I almost skimmed past this.

“Elon merges SpaceX and xAI.” Okay. Another Elon thing. Probably some vague “AI + space = future” headline. Move on.

But then I actually read the explanation. And it bothered me—in a good way.

Because it wasn’t about models, performance, or even “intelligence.” It was about… electricity.

Which is not the story anyone wants to tell. But it’s the real one.

1. The Part About AI Nobody Likes Talking About

When people talk about AI, they usually talk about bigger models, better benchmarks, smarter reasoning, and how AGI is coming “soon.”

They don’t talk about power grids, cooling systems, heat, and zoning permits.

But that’s where things are breaking.

Right now, AI lives in giant data centers. Not “the cloud”—actual buildings, full of machines that get very hot and need absurd amounts of power. And we’re starting to hit limits.

In a lot of places, utilities are basically saying: “We don’t have more electricity to give you.”

Which is wild, if you think about it. We’re in a world where intelligence is software, but intelligence is being capped by wires.

So when SpaceX said that “ground-based data centers are hitting physical limits,” that wasn’t PR. That was a constraint. Translation: we are running into physics.

Elon’s response was predictable in retrospect. If the problem is Earth, leave Earth.

In space, solar is constant, cooling is easier, and nobody complains. Nobody says, “Sorry, the grid is full.”

Suddenly, “AI in orbit” stops sounding like cosplay and starts sounding like infrastructure. And infrastructure is boring—but boring is where power lives.

2. Why This Isn’t About “Better Models”

Most AI companies are playing the same game: make the model better, make it cheaper, make it faster, repeat.

That works. Until everyone catches up. Then it turns into a knife fight over margins.

Elon doesn’t like knife fights. He likes terrain control.

So instead of asking, “How do we make Grok smarter?” he asks, “Where does Grok live?” And then he builds that place.

Look at the pieces: SpaceX launches things. Starlink connects things. xAI processes things. Tesla distributes things.

Individually, they’re impressive. Together, they form an environment—a closed loop of compute, network, distribution, data, and feedback.

Most startups build tools. He builds ecosystems.

And ecosystems are annoying to compete with. You can copy a feature. You can’t copy a landscape.

3. Vision Is Cheap. Execution Paths Aren’t.

Elon is famous for big ideas: Mars, self-driving, robots, neural interfaces. Some work. Some take much longer than promised.

The difference is rarely “belief.” It’s whether there’s a real path.

So instead of asking, “Is space AI brilliant?” ask, “What has to go right for this to work?”

A lot.

Regulators have to approve launches. Costs have to keep falling. Hardware has to survive radiation. Maintenance has to be manageable. Failure rates have to stay low.

If any of those break, the story weakens. If several break, it collapses.

This is how serious people evaluate ambitious projects. Not emotionally. Mechanically.

They turn visions into systems. They break ideas into steps, watch timelines, track constraints, and revise beliefs.

Most people never learn this skill. They either become fans or skeptics. Both are passive positions.

Builders can’t afford that.

Every meaningful project eventually becomes “a story.” And stories are dangerous. They seduce you into thinking progress is automatic.

Your job is to keep dragging the story back into reality—into schedules, bottlenecks, trade-offs, and work.

One Last Thought

This merger wasn’t about making AI smarter. It was about AI hitting limits: power, heat, regulation, physics.

When growth runs into walls, winners don’t just try harder. They change the map.

That’s what this is.

For independent builders, the lesson isn’t “think big.” It’s this: real advantage usually looks boring at first. It looks like infrastructure. It looks slow. It looks unsexy.

And by the time it looks impressive, it’s already too late to copy.

That’s how durable systems are built.

Quietly. Over time.

Links:

1. https://x.com/elonmusk

2. https://www.spacex.com

3. https://x.ai

4. WTF is happening at xAI | Sulaiman Ghori

xAI went from zero to $50 billion valuation in under 4 years. They built the world’s largest AI compute cluster in 122 days. They’re deploying AI employees that work 8x faster than humans.

An engineer who worked there just spilled everything in a 70-minute podcast. Then got fired three days later.

But before he left, he handed us their playbook. Let’s steal it.

The Speed vs. Smarts Fork in the Road

Every AI company faces the same question early on: Do we build the smartest possible model, or the fastest?

Most companies pick “smartest.” It’s sexier. Better for fundraising. OpenAI and Anthropic are racing to build models that think for 5 minutes before responding, producing essays that make you go “whoa, that’s impressive.”

xAI picked fast.

Their internal models run 8x faster than a human. Not 8x faster than other AIs—8x faster than an actual person doing the same task.

Here’s the fork they faced:

Option A: Build a huge model that can reason deeply, impress demos, win benchmarks. Deploy to maybe 10,000 users because inference costs are insane.

Option B: Build a small model that’s “good enough,” runs dirt cheap, and deploy to 1 million users tomorrow.

They picked B. And here’s what they gave up:

They can’t claim to have the “most intelligent” AI. When people compare models, xAI’s won’t top the charts. They won’t win the benchmark wars. Their demos won’t make people gasp.

But here’s what they got:

They can actually deploy at scale. While competitors are still figuring out how to afford serving their models, xAI can run theirs on Tesla car computers. While others need 5 minutes to think, xAI responds in 10 seconds.

The math is brutal: If your AI needs 5 minutes to do a task a human does in 5 minutes, nobody uses it. If your AI does it in 30 seconds, everyone uses it.

Sully mentioned they’re doing the same thing Tesla did with Full Self-Driving. Tesla didn’t build the biggest neural network. They built the smallest one that works, because it had to fit in a car.

Same logic: What’s the smallest model that can actually do the job?

This created a second-order effect they didn’t expect: small models iterate faster. Training a small model takes days, not weeks. They can run 20 experiments in parallel. They can test a new approach in the morning and deploy it that night.

While competitors spend months perfecting one big model, xAI ships 20 “pretty good” versions and learns what actually works.

Here’s the question for your own work:

When you’re building something—a presentation, a strategy, a new process—are you optimizing for “impressively smart” or “actually useful”?

Because those are different goals. And only one of them ships.

The Carnival Generator Gambit

xAI needed to build a massive data center in Memphis. Fast.

Standard timeline: 2+ years. Permits, environmental reviews, utility approvals, construction.

They had a choice:

Option A: Follow the proper process. Apply for industrial permits. Wait 18-24 months for approvals. Build a permanent, compliant facility.

Option B: Find a loophole.

Their lawyers discovered that “temporary events” like carnivals can use mobile generators without lengthy environmental approvals.

So xAI rented carnival generators, put them on trailers, and classified the entire data center as a “temporary facility.”

They built the world’s largest AI compute cluster using carnival equipment in 122 days.

Here’s what they traded:

They gave up legal certainty. The EPA is now investigating. They might face fines. They might be forced to rebuild properly. The “temporary lease” might become a permanent problem.

They gave up looking professional. There’s now a podcast where an engineer explains they’re basically running a carnival. Competitors can point and laugh. Investors might worry about regulatory risk.

But here’s what they got:

122 days instead of 730 days. That’s a 6X speed advantage. In AI, that’s multiple generations of models. That’s the difference between leading and following.

And here’s the part nobody talks about: by the time regulators figure out what to do, xAI will have already built three more data centers the “proper” way. The temporary one will have already paid for itself.

This is a pattern you see in every Musk company: Ask for forgiveness, not permission. But make sure you’re so far ahead by the time forgiveness is needed that it doesn’t matter.

SpaceX did this. Tesla did this. Now xAI is doing it.

But let’s be honest about the tradeoff: This only works if you can absorb the hit. If you’re a small startup and regulators come after you, you’re dead. If you’re xAI with billions in funding, you pay the fine and move on.

The question for your work isn’t “should I break rules?”.

The question is: What’s the “carnival generator” version of your project?

What’s the scrappy, probably-not-perfect-but-actually-works version you could ship in a week instead of waiting three months for the “proper” version?

Maybe it’s using free tools instead of waiting for enterprise software approval. Maybe it’s manually doing something that “should” be automated, just to prove it works. Maybe it’s testing with 10 users instead of waiting to test with 1,000.

The proper version will come. But the carnival version teaches you if you’re even building the right thing.

The Loose Lips Paradox

Sully got fired for one reason: he told the world exactly how far xAI has gotten.

But here’s the paradox: Everything he said was obvious to competitors.

Every AI lab knows small models are an option. Everyone knows Tesla cars have unused compute. Everyone’s testing AI agents internally.

Sully didn’t leak secrets. He leaked the progress bar.

He told everyone xAI is already:

• Running AI employees that fool humans

• Achieving 8x human speed in production

• Planning to deploy on Tesla’s car fleet

• Iterating daily on new model architectures

That last one is the killer. Competitors now know they’re not just behind—they know exactly how behind they are.

Here’s the fork Sully faced (probably without realizing it):

Option A: Stay quiet. Keep your head down. Ship great work in private. Build leverage through execution, not reputation.

Option B: Share openly. Build your personal brand. Attract opportunities. Show the world what you’re capable of.

Most advice tells you to pick B. “Build in public!” “Share your journey!” “Be authentic!”

And for 99% of people, 99% of the time, that’s right.

But there’s a threshold where openness becomes liability.

Sully crossed that threshold. He didn’t just share what he was working on (fine). He didn’t just share the approach (probably fine). He shared:

• Exact progress milestones

• Internal testing details

• Competitive timelines

• Legal gray areas

• Product roadmaps

He essentially handed every competitor a checklist: “Here’s what you need to catch up to xAI.”

He built a great personal brand, got podcast fame, and inspired thousands of engineers. But he lost his job and possibly his industry reputation.

Now, was it worth it? Depends on what he wanted.

If he wanted to work at xAI long-term, obviously not. If he wanted to become a known voice in AI and start his own company? Maybe yes. That podcast will open doors. VCs will remember his name.

But he didn’t make that choice consciously. He just got excited and talked.

Here’s the framework that would have helped:

Share the vision freely. “We’re building AI that works 8x faster than humans”—that’s inspiring, not revealing.

Share the approach selectively. “We’re using smaller models for speed”—that’s interesting, not damaging.

Never share the progress bar. “We’re already running 20 internal AI employees” + “We’re deploying on Tesla cars” + “We iterate daily”—that’s a roadmap for competitors.

The question for your work:

When you’re tempted to share your project publicly—on LinkedIn, in meetings, in Slack—ask: Am I sharing to build support, or am I accidentally giving competitors my playbook?

Because there’s a point where transparency stops being helpful and starts being stupid.

Sully found that point. Three days too late.

Outro

xAI’s playbook is simple: choose speed over perfection, use carnival generators when you can’t get real ones, and guard your progress bar like your job depends on it.

Because it does.

Sully proved that. He built incredible things, inspired thousands of people, and lost his job in 72 hours.

Sometimes the best lessons come from watching someone else step on the landmine.

Links:

1. https://x.com/elonmusk

2. https://x.com/sulaimanghori

3. https://x.ai

4. WTF is happening at xAI | Sulaiman Ghori

An 18-year-old made $2 million per month with a calorie-tracking app, eight months after launch.

But here’s the part that made me spit out my coffee: He didn’t build his own custom AI model. He just... connected two existing ones.

While I’m over here trying to figure out which Python library to use, he’s treating AI models like LEGO bricks.

Let me show you what he did that I completely missed.

1. The Build vs. Buy Decision (Spoiler: He Chose Neither)

Zach Yadegari wanted to track calories for his gym sessions. MyFitnessPal made him manually confirm every food item after taking a photo. He quit after three days.

Most technical people would think: “Time to build a custom model!”

I certainly would have. Six months of gathering data, fine-tuning, debugging why it thinks an apple is tikka masala...

Zach did something different.

He just plugged OpenAI’s GPT vision model and Anthropic’s Claude vision model together. Added some RAG (Retrieval-Augmented Generation) so the models could double-check themselves. Done.

90% accuracy. Better than industry average.

Total time spent training custom models: Zero hours.Here’s what I didn’t understand until I saw this: The question isn’t “build or buy.” It’s “compose or compete.”

Trying to build better vision models than OpenAI is like trying to manufacture better electricity than your local power company. Sure, you could. But why would you?

The companies competing on “better AI models” are burning through millions in compute costs. The companies competing on “better AI model composition” are four teenagers in a hacker house.

This creates a weird tradeoff most people miss:

If you build your own models, you own the technology but you’re competing with Google’s budget.

If you compose existing models, you own the application but you’re dependent on someone else’s API pricing.

Zach chose dependency. And hit $24M ARR in 8 months.

Meanwhile, there are startups that spent two years building custom models and never launched.

The hard part isn’t the AI anymore. It’s knowing what problem to solve and how to package it.

I spent so much time learning ML fundamentals that I missed the fundamental shift: In the age of model APIs, “knowing how to code AI” is less valuable than “knowing what to build with AI.”

Everyone has access to GPT now. Not everyone knows what to do with it.

2. When $770K in Marketing Spend Is Actually Market Research

Cal AI spends $770,000 per month on ads and influencer marketing.

That’s almost their entire profit going right back out the door.

My first reaction: “Oh great, another startup hemorrhaging cash for vanity metrics.”

But look at how they spent it:

Phase 1 (Early days): Partner with fitness influencers on TikTok. Performance-based payment plus referral codes.

Phase 2 (After validation): Scale with TikTok Ads and Apple Search Ads.

See the difference?

Phase 1 wasn’t marketing. It was paying influencers to tell them if their product was any good.

If the app sucked, influencers’ followers wouldn’t convert. The referral codes would show that immediately. Cal AI would know within days whether they built something people actually wanted.

This is the opposite of what I thought marketing was.

I thought: Build product → Validate with users → Then market it.

They did: Market to users → Validation happens automatically → Scale the marketing that works.

The tradeoff here is subtle but important:

Traditional validation is cheap but slow. You talk to users, run surveys, do beta tests. It takes months to know if you’re building the right thing.

Marketing-as-validation is expensive but fast. You pay influencers to promote your product and watch what happens. You know within days.

Cal AI spent $770K to learn what most startups spend 18 months trying to figure out: Who actually wants this? How much will they pay? What messaging works?

By the time they scaled to paid ads, they already knew:

• Which user segments converted best

• What pain points resonated

• How to position the product

• What price points worked

The expensive “marketing” in Phase 1 made the expensive marketing in Phase 2 way more efficient.

Here’s what clicked for me: Sometimes the fastest way to validate isn’t to build smaller—it’s to spend louder.

I keep trying to “validate cheaply” by talking to potential users, running surveys, building MVPs. But talk is cheap. Watching someone actually pull out their credit card? That’s real data.

Cal AI’s “reckless” marketing spend was actually the most conservative move they could make. Because finding out your product is wrong after 6 months of careful validation is way more expensive than finding out after 2 weeks of paid traffic.

3. The Reddit Hates You, App Store Loves You Paradox

Reddit absolutely destroys Cal AI.

“Predatory pricing!” “Hidden fees!” “Scammy subscription traps!”

The complaints are everywhere. Long, detailed posts about how the app tricks people into subscriptions they didn’t want.

Meanwhile, App Store rating: 4.8 stars. 150,000+ reviews. 30%+ retention rate.

What’s happening here?

Most founders would panic at the Reddit backlash. Try to fix the pricing. Make it more transparent. Add more free features. Chase universal approval.

Cal AI looked at the data and basically said: “Cool, Reddit isn’t our customer.”

They designed everything for people who value convenience over cost optimization:

• Hide the price until after onboarding (Reddit hates this)

• Use dynamic pricing based on user behavior (Reddit really hates this)

• Make the free version basically useless (Reddit loses their minds over this)

• Auto-charge after 3-day trial (Reddit wants to burn the building down)

But here’s the thing: The people complaining on Reddit and the people paying $49.99/year are two completely different humans.

Reddit users want maximum value per dollar. They’ll spend 30 minutes researching alternatives to save $5.

App Store users want minimum friction. They’ll pay $50 to avoid spending 30 seconds manually logging food.

Cal AI picked the latter and optimized everything for them. Made the experience so smooth that the price doesn’t matter. Built in just enough behavioral economics (sunk cost fallacy, dynamic pricing, fake freemium) to convert that audience at scale.

You can’t serve both audiences.

If you make Reddit happy with transparent pricing and generous free tiers, you lose the friction-hating users who were going to pay anyway. They need a little push.

If you optimize for conversion, you get Reddit threads full of angry people. But those people were never going to pay the premium tier price anyway.

This was hard for me to accept because I’m a Reddit person. I research everything. I optimize costs. I get offended by hidden pricing.

But I am not the target market for most consumer apps.

The real lesson: Controversy from the wrong audience is fine. Silence from your actual audience is death.

Cal AI would rather have 1,000 Reddit threads calling them a scam and 500,000 happy paying users than 1,000 Reddit upvotes and 5,000 users unwilling to pay.

They picked their lane. Optimized ruthlessly. Ignored everyone else.

Meanwhile, I’m over here trying to make everyone happy and ending up with a product nobody loves.

The Pattern I Keep Missing

So here I am, trying to build the “right way.” Custom models. Careful validation. Fair pricing. Universal appeal.

And this 18-year-old just composed some APIs, spent like a drunken sailor on marketing, pissed off half the internet, and made $24M ARR.

The pattern isn’t about being young or lucky or unethical.

It’s about understanding that the rules changed. Building your own AI isn’t the flex anymore. Solving a problem people will pay for is the flex. And sometimes solving it fast and dirty beats solving it perfect and slow.

The future belongs to people who can see past what they “should” do and just ship the thing that works.

Even if Reddit hates it.

Links:

1. https://x.com/zach_yadegari

2. https://www.calai.app

3. https://www.reddit.com/r/nutrition/comments/1in01pp/anyone\_have\_success\_using\_cal\_ai

4. https://techcrunch.com/2025/03/16/photo-calorie-app-cal-ai-downloaded-over-a-million-times-was-built-by-two-teenagers

5. https://lifehacker.com/health/ai-powered-calorie-counting-apps-worse-than-expected

6. https://www.cnbc.com/2025/09/06/cal-ai-how-a-teenage-ceo-built-a-fast-growing-calorie-tracking-app.html

7. https://www.eesel.ai/blog/cal-ai-pricing

8. https://dataintelo.com/report/calorie-counting-app-market

Imagine spending months building your product, then Anthropic drops a keynote that makes your roadmap obsolete. That’s what happened to Eigent in January 2026. Instead of folding, they posted a joke tweet that got 8.3k likes and 1.6M views. A day later, 1,000+ GitHub stars. Two weeks later, they’re #1 on GitHub Trending. Here’s how they pulled off what looks like magic but is actually just very good decision-making under pressure.

They Chose the “Wrong” Option (Which Made It Right)

Late night, London.

Three co-founders sit in their living room after watching Claude Cowork launch.

The product they’ve been building? Anthropic just announced something eerily similar.

Here’s where most founders panic-pivot.

Or write a dramatic “we’re shutting down” post.

Or try to out-feature the giant (LOL).

Eigent did none of these.

They doubled down on open source.

Now, I know what you’re thinking.

“Open source? How do you compete with a well-funded closed product by... giving yours away?”

This sounds like bringing a knife to a gunfight.

Except it’s not.

Here’s what I missed for years:

The best product for everyone is often the wrong product for someone specific.

Let me explain with an analogy that’s probably too honest.

You know how McDonald’s makes objectively good burgers? Fast, consistent, cheap.

But there’s a thriving market for $18 organic grass-fed burgers at local joints.

Why? Because some people don’t want McDonald’s.

Not because it’s bad. Because it’s wrong for them.

Eigent realized something crucial:

Not everyone wants Anthropic’s solution.

Who doesn’t?

Companies that can’t let data touch the cloud. (Banks, healthcare, government)

Teams that need deep customization. (The product has to fit their exact workflow)

Organizations allergic to vendor lock-in. (What happens if Anthropic changes pricing? Or shuts down the API?)

An 11,000-person company in the Middle East chose Eigent for local deployment.

A world-leading open-source data company picked them for customization.

Not because Eigent was “better.”

Because it was the only option that fit their constraints.

Here’s the thing I’m still processing:

When a giant enters your space, your job isn’t to beat them. It’s to become un-comparable.

You do this by serving people with needs the giant actively ignores.

Not niche as in “small.”

Niche as in “different shape.”

The tradeoff?

You give up mass market appeal.

You’ll never have Anthropic’s user base.

But you might have customers who can’t live without you.

Which actually matters more than I thought.

The Slow Accumulation That Became Sudden Speed

Okay, but here’s the part that makes this story actually interesting.

Eigent didn’t just have clever positioning.

They had three years of technical depth nobody else had.

March 2023: They publish CAMEL, a paper on multi-agent AI.

Gets 4,000 GitHub stars in a week.

Andrew Ng takes a photo of their poster at NeurIPS.

(If Andrew Ng photographs your work, you’ve made it.)

Then they just... kept going.

• CRAB: Teaching AI to operate phones and computers

• OWL: Browser automation

• OASIS: Simulating a million AI agents

• Loong: Generating high-quality training data

From the outside, this looks like random academic research.

But it’s actually a tech stack disguised as papers.

When Claude Cowork launched, Eigent didn’t scramble.

They had:

Ready-to-use browser automation from OWL.

Multi-agent coordination from CAMEL.

Enterprise benchmarks from actual client work.

I used to think research was separate from building products.

Like, you either do research (slow, academic) or ship products (fast, practical).

Turns out that’s wrong.

Research IS product development.

Just on a longer time horizon.

Here’s the pattern:

Shallow moat = “We have nice UI”

Deep moat = “We spent years solving problems you don’t even know exist yet”

When you do research, you’re not just publishing papers.

You’re building capabilities that compound.

Each paper teaches you something.

Each project adds a tool to your arsenal.

By the time competitors try to catch up, you’re not one step ahead.

You’re in a different dimension.

This applies to your work too.

That “extra” project you do on the side?

That skill you’re learning that seems irrelevant?

That’s your research phase.

It won’t pay off this quarter.

But in two years, when everyone’s scrambling to learn what you already know?

You’ll have options they don’t.

The Pause That Accelerated Everything

May 2024.

The product MVP is looking good.

Time to launch and grow, right?

Guohao writes a doc: “Lay the Foundation.”

The directive: Stop building the product. Fix the infrastructure.

Context: They have TWO engineers.

Not twenty. Two.

Every startup playbook says: Ship fast. Iterate. Get users. Grow.

This is the opposite.

They spend an ENTIRE YEAR making their framework production-ready.

No product launches.

No user growth.

Just engineering.

When I first read this, I thought: That’s career suicide.

A year with two engineers and nothing to show customers?

But here’s what actually happened:

When they finally launched, they didn’t have a duct-taped demo.

They had a system that enterprises could actually deploy.

Without a sales team, they started landing clients:

• 11,000-person Middle Eastern company

• Major open-source data company

• Google inviting them to test Gemini early

Why?

Because enterprise buyers care if your thing works, not how fast you shipped.

Nobody wants to be your beta tester when real money is on the line.

They want boring reliability.

This breaks every “move fast” rule I was taught.

But the tradeoff makes sense:

You can go fast with a weak foundation and collapse later.

Or go slow with a strong foundation and accelerate later.

Most people pick fast → collapse.

Because slow → accelerate requires patience nobody has.

But here’s what I’m learning:

Real speed isn’t about how fast you ship version 1.

It’s about how fast you can scale version 10.

A shaky foundation means every new feature is a negotiation with your technical debt.

A solid foundation means you can stack features like Legos.

The first approach feels fast for six months.

Then turns into molasses.

The second approach feels slow for six months.

Then becomes unstoppable.

For your work:

That boring refactor you keep postponing?

That documentation you “don’t have time for”?

That’s your foundation.

Skip it, and you’ll spend next year fixing what you built this year.

Build it, and next year you’ll be building on top of what you built this year.

Different game entirely.

The Real Game

January 2026, Eigent hits #1 on GitHub Trending.

But that’s not the point.

The point is what they gave up to get there—and what they gained instead.

They gave up: Trying to beat Anthropic head-on. Mass market dreams. The “move fast” playbook everyone follows.

They gained: A defensible position. Enterprise clients who can’t leave. Technical depth nobody can copy quickly.

The lesson isn’t “do what Eigent did.”

It’s “understand what you’re trading.”

Every choice has a cost. The question is whether you’re paying it on purpose.

When the giant launches, will you have something they can’t replicate? Not because you’re better—because you’re different in a way that matters to someone specific.

That’s not a consolation prize. That’s the whole game.

Links:

1. https://x.com/guohao_li

2. https://www.eigent.ai

3. CAMEL: Communicative Agents for “Mind” Exploration of Large Language Model Society

4. CRAB: Cross-environment Agent Benchmark for Multimodal Language Model Agents

5. OWL: Optimized Workforce Learning for General Multi-Agent Assistance in Real-World Task Automation

6. OASIS: Open Agent Social Interaction Simulations with One Million Agents

7. Loong: Synthesize Long Chain-of-Thoughts at Scale through Verifiers

A patient in Liverpool waited 91 hours on a hospital gurney for a bed. Not 91 minutes—91 HOURS. Four days listening to footsteps and beeping machines. This wasn’t an outlier. This was winter 2024 in the UK’s NHS. But then 50 hospitals did something weird: they stopped looking at other hospitals for answers and started copying Amazon instead.

The Problem: Hospitals Only Know How To Live In The Present

Here’s how hospitals traditionally work:

Patients show up. Staff scramble.

More patients show up. Staff scramble harder.

Everyone scrambles until someone collapses or the day ends.

The entire system lives in the present tense. What’s happening NOW, we handle NOW. No bandwidth for tomorrow.

This works fine during “normal times” (that mythical two-week period in April when nobody gets sick).

But winter? Winter is chaos.

One ER director’s typical morning: Arrive at 8 AM. Discover 200 patients need care today. Remember you only scheduled 150 staff. Start frantically calling people.

First your own nurses: “Want to work a double shift?”

Then the temp agencies—those mercenary outfits that provide contract staff at 2-3x the cost. And they’re not even guaranteed to show up because every other hospital is calling the same agencies, fighting over the same pool of temps like it’s Black Friday for healthcare workers.

But even if you find staff, you still have a problem: Where do you put the patients?

The ER can treat them, sure. But then they need to go to a hospital ward. Except the wards are full. Why? Because the patients IN the wards can’t leave.

Not because they’re sick. Many are medically cleared. But they’re waiting for prescriptions, or rides home, or community care arrangements.

This is the “exit block problem.” (Polite hospital-speak for “the entire system is constipated.”)

Imagine a parking garage where cars keep entering, no spaces are available, and the exit is blocked by cars that COULD leave but are waiting for... reasons.

That’s a hospital in winter.

The patient who waited 91 hours? That wasn’t because doctors weren’t trying. It was because the system had no ability to plan ahead.

Like running a restaurant where you wait until you’re completely slammed, THEN run to the grocery store to buy ingredients and hire a chef.

That restaurant would fail immediately.

Hospitals just suffer.

The Solution: Time Travel (Sort Of)

The NHS rolled out an AI tool that predicts ER demand three weeks in advance.

Here’s what surprised me: Accuracy doesn’t really matter.

Whether it predicts 30% more patients or 25% more—who cares?

What matters is having ANY advance warning at all.

The system looks at weather forecasts (temperature drops = respiratory spikes 3-7 days later), school holidays (kids spread disease like it’s their job), historical patterns, flu transmission rates.

Then it tells you: “Next Monday 10 AM-2 PM, you’re getting slammed” or “Three weeks from now, cold snap + flu season = 30% surge.”

With three weeks of warning, hospitals can finally do three things:

One: Stop playing schedule roulette.

Traditionally, hospitals schedule staff weeks ahead based on last year’s patterns or vibes. If something unexpected happens, panic-call temp agencies.

Now they can see the staffing gap coming and post overtime shifts to “bank staff”—regular employees who want extra cash.

Using internal staff costs WAY less than temp agencies. Plus you can optimize: pediatric surge predicted? Schedule more pediatric nurses specifically.

Two: The great bed liberation.

If you know a wave is coming in three weeks, start planning discharges NOW.

Which patients are medically ready but stuck waiting? Get them out. Which elective surgeries can be postponed? Reschedule them.

You’ve turned “exit block” from a passive disaster into active management.

Three: Surgery Tetris.

Non-emergency surgeries get scheduled dynamically based on predicted ER pressure. Light week? More surgeries. Busy week? Scale back, save resources for emergencies.

One pilot hospital called this “groundbreaking.” Translation: “Holy shit, we can actually plan things now.”

The real breakthrough isn’t the AI. It’s that AI changed how hospitals think about time.

Old hospitals operated in linear time: present moment only.

New hospitals got a new dimension: the next three weeks.

Those three weeks aren’t for waiting. They’re for preparing.

It’s like the difference between driving without GPS (you only see what’s directly ahead) versus with GPS (you can see three kilometers ahead and adjust before problems hit).

GPS doesn’t change your car. It changes your understanding of the road.

AI didn’t change the hospital’s resources. It changed their understanding of time.

The Real Lesson: They Learned From The Wrong Industry

Here’s what’s wild about this case:

This playbook isn’t new. It’s exactly what e-commerce has been doing for years.

Amazon’s “predictive dispatch” looks at your browsing history and buying patterns, then moves items to local warehouses BEFORE you click buy. That’s how they do 2-hour delivery.

NHS hospitals are basically doing e-commerce logistics with nurses and beds.

The difference: E-commerce moves packages. Hospitals move nurses. E-commerce can stock inventory in advance. Hospitals can only prepare the ABILITY to handle patients.

Which makes this approach even MORE valuable in healthcare.

But here’s what struck me:

Healthcare didn’t learn this from other hospitals. They learned it from retail.

The innovation wasn’t medical. It was operational.

The best hospitals didn’t benchmark against other hospitals. They benchmarked against Amazon.

They stopped asking “What are other hospitals doing?” and started asking “What are the best operators in ANY industry doing?”

That’s the move.

When your entire industry operates one way, look outside your industry for different playbooks.

The constraint isn’t usually what you think it is.

Hospitals thought the constraint was nurses and beds (resources).

Turns out the real constraint was TIME. Specifically: not having advance warning.

E-commerce figured out how to manufacture certainty from data. Hospitals just copied the homework.

And it’s working: ambulances stopped piling up at entrances, patients move through faster, temp agency costs drop.

This is now running in 50+ UK hospitals. It’s a flagship government project.

But the playbook isn’t government or healthcare specific.

Any business facing demand uncertainty can use this:

Call centers predicting volume spikes. Airports predicting delays. Supply chains predicting stockouts. Cities predicting traffic jams.

When everyone’s firefighting, you’re already preventing fires.

That’s not just an advantage. That’s a different game.

The Punch:

NHS hospitals didn’t just steal Amazon’s playbook. They learned how to pull the future into the present. And the best part? They didn’t need to invent anything new—they just needed to look at who was already solving similar problems in completely different contexts. Sometimes the best answer to your problem is sitting in an industry that has nothing to do with yours.

Links:

1. https://www.gov.uk/guidance/ai-exemplars-programme

2. https://www.gov.uk/government/news/faster-treatments-and-support-for-health-workers-as-ai-tackles-ae-bottlenecks

Anthropic tripled their engineering team in one year.

Each engineer became 70% more productive.

The tool behind it? Claude Code.

Here’s the weird part: 80% of Claude Code is written by Claude Code itself.

Today we’re stealing Boris Cherny’s playbook. Not the “what” (you already know Claude Code won). The “how”—the forks in the road where they could have zigged but zagged instead.

And why those zigs would have killed them.

Crossroad #1: The Beautiful UI vs The Black Screen

Picture this:

It’s early 2024. Every AI coding tool is racing to build the prettiest interface.

Cursor has that sleek split-pane editor. Windsurf has beautiful diffs. Everyone’s designing for the Figma generation.

Then Boris builds... a command line tool.

A terminal.

Black screen. Green text. Zero GUI.

His manager at Anthropic, Ben Mann, could have said: “Nobody’s going to use this. Make it pretty.”

Instead he said: “Don’t build for today’s model. Build for the model 6 months from now.”

This is the fork in the road.

Option A: Build the polished IDE everyone expects.

• Pro: Non-technical users can use it day one. Lower learning curve. Easier adoption.

• Con: You have to build ALL the scaffolding. File trees. Syntax highlighting. Git integration. Settings panels. This takes months. And limits what the AI can do to what you’ve built UI for.

Option B: Give Claude raw terminal access and nothing else.

• Pro: Claude can do literally anything you can do in a terminal. Zero limitations. Also, you ship in weeks not months.

• Con: Terrifying for non-technical users. Sales team sees it and goes “...what is this?”

Boris chose Option B.

Here’s why it worked:

When you give Claude a terminal, you give it the same interface YOU use.

Anything you can do, Claude can do.

• Need to check git history? git log

• Need to run tests? npm test

• Need to deploy? ./deploy.sh

No custom tools. No wrapper APIs. No “we need to build that feature.”

Just bash.

The tradeoff is brutal though.

Cat (the PM) told a story about their marketing team member trying Claude Code. She got 30 permission pop-ups in a row because she’d never used a terminal before.

They almost lost her.

But here’s what I didn’t expect:

The terminal actually made it more flexible, not less.

Because non-technical people could now do technical things.

Brandon, their data scientist, had never used a terminal. Now he runs multiple Claude Code instances writing SQL.

Sales people use it to query Salesforce.

Someone used it to decide which sofa to buy.

Nobody expected this.

If they’d built a polished coding IDE, it would have stayed a coding tool.

The terminal made it more accessible long-term, not less.

This is backwards from how we usually think.

We think: Polish → Accessibility.

But actually: Flexibility → Accessibility (once users learn the basics).

The lesson for you:

When you’re building something (a workflow, a template, a system), you face this choice:

Make it pretty and constrained? Or make it ugly and flexible?

We default to pretty.

Pretty feels professional. Pretty feels done.

But pretty also means locked-in.

Every design choice you make is a constraint you’re imposing.

Sometimes the right move is to give people the raw tools and get out of the way.

Yes, the learning curve is steeper.

But the ceiling is way higher.

Crossroad #2: The Smart Database vs The Dumb Search

Early Claude Code had a problem:

How does Claude find relevant code in a massive codebase?

Every AI coding tool faces this. Your codebase is too big to fit in context.

The obvious answer: Use RAG (Retrieval Augmented Generation).

Index the codebase. Create vector embeddings. Build semantic search. Store it in a database.

This is what everyone does. This is the “smart” way.

Boris tried it. It worked... okay.

But there were problems:

Problem 1: The index goes stale. You change code. The index doesn’t update. Now Claude is working with old information.

Problem 2: Security. Your index has to live somewhere. A third-party provider? Now you’re uploading your codebase to someone else’s servers. Build it yourself? Now you’re maintaining database infrastructure.

Problem 3: Complexity. More moving parts. More things to break. More things to maintain.

So Boris hit a fork in the road:

Option A: Invest in making RAG really good. Build auto-indexing. Build security. Build infrastructure. Optimize retrieval. This is technically sophisticated. This is what a “real” company would do.

Option B: Delete all the RAG code and just let Claude search with grep and bash. This is caveman simple. This seems worse. Grep is brute force. It’s slow. It’s expensive (more tokens, more API calls).

Boris chose Option B.

He deleted the RAG system.

Now when Claude needs to find something, it just... searches. With regular code search. With grep. With find.

Agentic search.

Claude decides what to search for. Runs the search. Looks at results. Searches again. Iterates.

This seems insane, right?

You’re burning tokens and time on something a database could do instantly.

But here’s what happened:

It outperformed RAG by a lot.

Boris’s words: “It outperformed everything by a lot.”

Why?

Because the model is smart enough to know what to search for.

RAG is dumb retrieval. You ask for “authentication” and it gives you every file with that word.

Agentic search is smart exploration. Claude searches, evaluates, searches differently, narrows down.

It understands what it’s looking for.

Plus:

• No index to maintain

• No security issues (everything stays local)

• No drift between code and index

• Works immediately (no indexing step)

The tradeoff:

You’re trading tokens and latency for simplicity and security.

This trade makes no sense in 2020.

But in 2026? When model calls are cheap and getting cheaper?

When security is paramount?

When developer time is more expensive than compute?

Suddenly burning tokens to avoid infrastructure complexity is the smart move.

Here’s what this means for you:

You probably have some system that’s “optimized.”

Some spreadsheet with elaborate formulas. Some process with seven steps to avoid one manual check.

You built complexity to save effort.

But what if the effort you saved is now cheaper than the complexity you’re maintaining?

I have a friend who built an elaborate automation to avoid checking something twice a week.

The automation breaks constantly. She spends more time fixing it than she saved.

She won’t delete it because “I spent so much time building it.”

Sunk cost fallacy, optimized edition.

Sometimes the dumb solution is the smart solution.

Especially when the dumb solution is maintainable and the smart solution is brittle.

Boris’s rule: Simple today beats optimized tomorrow.

Because tomorrow the constraints will change anyway.

Crossroad #3: The Big Team vs The Tiny Team

Here’s a weird one:

Anthropic raised a ton of money. They’re scaling fast. They’re hiring aggressively.

Claude Code is their most-used internal tool. 70-80% daily active users among engineers.

Standard playbook: Throw 50 engineers at it. Scale it up. Build everything.

Instead?

They kept the team tiny.

Cat (the PM) explains their philosophy:

“At Anthropic, our product principle is: do the simple thing first. You staff things as little as you can and keep things as scrappy as you can because the constraints are actually pretty helpful.”

This is bonkers.

You have the money. You have the demand. You have executive buy-in.

Scale up!

But they didn’t.

Why?

Fork in the road:

Option A: Staff up to 50+ engineers.

• Pro: You can build everything. Every feature request. Every integration. Polish everything. Move fast on execution.

• Con: More people = more coordination = more process = slower decisions. Also, more features = more complexity = more maintenance = more bugs.

Option B: Keep the team small.

• Pro: Fast decisions. Everyone knows everything. No coordination overhead. Forces you to say no to most features.

• Con: Slow execution. Can’t build everything. Some obvious features won’t ship. Users will complain.

They chose Option B.

The current team is still small. Under 10 people.

Boris said something wild: “Most of our features are just people building the thing they wish the product had.”

It’s not top-down roadmap planning. It’s engineers scratching their own itches.

Dixon wanted to get Slack notifications from Claude Code. Built hooks.

Daisy wanted a plugin marketplace. Built plugins.

Someone wanted Vim mode. It went viral.

But here’s the tradeoff:

By staying small, they can’t build everything.

Cat mentioned they’re still figuring out the subscription model. Still working on better pricing communication. Still building out enterprise features.

These are obvious things a bigger team would have shipped.

But staying small forces them to choose.

And choosing is how you maintain product clarity.

Big teams build everything. Small teams build the right things.

There’s another reason this works:

80% of Claude Code is written by Claude Code.

They don’t need 50 engineers.

They need 5 engineers managing 50 Claude instances.

Boris’s morning routine now:

• Wake up

• Open Claude Code mobile app

• Start 3-5 agents working on different tasks

• Go eat breakfast

• Come back, review their work

He’s not writing code. He’s orchestrating agents.

Fiona, the team’s manager, hadn’t coded in 10 years. Now she ships PRs weekly.

When your engineers have 10x leverage from AI, you don’t need 10x engineers.

You need really good judgment about what to build.

And small teams have better judgment than large teams.

Always.

Here’s what this means for you:

When you get resources (budget, headcount, time), the default is to use them.

“We have the budget, so let’s hire.”

“We have the time, so let’s build more features.”

“We have the people, so let’s add more projects.”

But resources are a trap.

More resources = more complexity = more coordination = slower decisions.

Constraints force clarity.

When you can only do three things, you pick the three most important things.

When you can do thirty things, you do thirty mediocre things.

The best projects I’ve worked on had too few people and too little time.

We had to be ruthless about scope. We had to make hard choices.

We built something focused and good.

The worst projects had plenty of everything.

We built a sprawling mess that nobody loved.

Boris’s insight: Treat constraints as features, not bugs.

If you don’t have enough people, good. Forces you to automate.

If you don’t have enough time, good. Forces you to ship simple.

If you don’t have enough budget, good. Forces you to be creative.

The question isn’t “how do I get more resources?”

The question is “how do I do more with what I have?”

That’s where the leverage is.

Outro

Three crossroads. Three counterintuitive choices.

Raw terminal over polished UI. Dumb search over smart database. Tiny team over big team.

Each one traded short-term convenience for long-term leverage.

Each one only works if you trust the model will get better.

And it will.

Your move: Find one place where you’ve over-optimized. Delete it. Replace it with the dumb, simple version.

Watch what happens.

Links:

1. https://x.com/bcherny

2. https://claude.ai/code

3. Claude Code: Anthropic’s CLI Agent

4. A conversation on Claude Code

5. The future of agentic coding with Claude Code

6. The Secrets of Claude Code From the Engineers Who Built It

7. Boris Cherny (Creator of Claude Code) On What Grew His Career And Building at Anthropic

Meet Connor.

He is 23 years old.

While I was celebrating my 23rd birthday by panicking because I didn't know what a '401k' was (I thought it was a running race), Connor is currently generating $185,000 a month.

That is over $2 million a year. From his laptop.

He doesn’t have a giant staff. He doesn’t have venture capital funding. He just has a very specific way of looking at the world that makes me feel both incredibly inspired and deeply, deeply lazy.

Today, we are going to surgically remove the top of his skull (metaphorically, legal team, calm down) to see how he built an empire before his prefrontal cortex even finished developing.

1. The Death of the “Coder” (and the Birth of the Vibe)

I usually treat coding like a high-stakes bomb defusal. One wrong wire (or missing semicolon), and the whole thing explodes.

Connor treats coding like he’s ordering a sandwich.

He uses AI tools like Cursor and Claude. But he doesn’t just ask them to “write code.” He treats the AI like a highly talented, extremely literal intern who has had way too much coffee.

The Workflow:

• Step 1: He downloads 20 successful apps in a niche.

• Step 2: He screenshots the best parts. The onboarding flow from App A. The payment screen from App B.

• Step 3: He throws those screenshots into a Figma file to create a Frankenstein monster of “good design.”

• Step 4: He feeds the design to the AI and says, “Make it look like this.”

He does not write code. He conducts it.

If the button is the wrong color, he doesn’t hunt for the hex code in a CSS file. He tells the AI, “Make it pop more. Like, angry red.”

He has sacrificed Precision (understanding exactly how the engine works) for Speed (driving the car at 200mph).

He realized that in 2026, writing code is a commodity. The AI can do the syntax.

The scarcity is Taste.

His job isn’t to build the wall; his job is to tell the bricklayer exactly where to put the brick so the house looks cool.

2. The “Marriage” Proposal

Okay, so he Vibe Coded an app in two weeks. Great.

Now he has the classic problem: The Empty Party.

He built a cool house, but nobody is there.

Usually, this is where we panic and set a pile of money on fire by giving it to Mark Zuckerberg for ads.

Connor realized something counter-intuitive: Paying for attention is for suckers.

He looks for influencers who have:

1. Massive, niche audiences (like “hardcore mountain bikers”).

2. Terrible monetization skills (they are selling t-shirts and making $50).

He slides into their DMs. But he doesn’t offer to pay them $500 for a shoutout.

He offers them Equity.

He says: “I built this app for your fans. I handle the tech. You handle the talking. We split the company.”

When you pay an influencer, they post once and forget you exist.

When you give them equity, they become a Co-Founder. They hustle. They make content because their rent depends on it.

He realized that giving away 50% of the company is actually the greedy move.

Because 50% of a rocket ship is worth a lot more than 100% of a PowerPoint presentation.

3. The “Don’t Fall in Love” Rule

I treat my projects like my actual children. I nurture them. I worry about them. I refuse to admit they might be ugly.

Connor treats his apps like stocks in a portfolio.

He currently has a main app (Payout) making bank. But he has others that failed.

Instead of trying to force a bad app to work, or letting it rot on his hard drive, he plays the leverage game.

If an app is okay but he’s bored of it? He sells it.

If an app needs marketing but he doesn’t want to do it? He finds a partner and gives them the keys.

He separates “ The Builder” from “The Operator.”

He knows his superpower is the 0-to-1 build (the Vibe Coding). Once it hits 1, he gets bored.

Instead of forcing himself to be the manager of a mature company (which sounds like a real job, gross), he just spins up the next plate.

He optimizes for his own attention span.

So, here is the plan.

Step 1: Stop obsessing over your syntax errors.

Step 2: Find a niche influencer who needs an app but hates computers.

Step 3: Let the AI do the heavy lifting while you take the credit.

The gatekeepers are gone, my friends. The only thing stopping us is our own insistence on doing things the hard way.

Links:

1. https://x.com/BusDownBonnor

2. https://www.trypayout.app

3. https://sfstandard.com/2025/03/06/startups-grindcore-ai-agents-hayes-cerebral-valley/

4. I vibe coded a $20K/month mobile app in 14 days

5. 23-YEAR-OLD MAKES $1 MILLION/YEAR WITH APPS BUILT IN JUST 2 WEEKS!

6. This 23 Yr Old Genius Makes $1M/Year with No-Code Apps. Here’s How

7. https://www.indiehackers.com/post/tech/growing-a-portfolio-of-mobile-apps-to-185k-mo-hZ4hqICtByIljkiJECQv

Morning, CEO!

Imagine walking into your sales department. There are ten desks. They used to be occupied by loud, caffeinated humans ringing gongs and complaining about lead quality.

Now? Silence.

Just ten empty chairs and the faint hum of a server rack.

Jason Lemkin, the godfather of B2B SaaS, didn’t just fire his sales team. He replaced 10 humans with 20 AI agents and 1.2 humans.

The result? Revenue didn’t drop. Efficiency went to the moon. And the office drama went to zero.

Let’s peek under the hood.

1. The “Meh” Middle is Lava

I have a confession. I prefer code to people. Code does what you tell it to do. Cold leads just hang up on you or report you as spam.

But for a long time, the business world ran on young people doing exactly that. We hired armies of SDRs (Sales Development Representatives). Their job was to exist. To “Show Up.” To spam inboxes and hope for a 1% reply rate.

It was boring. It was soul-crushing. But it was necessary.

Jason Lemkin’s experiment proved something terrifying: That job is dead.

Not “dying.” Dead. Buried. The tombstone is already engraved.

Why? Because the “Mediocre Middle”—the zone where you just go through the motions—is where AI thrives.

The SaaStr bots don’t just send emails. They read a prospect’s LinkedIn. They digest the company’s Q3 earnings report. They notice that the prospect just adopted a golden retriever. Then, they write a hyper-personalized email that sounds more human than the actual humans used to write.

And they do it in 4 seconds. At 3:00 AM on a Sunday.

This creates a brutal bifurcation in the workforce.

The Safe Zone A (The Top 1%): The humans who can close a million-dollar deal over a steak dinner because they possess charisma, empathy, and strategic genius. AI cannot eat steak (yet).

The Safe Zone B (The Wranglers): The people building the bots.

The Danger Zone (Everything Else): If your value prop is “I process information and move it from Spreadsheet A to Spreadsheet B,” you are standing on a trapdoor. And the AI just found the lever.

We can no longer build a career on being “competent.” Competence is now a commodity. We have to be exceptional, or we have to be the architect.

2. Your New Job: Managing Compute, Not Egos

Let’s talk about that “1.2 humans” statistic.

The “1” is a human Account Executive who handles the final, high-stakes closing conversations.

The “0.2” is the fascinating part.

Her name is Amelia. Jason calls her the “Chief AI Officer.”

She spends roughly 20% of her time managing this army of 20 robots.

Think about that. One person. Ten full-time employees’ worth of output.

This fundamentally shifts what it means to be a manager.

In the old world (2025), a manager was a therapist. You managed burnout. You managed vacation schedules. You managed that one guy who kept heating up fish in the office microwave.

In Jason’s world, Amelia doesn’t manage feelings. She manages orchestration.

She is a Robot Wrangler.

Her job is to:

• Feed the bots the right data (ingestion).

• Check their work for hallucinations (QA).

• Tweaking their prompts when they start acting weird (training).

It’s like raising a puppy. A very fast, very literal puppy that never sleeps.

At first, you have to watch the puppy constantly so it doesn’t chew on the electric cords (send a hallucinated discount to a client). But unlike a human employee who might get trained up and then leave for a competitor, the AI gets smarter and stays.

The “institutional knowledge” isn’t walking out the door at 5 PM. It’s hard-coded into the system.

For us running our own shops, this is the unlock. We don’t need to hire a junior assistant to offload the boring work. We need to become the Chief AI Officer of our own tiny empires. We need to stop doing the work and start orchestrating the compute power to do it for us.

3. The “Perfect” Trap (And the Rise of the FDE)

I have a bad habit. I like to wait until I have a perfect plan before I start. I will spend three hours researching the “best” font for a website I haven’t built yet.

If I tried to deploy AI sales agents, I would spend three weeks trying to get the workflow perfect before sending a single email.

Jason says this is the fastest way to die.

“The biggest mistake is waiting for it to be perfect.”

The companies winning right now aren’t the ones with the smartest AI models. They are the ones willing to get their hands dirty.

Jason highlights a specific role that is becoming the secret weapon of the AI era: The Forward Deployed Engineer (FDE).

This isn’t just a coder sitting in a dark room. And it isn’t a sales guy in a nice suit.

It’s a hybrid.

An FDE is someone who goes to the client, understands their messy, chaotic data, and manually wires the AI to fix it. They are the bridge between the “magic” of AI and the “reality” of business process.

They don’t just sell the software; they guarantee the outcome by building the integration themselves.

This is counter-intuitive because we are taught that software should be “plug and play.” We want passive income. We want one-click solutions.

But the real money right now isn’t in the software itself—it’s in the configuration of that software.

If you are looking for a way to position yourself, don’t just be the person who uses AI. Be the FDE. Be the person who walks into a mess, untangles the wires, and sets up the machine so it hums.

The gap between “people who use AI” and “people who wait for AI to be ready” is no longer a gap. It’s a different species.

The Bottom Line

We used to ask, “Will AI replace me?”

Jason’s story suggests that’s the wrong question.

The right question is: “Am I the one being replaced, or am I the one holding the leash?”

The middle ground has collapsed. You are either the Top 1% Talent or the Chief Orchestrator.

Choose your fighter.

Links:

1. https://x.com/jasonlk

2. https://www.saastr.com

3. Building a world-class sales org | Jason Lemkin (SaaStr)

4. Replacing Your Sales Team with AI Agents (Jason Lemkin, Founder & CEO @ SaaStr)

5. We replaced our sales team with 20 AI agents—here’s what happened next | Jason Lemkin (SaaStr)

Morning, CEO!

A startup called LMArena announced they raised $150 million at a $1.7 billion valuation.

If you’re keeping score at home, that is a unicorn status for a website that is, essentially, a “Hot or Not” for chatbots.

They don’t build the models. They don’t sell the chat apps. They are the referee.

And usually, the referee is the guy getting yelled at by drunk dads at a Little League game, not the guy walking away with a billion-dollar check.

So, how did a tiny UC Berkeley research project turn into a unicorn in roughly the time it takes me to decide what to make for dinner?

Let’s steal their playbook.

1. The “Memorizing the Eye Chart” Problem

Back in the dark ages (2023), if you wanted to know if an AI was smart, you gave it a standardized test.

We called them benchmarks. They were the SATs for robots. Multiple choice. Math. History. Coding.

The problem is that AI models are professional cheaters.

Imagine a student who refuses to learn biology but manages to memorize the answer key to the final exam.

Question 1: B.

Question 2: A.

The student gets 100%. The teacher thinks the student is the next Doogie Howser. Then you ask the student to perform open-heart surgery, and well... it’s messy.

In the AI world, we call this overfitting.

Because these models are trained on the entire internet, they had often already seen the test questions during training. They weren’t solving the problem; they were remembering the answer.

So, every company started claiming they were “Smarter than GPT-4.” But when I used them to debug my Python script, they hallucinated a library that didn’t exist.

LMArena realized the “Exam” was broken. So they pivoted to the “Thunderdome.”

They built a “Blind Test” arena.

You type a prompt. Two anonymous models generate answers. You vote for the winner.

You can’t cheat on this test because the test is you. The test is 5 million humans asking 60 million weird, messy, unpredictable questions every month.

When the standard metrics (exams) become commodities that everyone can game, the only valuable metric left is raw, subjective user preference. LMArena cornered the market on “what humans actually like.”

2. The “Nano Banana” Pivot

If LMArena was just a leaderboard, it would be a cool Wikipedia page. Not a billion-dollar business.

The real money—the “I’m buying a yacht” money—comes from a boring sounding thing called Routing.

Let me explain with a banana.

A while back, a mysterious model appeared on the leaderboard under the code name “Nano Banana.” It started crushing everyone.

It turned out to be a small, efficient preview model from a major lab.

This highlighted a massive inefficiency.

Right now, most companies are lazy. They pay for the biggest, smartest, most expensive model (like GPT-5.2 or Claude 4.5 Opus) for everything.

It’s like hiring Einstein to do your taxes.

Sure, he can do them. But he costs a fortune. Meanwhile, there’s a guy named Bob (a smaller model) who is actually better at taxes than Einstein, and he costs $5.

But how do you know when to use Einstein and when to use Bob?

LMArena knows.

They have millions of data points on exactly which models are good at exactly which questions.

So they built a product called “Prompt to Leaderboard.” It’s an API. When you send a query, LMArena instantly analyzes it and says:

“Okay, this is a basic summary task. For this, the cheap open-source model is 99% as good as the expensive model. Route it to the cheap one.”

Boom. You just saved 50% on your bill.

They aren’t selling the ranking. They are selling efficiency. In an Agency of One, your value often isn’t doing the work; it’s knowing exactly which tool (or subcontractor) can do the work cheapest and fastest without sacrificing quality.

3. The Death of the “Average” Human

Here is the existential part.

Right now, the leaderboard tells you what the “average” human likes.

But the average human doesn’t exist.

If I ask an AI to “Write a function to sort a list,” I want a dry, efficient code block. I’m an engineer. I don’t want small talk.

If a beginner asks the same question, they want a long, patient explanation with analogies about sorting laundry.

If the AI gives the dry code block to the beginner, the beginner votes “Bad.”

If the AI gives the long explanation to me, I vote “Bad” (and roll my eyes).

Subjectivity is a mess.

LMArena realized that a single “King of AI” is a myth.

So, they are moving toward Personalized Evals.

They want to look at your history and say: “Okay, Hannah likes her answers funny, short, and slightly cynical. She hates emojis. She loves analogies.”

Then, they route your questions to the model that fits your specific brain.

The future isn’t about the “best” model. It’s about the best model for the specific user context. If you can map your client’s specific context better than they can, you win.

The Punchline

We spent the last few years obsessing over who would build the “God Model.”

But while the giants were fighting over who could build the biggest brain, a bunch of students at Berkeley realized something smarter:

It doesn’t matter how big the brain is if it doesn’t make the human happy.

LMArena didn’t win by building AI. They won by building a mirror that reflects what we actually want.

And it turns out, knowing what humans want is worth exactly $1.7 billion.

Links:

1. https://x.com/ml_angelopoulos

2. https://lmarena.ai

3. Beyond Leaderboards: LMArena’s Mission to Make AI Reliable

4. LMArena Co-Founders on the Future of AI Rankings

5. [State of Evals] LMArena’s $1.7B Vision — Anastasios Angelopoulos, LMArena

6. https://www.theinformation.com/articles/ai-evaluation-startup-lmarena-valued-1-7-billion-new-funding-round?rc=epv9gi

What if I told you the smartest person alive spends 70% of his time doing glorified janitorial work?

I recently watched an interview of Terence Tao.

If you don’t know him, he’s basically the Beyoncé of mathematics. The Mozart of numbers. The guy who makes other geniuses feel like they are eating crayons.

He’s transforming a 2,000-year-old “handcraft” industry into a high-speed, automated factory of truth.

And in the process, he’s rewriting the job description for every knowledge worker on the planet.

Including you.

1. The Audit: Identifying the “70% Tax”

We have this romantic idea of “The Expert.”

I picture Terry Tao staring out a rainy window, listening to Bach, waiting for a lightning bolt of inspiration that unlocks the secrets of the universe.

The reality?

Terry admits he spends 70% of his time on “drudgery.”

Formatting papers. Checking citations. Debugging typos. Translating variables.

It’s the intellectual equivalent of scrubbing the toilets so you can host a dinner party.

In your “Agency of One,” you are likely hoarding the drudgery. You think checking the details is “the work.” You think grinding is where the value is.

But Terry realized that manual perfectionism was costing him volume.

He could either be:

A) A perfectionist craftsman who produces one perfect vase every 10 years.

B) An architect of a factory that produces perfect vases daily.

He chose option B. He is aggressively adopting Lean (a coding language for math) and AI to handle the grunge work.

If the Mozart of Math is okay with outsourcing the “hard part” (the rigorous checking) to a machine, why are you still manually formatting your own spreadsheets?

You aren’t protecting your quality. You’re just paying a 70% tax on your own potential.

2. The Scale Problem: Breaking the “Star Graph”

Before AI, collaboration in math was a nightmare.

It looked like a “Star Graph.” One genius in the middle (usually exhausted) and a bunch of helpers reporting back to the genius.

Why?

Because of the Trust Gap.

If Helper A writes a proof, the Genius has to check it manually. Because if Helper A had a bad day, or a hangover, or just sucks at math, the whole project explodes.

Verification was manual. Trust was personal. And that is a massive bottleneck.

Tao is moving to a model of “Trust the Code, Not the Person.”

By using formal verification software (and AI to write it), he doesn’t need to know you. He doesn’t need to trust you.

If your code compiles, your logic is true.

This turns math into LEGOs. I can build a brick. You can snap it onto your spaceship. We never have to speak.

We usually think business relationships must be built on “trust.”

But actually, trust is friction.

Trust requires coffee chats, vetting, and reputation management. Scalable systems rely on verification.

As you build your operation, stop trying to hire people you “trust.” Build systems that verify the output so you can work with anyone.

3. The Pivot: From “Rigorous” to “Post-Rigorous”

This is my favorite part because it makes me feel better about being a chaotic thinker.

Terry has a framework for the three stages of learning:

• Stage 1: Pre-Rigorous (The Drunk Toddler). You have intuition, but you’re wrong half the time. You think you can fly. You cannot.

• Stage 2: Rigorous (The Nervous Lawyer). You learn the rules. You follow them perfectly. You lose your creativity because you are so terrified of making a mistake.

• Stage 3: Post-Rigorous (The Jedi). You know the rules so well you can ignore them. You return to intuition, but now it’s accurate.

Most mid-career pros get stuck in Stage 2.

We are obsessed with being “Rigorous.” We want to be right. We obsess over the details because that’s how we got promoted.

Terry sees AI as the ultimate Stage 2 Employee.

AI loves rules. It loves syntax. It loves checking for errors. It is the ultimate Nervous Lawyer.

By letting the AI handle the Rigor, you might feel lazy. You aren’t doing the heavy lifting anymore.

But the upside? You get to skip straight to Stage 3.

You get to be the Jedi.

Don’t compete with AI on “correctness.” You will lose.

Compete on “vision.”

The goal of your Agency isn’t to be the best rule-follower. It’s to be the one deciding which game we’re playing.

The Bottom Line

We are witnessing the Industrial Revolution of Logic.

Terry Tao isn’t trying to be the smartest worker on the factory floor anymore. He’s designing the assembly line.

Your job this week?

Find your “70%.” Find the rigorous, boring, necessary tasks that make you feel productive but are actually just keeping you busy.

Fire yourself from those tasks. Hire the robot.

And go be the Architect.

Links:

https://terrytao.wordpress.com

https://github.com/teorth

Terence Tao on the cosmic distance ladder

Terence Tao: Hardest Problems in Mathematics, Physics & the Future of AI

Terry Tao on the future of mathematics | Math, Inc.

While normal people were making New Year’s resolutions they definitely won’t keep, the team at DeepSeek was busy rewriting the laws of physics for AI.

On New Year’s Eve, they dropped a paper that solves a problem so annoying it has plagued AI engineers for a decade. By fixing it, they didn’t just stabilize their models; they created a blueprint for scaling intelligence without losing your mind.

If you want to know how to handle massive growth without crashing—whether you’re a neural network or a human managing three freelance clients and a needy golden retriever—read on.

Here is how we steal their playbook.

1. The “Single-Lane” Problem

To understand what DeepSeek did, you have to look at how we usually build these things.

For the last decade, the industry standard for building AI has been the Residual Connection.

Think of it like a very reliable, strictly regulated corporate mailroom.

In the old days, messages (signals) would get lost or garbled as they moved up the chain of command. The Residual Connection fixed this with a “Safety Rule”: “You can process the mail, but you must strictly preserve the original package and pass it along with your changes.”

This ensures the signal never dies. It flows from Layer 1 to Layer 100 perfectly intact. It is safe. It is reliable. This is why tools like ChatGPT exist today.

But it has a bandwidth problem.

Because the standard Residual Connection treats the information flow like a single, protected stream, it limits how much information can flow at once. It’s like a polite, single-lane road. The cars (data) move smoothly, but you can only fit so many cars on the road before you hit a limit.

Engineers wanted more. They didn’t just want a safe road; they wanted a massive logistical network.

2. The “Toddler on Espresso” Trap

So, the engineers got ambitious. They thought, “Let’s turn this single lane into a twelve-lane superhighway!”

They called this Hyper-Connections (HC).

The idea was to widen the path so more information could flow at once. Total freedom! Maximum bandwidth! Everyone talking to everyone!

I tried this strategy in my 20s. I said “yes” to every project, every social event, and every hobby.

The result for me was a panic attack in a grocery store.

The result for DeepSeek? Their AI had a mathematical panic attack.

When they opened up these “Hyper-Connections” without any rules, the signal inside the model exploded. The amplification factor hit 3,000x.

Imagine whispering a secret to a friend, and by the time it reaches the last person in the line, they are screaming it through a megaphone while setting off fireworks.

The model became unstable. The “gradient norm” (basically the AI’s cortisol level) went haywire.

It turns out, unconstrained freedom isn’t genius. It’s just noise.

3. The Pizza Law of Sanity

DeepSeek realized they needed the best of both worlds: the massive bandwidth of the superhighway, but the safety and stability of the old single-lane road.

So, they dug up a piece of math from 1967 called the Sinkhorn-Knopp algorithm. They applied a new rule called Manifold-Constrained Hyper-Connections (mHC).

I know. “Manifold-Constrained” sounds like something that happens to a spaceship in Star Trek.

But the concept is actually hilarious. I call it the Pizza Law.

They forced the model’s wide connections to behave like Doubly Stochastic Matrices. Here is what that means in plain English:

• The Giver Rule: You (a neuron/channel) have exactly one pizza of energy to give. You can give a slice to Channel A and a slice to Channel B. But you cannot give everyone a whole pizza. Your total output must equal 100%.

• The Receiver Rule: You can only eat exactly one pizza. If 50 people try to feed you, you have to shrink their slices down until they all fit on one plate. Your total input must equal 100%.

That’s it.

By forcing every connection to follow the Pizza Law, they created a closed loop of energy conservation.

The result? That 3,000x signal explosion dropped to a polite 1.6x. The model stopped screaming and started thinking. It retained the capacity of the superhighway but regained the stability of the single lane.

The Takeaway

We usually assume that Scale = More Freedom.

“If I widen the road, I can drive however I want.”

DeepSeek proves the counter-intuitive truth: Scale requires stricter Constraints.

They successfully built a massive “12-lane superhighway,” but it only worked because they mathematically forced every car to stay in its lane (the Pizza Law). Without this constraint, the signal didn’t just grow; it exploded into noise (3,000x amplification).

They paid a tiny 6.7% “stability tax” in speed to unlock scalability.

This is the ultimate lesson for any complex system, whether it’s a neural network or your career:

Unconstrained freedom is just chaos waiting to happen.

To go wide, you must first lock down your rules.

Go define your Pizza Law.

Links:

1. https://www.arxiv.org/abs/2512.24880

While I spent the last week of December trying to determine if “leftover pie” is a legitimate breakfast food (the data suggests: yes), Mark Zuckerberg was busy rewriting the laws of physics for our careers.

Meta dropped billions to buy Manus.

Why does this matter to you, sitting there with your coffee?

Because the rules of the game just changed. And if we don’t update our internal operating systems, we’re going to be playing Chess while the rest of the world is playing Laser Tag.

Let’s dig in.

1. The Death of the “Smartest Person in the Room”

For the last few years, we’ve been obsessed with IQ.

We looked at AI like it was a super-genius Professor trapped in a glass box. We asked it hard questions. We marveled at its essays. We felt insecure because it knew more about 14th-century French poetry than we did.

But here’s the thing about the Professor in the glass box: He can’t open the door.

Manus—the company Zuck just bought—didn’t try to make the Professor smarter. They gave the Professor hands.

They realized that Action Intelligence is worth infinitely more than Chatty Intelligence.

Manus doesn’t just talk about market research; it opens a browser, clicks the buttons, fights the pop-ups, downloads the spreadsheet, and makes the graph.

The Wisdom for You:

Stop trying to be the “Smartest Person in the Room.”

In your current role, you might be tempted to hoard knowledge. To be the one who knows the obscure Excel formula or the historical context of the Q3 project.

That asset is depreciating faster than a new car driven off a cliff.

The market no longer pays a premium for the person who knows the answer. They pay for the person who executes the solution.

We are moving from an economy of Explanation to an economy of Completion.

If your deliverables are just “really good ideas” or “thorough analysis,” you are the Professor in the glass box.

Break the glass. Don’t just suggest the strategy; build the prototype. Don’t just analyze the problem; deploy the fix.

Be the hands, not just the mouth.

2. Inefficiency is the New Luxury

Here is a weird thing about Manus.

It is incredibly inefficient.

To do a simple task, the AI might loop fifty times. It checks its work. It realizes it clicked the wrong link. It goes back. It tries again. It burns through computing power like a bonfire burns dry pine.

A year ago, everyone said, “That’s too expensive! You’re wasting resources!”

But the founders made a counter-intuitive bet: Compute is cheap. Trust is expensive.

They realized that a client doesn’t care if you wasted 10,000 units of energy to get the job done. They care that the job is done right and they didn’t have to babysit you.

The Wisdom for You:

We have been trained to be efficient. To minimize effort. To get things done “quick and dirty.”

But in a world of infinite, cheap labor (AI), Reliability becomes the ultimate luxury good.

If you are running your career like a business, stop obsessing over how fast you can do a task. Start obsessing over the fidelity of the result.

If you turn in work that is 90% there, you are creating a “management debt” for your client. They have to check it. They have to fix that last 10%.

That 10% is annoying. It creates friction.

The “Manus Mindset” is to burn your own energy—your own “compute”—to vigorously self-correct, verify, and polish before you ever hit send.

Be the person who creates Zero Friction.

If your boss knows that when they hand you a ball, you will not drop it—even if you have to run in circles five times to catch it—you become invaluable.

Inefficiency in the process is fine, as long as it guarantees perfection in the outcome.

3. The “Conductor” Theory of Value

Manus had about 100 employees when it was bought. It was generating $100 million a year.

That math shouldn’t exist.

How? Because 80% of their code was written by AI.

The humans weren’t “coders” in the traditional sense. They were Commanders.

They weren’t playing the instruments; they were conducting the orchestra.

This is the scary part for those of us who pride ourselves on our craft. We like “doing the thing.” We like writing the sentence, designing the slide, writing the function.

But the leverage has shifted.

The Wisdom for You:

The most valuable skill in 2026 isn’t Creation; it’s Volition.

Volition is the will. It is the ability to decide what matters.

If an army of digital interns can do 80% of the work, your value is entirely contained in the remaining 20%.

And that 20% isn’t “checking for typos.”

That 20% is Taste and Direction.

• Taste: Knowing what “good” actually looks like so you can tell the AI when it’s wrong.

• Direction: Knowing which mountain is worth climbing in the first place.

You need to stop viewing yourself as the “Worker Bee” who is judged by how much pollen you carry.

You are the Queen Bee. You are judged by where you tell the hive to go.

If you don’t have a strong opinion on where your project/department/company should go, you are liable to be replaced by a script that can run faster than you.

Develop a point of view. That is the one thing the machine can’t generate.

The Punchline:

The era of “I’m valuable because I know things” is over.

The era of “I’m valuable because I get things done” has begun.

The tools will change next month. But the human who has Taste, Reliability, and the Guts to make a decision?

That human is never going out of style.

Now, if you’ll excuse me, I have to go ask an AI to help me figure out how to burn off this holiday pie.

Links:

1. https://zh.wikipedia.org/wiki/肖弘

2. https://manus.im

3. https://www.bbc.com/news/articles/ce3k11q9qe1o

4. https://www.wsj.com/tech/ai/meta-buys-ai-startup-manus-adding-millions-of-paying-users-f1dc7ef8

I spent most of my weekend feeling proud that I reorganized my dishwasher loading strategy. It felt like a big win for efficiency.

Then I read an interview with Mike Krieger, the co-founder of Instagram and current CPO of Anthropic.

He casually dropped a statistic that made my inner Career Crisis Committee call an emergency meeting:

At Anthropic, AI now writes 90% of their production code.

Read that again. 90%.

While I was feeling smug about stacking plates better, the engineers building Claude are using Claude to do almost all of their actual work.

We are looking at a fundamental shift in the economics of effort.

Here is the playbook on how Krieger is navigating 2026, and how we can steal it to stop doing grunt work and start charging for our brains.

1. The “How” is Dead. Long Live the “Vibe.”

For the last 20 years of my career, the biggest monster under my bed was the “How.”

I want to build an app that runs on my parents’ phones, records their stories, and writes memos for our family archive.

The AI part? Easy.

The app part? Nightmare. Dealing with microphone permissions, UI layouts, and building an APK file sounds like a form of torture I am not willing to endure on a Saturday.

So the idea dies, and the stories stay untold.

The “How” was a giant, expensive wall standing between my Brain and Reality.

Krieger says the cost of “How” has dropped to roughly zero.

He calls this the era of “Vibe Coding.”

You don’t need to know the syntax. You don’t need to remember where the semicolon goes or why the computer is yelling at you.

You just need the vibe—the intention, the logic, the “thing you want to happen.”

Mike shared a story. He needed a “Secret Santa” app for his family. In the old world, he wouldn’t build it. The friction was too high. His brain’s strict "Low Power Mode" would have kicked in and convinced him to just draw names out of a hat.

In this new world? He built the app over breakfast. While cooking eggs.

The Strategy for Us:

If you are still telling your boss that you can’t solve a problem because you “don’t have the technical skills,” you are lying to them.

And worse, you’re lying to yourself.

The barrier isn’t code anymore. It isn’t budget. It’s just inertia. The new skill isn’t typing; it’s having the guts to describe what you want and hitting “Enter.”

2. The Bottleneck Paradox

There is a concept called The Theory of Constraints.

It basically says: A system can only go as fast as its slowest part. If you widen the bottleneck, the flow speeds up until it slams into a new, different bottleneck.

So, if AI writes 90% of the code instantly, where did the bottleneck go?

Krieger says it moved upstream. The new bottleneck is Taste.

At Anthropic, they realized generating code was easy. The hard part was the “Merge Queue”—the line to get that code approved.

They had a firehose of AI-generated work and a tiny straw of human approval.

This is the counter-intuitive insight: When creation is free, curation becomes the most expensive resource in the universe.

Think about your current role.

If you are the person who do the work (writing the email, crunching the Excel sheet), you are in the danger zone. That work is going to zero.

But if you are the person who decides what work gets done, and verifies if it’s actually good? You just became 10x more valuable.

The Strategy for Us:

Stop trying to be the best bricklayer on the construction site.

Be the foreman.

Your value to your “client” is no longer your ability to grind out output. It is your ability to look at 50 AI-generated options and say, “That one. That is the one that won’t get us sued. Ship it.”

Taste is the new technical skill.

3. From “Chatting” to “Actual Work”

If 2024 was the year of the Chatbot (”Write me a poem about a potato”), and 2025 was the year AI came for the Junior Developer’s job...

What is 2026?

Mike’s prediction: 2026 is the year AI reliably takes work off your plate.

Notice the word reliably. And notice the difference in the relationship.

• A tool is a hammer. You have to swing it.

• A colleague is someone you email and say, “Hey, handle this,” and then you go to lunch.

Mike talks about “Horizontal Agents.” This is the boring-sounding term for the most exciting shift in our careers.

It means the AI isn’t just trapped in a little chat window on the side of your screen like a digital pet. It’s logging into your Salesforce. It’s fixing the messy data on the legacy server from 2019. It’s navigating the bureaucracy.

It is the difference between asking an intern, “How do I do this?” and telling a Senior Manager, “Go fix the Q3 sales report, email it to the board, and don’t bother me.”

This explains the “Productivity Paradox” we’ve all felt. You know, that study that says AI hasn’t made companies more productive yet?

It’s because we are treating it like a toy. We are sprinkling AI on top of broken processes.

The Strategy for Us: To win in 2026, you have to stop chatting with the AI and start delegating to it. You have to give it permission to enter your systems and do the dirty work.

To recap for the road:

1. Vibe Code: The barrier to building is zero. Stop making excuses.

2. Cultivate Taste: Be the editor, not the writer. Be the architect, not the bricklayer.

3. Delegate, Don't Chat: Move from swinging the hammer to directing the crew.

I’m going to go “Vibe Code” a script to organize my Spotify playlists. Or maybe I’ll just ask Claude to explain my dishwasher to me.

Links:

1. https://www.linkedin.com/in/mikekrieger

2. https://claude.ai

3. 30-Minute Masterclass on Product Thinking | Instagram Co-Founder & Anthropic CPO, Mike Krieger

4. Anthropic’s CPO on what comes next | Mike Krieger (co-founder of Instagram)

5. Anthropic CPO: How AI Will Build the Next $100M Companies | Mike Krieger

6. How AI Starts Doing the Work in 2026 with Anthropic CPO Mike Krieger

Morning, CEO!

Here is a number that should make you sweat: 40,000 hours.

That’s how long NVIDIA’s new AI model, NitroGen, spent watching YouTube videos to learn how to function.

Usually, if a company wants to build a “unicorn” product (or just a functional robot), they burn cash hiring humans to label data. It’s the artisanal, hand-crafted approach to AI—slow, expensive, and difficult to scale.

But NVIDIA achieved a massive result—training an AI to master 1,000+ different environments—by spending practically $0 on manual data labeling.

They didn’t work harder. They just looked where no one else was looking.

Here is the playbook on how they turned other people's Saturday night entertainment into a business asset, and how you can steal it.

1. The “Digital Exhaust” Strategy

Here is the biggest problem with AI right now: It’s a Brain in a Jar.

It has read every book in the library, so it knows the theory of everything. But it has no hands. It doesn’t know how to do anything.

If you want an AI to play a video game (or run your inbox, or navigate a spreadsheet), you need to show it Cause and Effect. “When the screen looks like this, my finger pushes that.”

The “Enterprise” way to solve this is to hire a thousand interns to record their keystrokes. Expensive. Boring. Soul-crushing.

NVIDIA’s researchers did something smarter. They asked: “Where is this data already happening for free?”

They realized that on Twitch and YouTube, hard-core gamers use “Gamepad Overlays”—little graphics in the corner of the screen that light up when they press buttons.

To the viewers, this is just a cool visual.

To the researchers, this was the Rosetta Stone.

They wrote a script to watch 40,000 hours of gameplay, crop out that little gamepad, and use it to teach the AI exactly what buttons to press for every situation.

The Playbook:

They didn’t create new data. They harvested “digital exhaust.”

As an Agency of One, you are likely sitting on a mountain of “gamepad overlays.”

• Do you record your Zoom calls?

• Do you have a folder of “Sent” emails?

• Do you have commit logs?

You think it’s trash. But that is your proprietary dataset. Stop trying to “build” a process document from scratch. Scrape your own exhaust.

2. The “Street Smarts” Pivot

I have a confession. I am terrible at video games.

If I memorize a level in Super Mario, I can beat it. But if you drop me into Halo, I will spend 20 minutes walking into a wall before falling off a cliff.

Old AI was like me. It could beat the world champion at Go, but it couldn’t play Checkers to save its life. It was a specialist.

NitroGen is different. Because NVIDIA fed it 1,000 different games—from 2D platformers to 3D open worlds—it didn’t just memorize maps.

It learned intuition.

It learned the “physics of winning.”

• Red bar getting low? Run away.

• Spikes on the floor? Jump.

• Shiny thing? Pick it up.

The results are wild. Because NitroGen had already seen 1,000 other games, it had a massive head start. When tested on a brand new 3D game, it was up to 52% more successful than a model starting with a blank slate.

The Playbook:

Stop selling yourself as a “Specialist.”

In the freelance world, we are told to “niche down.” Be the “Email Marketer for Gluten-Free Bakeries in Austin.”

But the real leverage comes from Transfer Learning. The most valuable “Me Inc.” businesses are the ones that learn the underlying physics of a problem (Sales, Operations, Strategy) and can drop into any client’s messy “game” and figure it out 52% faster than the new guy.

Be a Generalist with a library of solutions.

3. Embrace the “Monkey Brain”

The human brain has two systems (shout out to Kahneman).

System 1: Fast, instinctive, “The Monkey.” (Dodging a ball).

System 2: Slow, logical, “The Professor.” (Doing taxes).

We usually think AI is “The Professor.” We want it to write code and solve cancer.

But the NitroGen paper explicitly says: This model is a System 1 thinker. It doesn’t plan. It doesn’t contemplate the philosophical implications of shooting a zombie. It just reacts.

It sees a frame of video, and it outputs an action. Immediately.

And honestly? That is where the money is.

Most of us are drowning in work because we are trying to use our human “Professor Brain” for “Monkey Brain” tasks.

• Sorting emails? Monkey task.

• Formatting data? Monkey task.

• Scheduling meetings? Monkey task.

NVIDIA proved that if you have enough data, you can train a model to handle the high-speed, low-context reactions perfectly.

The Playbook:

Don’t try to get AI to be the CEO of your life yet. It’s not ready to do the “Deep Work” planning.

Instead, build a “System 1” layer. Automate the reactions. Train your tools to handle the immediate inputs (client emails, bug reports, invoice requests) so your poor, tired Professor Brain can actually think for five minutes.

To recap:

NVIDIA looked at 40,000 hours of teenagers screaming at screens and saw the future of robotics. That is the level of resourcefulness we need.

Stop waiting for the perfect client or the clean dataset. Look for the “overlays” in your current mess.

The map to your next level of leverage is probably hiding in the corner of your screen, right where you aren’t looking.

Links:

1. https://nitrogen.minedojo.org

2. https://nitrogen.minedojo.org/assets/documents/nitrogen.pdf

3. https://github.com/MineDojo/NitroGen

4. https://huggingface.co/nvidia/NitroGen

5. https://huggingface.co/datasets/nvidia/NitroGen

Morning, CEO!

Here’s a number that made me spill my third cup of coffee: 100 Million RMB (about $14M USD).

That’s how much funding Sengine Technology raised.

Here’s the kicker: Their main product—a game called SenBOX—hasn’t even officially launched yet. But it already has millions of fans and is generating revenue.

The founder, Zidong Liu, is a Gen Z Ph.D. dropout who looked at the AI landscape, saw Google and OpenAI fighting to the death, and said, “Nah, I’m good.”

Today, we are stealing his playbook on how to win by refusing to fight.

1. Stop Polishing the Old Battlefield

If you’re anything like me, you probably spend a decent chunk of your week trying to get slightly better at things that everyone else is also doing.

We optimize our email subject lines. We tweak our Midjourney prompts to get the lighting just right. We try to squeeze 5% more productivity out of a Tuesday.

Zidong Liu calls this “fighting on the Old Battlefield.”

He looked at the history of the internet like a ladder of “Information Density”:

• Text (1.0): Blogs. Low bandwidth.

• Images (2.0): Instagram. Prettier, faster.

• Video (3.0): TikTok. The current king.

Right now, every major AI company is bludgeoning each other trying to win the Video war. It’s a bloodbath.

Liu realized that if he tried to build a “better video generator,” he’d just be another casualty. So he went to the Next Continent: Space.

Not outer space. 3D Space.

While everyone is fighting over pixels, he’s betting on Voxels (3D pixels). He realized that “Space” is the only modality left that offers higher dimension and interaction than video.

The CEO Takeaway:

Look at your current “business” (your job). Are you trying to be the best at a skill that is currently being commoditized? Are you fighting trench warfare?

Stop trying to do the “Old Modality” 10% better. Find the “New Modality” where no one is looking yet. Be the only person in your department using AI to generate video updates while everyone else is still typing memos.

2. The “Skeleton” vs. The “Skin”

Here is where my brain did a little backflip.

Usually, when we think of 3D AI, we think of “making things look pretty.” We think of textures, lighting, and rendering. We think of the Skin.

Liu thinks this is a trap.

There are massive labs (like Fei-Fei Li’s World Labs) working on visually perfect 3D scenes.¹ But Liu argues that a pretty room is useless if the AI doesn’t understand how humans actually live.

If you ask an AI to “generate a living room,” a visual model might give you a stunning image where the sofa is facing the wall and the TV is on the ceiling. It looks real, but it’s logically insane.

Sengine ignores the Skin and focuses on the Skeleton:

1. What do we place?

2. Where do we place it?

He treats furniture layout not as a design problem, but as a Recommendation System problem. Just like TikTok recommends the next video based on your viewing history, Sengine recommends the perfect spot for that armchair based on flow, privacy, and sunlight.

The CEO Takeaway:

In your work, are you focusing on the “Skin” (the deliverables, the slide deck design, the final report)?

The value is in the “Skeleton” (the logic, the structure, the why). AI can generate the skin instantly. The person who owns the skeleton owns the project.

3. The “Rainforest” Strategy (Or: How to Trick Robots)

This is my favorite part. It’s devious.

Liu wants to train robots. Specifically, domestic robots that can fold laundry and find your lost keys.

The problem? Data scarcity.

Self-driving cars have endless hours of GoPro footage from roads. But we don’t have cameras inside millions of people’s bedrooms (thankfully). So, how do you train a robot to navigate a messy bedroom without crashing?

He built a video game.

SenBOX is basically an AI-powered Minecraft or The Sims. Users—mostly young women—spend hours designing their dream homes, placing furniture, and creating layouts.

They think they are playing a game.

Liu knows they are generating training data.

He calls this the “Rainforest Strategy.”

• SaaS is a desert where only the biggest survive (Winner Take All).

• Gaming is a rainforest. There are elephants (AAA games), but there are also millions of ants (indie games). Even the ants eat well.

He builds a game to survive and make cash. The game generates the data. The data trains the “Spatial Intelligence” model. Then, he sells that intelligence to B2B giants like Panasonic and robot manufacturers.

He treats the B2B revenue as a “side hustle” while the game keeps the lights on.

The CEO Takeaway:

What is the “Game” you can play that generates value for your “Real Mission”?

Maybe you start a newsletter (The Game) to force yourself to learn AI (The Mission). The newsletter might make a few bucks, but the data (the skills you learn) is what gets you the promotion or the big contract.

The Bottom Line

Zidong Liu didn’t win by fighting harder. He won by looking at the map, seeing where the crowd was heading, and walking the other way.

He bet on Space over Video. He bet on Logic over Looks. He bet on Play over Pitching.

I’m going to go stare at a wall now and think about my life choices. Or maybe I’ll just go play SenBOX and pretend it’s market research.

Links:

1. Zidong Liu

2. https://sengine.ai/home

3. https://sz.ifeng.com/c/8h4iqdSIxac

4. https://www.douyin.com/video/7572100056153689394

5. https://mp.weixin.qq.com/s/f2OkjTa2qQh4CuP6EwSaFQ

Morning, CEO!

Imagine staring at an extinction-level event.

That was Fedor Pak in late 2022. He took over a chatbot company right before ChatGPT dropped. His product became obsolete overnight.

Most people would curl into a ball and weep. Fedor? He pivoted.

By December 2025, his new venture, Chatfuel Clinic Suite, hit $1M ARR.

He didn’t just survive the AI meteor; he built a bunker, filled it with gold, and sold tickets.

Here’s the playbook we’re stealing today.

1. Your “Product-Market Fit” Has an Expiration Date (And It Smells Like Old Milk)

I’m 41. I know a thing or two about things that used to work suddenly not working anymore (my knees, my tolerance for loud bars, my ability to stay awake past 10 PM).

Fedor learned this the hard way.

Chatfuel was the king of “if-this-then-that” chatbots. They had 1.5 billion conversations in the bank. They were the cool kids.

Then Generative AI arrived. Suddenly, “rules-based bots” looked like a horse and buggy on a Formula 1 track.

Fedor realized that resting on your laurels is a great way to get crushed.

He didn’t try to fight the wave. He went back to square one.

He literally walked into a small, three-room dermatology clinic. He watched the owner drowning in WhatsApp messages.

The owner wasn’t treating skin. He was playing schedule Tetris.

The Steal for You:

Your current job description? It’s probably perishable.

The skills your boss hired you for three years ago might be the equivalent of that rules-based chatbot.

Don’t wait for the meteor.

Audit your services today. Are you offering 2022 value in a 2025 market? If so, find your version of the “dermatology clinic.” Find the new pain point. Pivot before you’re pushed.

2. Nobody Wants to Buy a Drill (They Just Want the Hole in the Wall)

I have a toxic trait.

When I build something cool with AI, I want to explain the architecture. I want to tell you about the vector database, the RAG pipeline, and the context window.

I want you to be impressed by the how.

Fedor made this mistake too.

Early on, he tried to sell “AI capabilities” to doctors. He talked tech.

The doctors looked at him like he was speaking Klingon.

Nothing landed.

So, he stopped talking about the drill (AI). He started talking about the hole (Revenue).

He changed his pitch to: “I will stop you from losing 35% of your revenue to no-shows.”

Suddenly, wallets opened.

He realized the market rewards clarity, not complexity. He stopped selling “Automation” and started selling “Peace of Mind.”

The Steal for You:

Stop telling your boss/client how you used AI to finish that report.

They don’t care about your prompt engineering. They don’t care that you used Claude Opus 4.5 or GPT-5.2.

They care that the report is done 3 days early. They care that you found a $10k cost saving.

Market the outcome, not the tool. Be the person who solves the headache, not the person who explains the chemistry of Aspirin.

3. The “Centaur” Strategy (Or: How to Not Hate Your Job)

I am an introvert. If the doorbell rings unexpectedly, I don't answer it; I drop to the floor and army-crawl to the kitchen.

So, the idea of AI handling communication sounds like heaven.

But Fedor’s approach wasn’t about replacing humans because humans are annoying. It was about replacing humans because humans are expensive and distracted.

He built a system where AI handles the low-leverage junk:

• Scheduling? AI.

• Reminders? AI.

• “Where are you located?” questions? AI.

This wasn’t to fire the doctors. It was so the doctors could actually doctor.

The AI handles the admin layer so the human can handle the empathy layer.

Fedor calls it the “Business Autopilot.”

The Steal for You:

Look at your workday.

How much of it is “Admin Drudgery”? Scheduling meetings, summarizing emails, formatting data?

That is low-leverage work. It keeps your “Me Inc.” running, but it doesn’t grow the business.

Your goal is to build your own “Autopilot.”

Use AI to strip away the 30% of your job that makes you want to scream.

Reinvest that time into the stuff that actually gets you promoted (or keeps you sane enough to binge British detective shows, which is my personal KPI).

To Recap:

1. Pivot Fast: Your skills expire. Refresh them before they stink.

2. Sell Outcomes: Don’t explain the tech. Explain the win.

3. Automate Drudgery: Let AI be your secretary so you can be the CEO.

Fedor turned a crisis into $1M.

You can at least turn it into a Friday afternoon off.

Links:

1. https://www.linkedin.com/in/fedor-pak-chatfuel

2. https://chatfuel.com

3. https://www.aitoolssme.com/interviews/fedor-pak

4. https://medium.com/authority-magazine/fedor-pak-of-chatfuel-5-ways-to-use-marketing-automation-to-improve-team-efficiencies-ba7ff34d6ab9

5. https://www.starterstory.com/stories/chatfuel

Morning, CEO!

I consider myself a creature of habit. I like my tea hot, my code commented, and my British detective series running for at least 10 seasons.

Then I listened to Elena Verna.

Elena is a growth legend. She led growth at Dropbox, Miro, and Amplitude. She knows everything.

And yet, after joining Lovable (which hit $200M ARR in under a year), she admitted that only 30% to 40% of her decades of experience actually applies anymore.

The rest? She had to throw it in the trash.

If a growth titan has to burn her playbook, we probably should too.

Here is how Elena is navigating the chaos, and how we can steal her brain.

1. Stop Polishing the Door Handle (Optimization is Dead)

In her past lives, Elena spent 95% of her time “optimizing.”

You know the drill. Tweaking the funnel. Changing the button color from blue to slightly-more-urgent blue. Shaving 0.2 seconds off a process.

At Lovable? She says that’s a waste of time.

She now spends 95% of her time on Innovation and only 5% on Optimization.

Why? Because the market is moving too fast. If you spend six months optimizing a feature, by the time you’re done, AI has rendered that entire feature useless.

I feel attacked.

My comfort zone is optimization. I love organizing my desktop folders. I love color-coding my calendar. It feels like work, but it’s actually just “procrastination with better lighting.”

The Takeaway for Us:

Stop optimizing your current job description.

If you are an “Agency of One,” you can’t spend your day making your existing tasks 1% faster. That’s a race to the bottom because AI will eventually do those tasks 100% faster for free.

You need to ship new features.

Don’t just format the weekly report better. Build a new dashboard that predicts next week’s problems. Don’t just answer emails faster. Build an agent that drafts the answers for you.

Optimization is maintenance. Innovation is equity.

2. Your Tenure Resets Every 90 Days

Elena dropped a bomb that made me want to crawl under my weighted blanket.

She said: “You have to re-earn Product-Market Fit every three months.”

In the old days (2022), you found a fit, and then you scaled it for five years. You rode the wave.

But in AI, the underlying tech changes every quarter. User expectations skyrocket. What was “magic” in Q1 is “boring” in Q2.

If Lovable—a company growing faster than a bacterial culture in a petri dish—has to panic about relevance every 90 days, so do we.

We love the idea of “Tenure.” We think, “I’ve been here 10 years, I know where the bodies are buried, I am safe.”

Elena’s logic suggests that tenure is now a liability. It makes you slow. It makes you defend the “old way.”

The Takeaway for Us:

Your boss doesn’t care what you did in 2019.

Treat your employment contract like a renewable 3-month subscription.

Every quarter, ask yourself: “If I had to pitch my services to this company today, from scratch, would they hire me?”

If the answer is “maybe,” you are already churning.

You need to re-launch “Me Inc.” every season. New capabilities, new offers, new value.

(I know. I’m tired too. I need a nap.)

3. Hire Yourself for “Chaos Clarity”

When Elena hires for her team, she doesn’t look for people who fit a neat little box.

She looks for “High Agency” people who can “create clarity from chaos.”

She says fast-moving AI companies don’t have stable roadmaps. They don’t have instruction manuals. If you need to be told exactly what to do, you will drown.

She explicitly looks for people who don’t wait for permission.

This is the ultimate “Agency of One” mindset.

Most employees wait for the ticket. They wait for the email. They wait for the “all hands” meeting to tell them the strategy.

But in a chaotic market, the person who waits is the person who gets left behind.

Elena is saying that value is created by the people who walk into the mess, grab a shovel, and say, “I’m going to build a path right here.”

The Takeaway for Us:

Stop asking your boss for clarity. Sell them clarity.

Don’t go to them asking, “What should I do about this AI thing?”

Go to them and say, “I looked at the chaos, and here are three ways we can use AI to fix it. I’ve already started on number one.”

That is High Agency. That is how you become indispensable.

To recap:

1. Stop Optimizing: Polishing a dying workflow is a waste of time. Build something new.

2. The 90-Day Rule: Your skills have a shelf life of a banana. Refresh them constantly.

3. Eat Chaos for Breakfast: Don’t wait for instructions. Create clarity.

I’m going to go create some clarity (by which I mean, organize my fridge).

Links:

1. https://x.com/elenaverna

2. https://lovable.dev

3. Building Lovable: $10M ARR in 60 days with 15 people | Anton Osika

4. Lovable CEO, Anton Osika: The State of Foundation Models, Grok vs OpenAI, and Replit vs Bolt

5. The new AI growth playbook for 2026 | How Lovable hit $200M ARR in one year

Morning, CEO!

Today we’re looking at Kaijie Chen. He’s 26. He’s on his third startup. His latest creation, Macaron AI, racked up 300,000 users and churned out 200,000 user-generated apps in just a few months.

Meanwhile, I’m over here celebrating because I remembered to cancel a free trial before my credit card got hit. We are not the same.

But that’s okay. We don’t need to be the 26-year-old prodigy; we just need to ruthlessly steal his strategy for our own operations. Let’s look at how he’s building a personalized AI empire—and how we can use his logic to upgrade “Me Inc.”

1. Stop Reading the Manual, Start Whacking the Ball

Here is the biggest trap I fall into—and I bet you do too. I call it “Procrastination by Preparation.”

We read the books. We watch the tutorials. We consume the strategy. We feel productive because our brains are filling up with data.

Kaijie Chen calls this the “Pre-training Era.” It’s how early AI models were built. You feed the machine the entire internet, and hope it gets smart. But in 2025? We’ve run out of internet. The data is exhausted. Reading more isn’t making the models smarter.

So, Chen pivoted to the “Era of Experience.”

He uses this tennis analogy that hit me right in the gut:

You can watch 10 hours of Roger Federer playing tennis. You will intellectually understand the game.

But if you pick up a racquet and swing once? Even if you miss? That single second of failure contains more data than the 10 hours of watching.

Your arm feels the weight. Your eyes judge the depth. Your brain gets an immediate “Error! Adjust Left!” signal.

This is Reinforcement Learning (RL). It’s learning through the feedback loop of doing, failing, and adjusting.

The Takeaway for Us:

Your “Agency of One” is stalling because you’re still pre-training. You’re over-consuming information.

Chen’s entire tech stack is built on High Information Gain. He doesn’t want more data; he wants better data. The only way to get that is to enter the arena.

Stop researching the perfect cold email strategy. Send ten terrible emails today. The rejection you get is the “Reward Model” signal you need to optimize the next ten.

Stop planning the perfect project roadmap. Ship the messy beta.

The market provides the “RL,” but only if you provide the swing.

2. Your “Vibe” is the Moat (Product Taste)

We live in a world where everyone has access to the same “Intellectual Capital.”

You have ChatGPT. Your competitor has ChatGPT. I have ChatGPT (and a mild caffeine addiction).

If the raw intelligence is commoditized, why does anyone pick Macaron AI over the giants like OpenAI or Google?

Taste.

Chen realized that most AI tools are just cold, efficient assistants. They sit there, blinking, waiting for a command. They are reactive.

He built Macaron to be Proactive, Dynamic, and Vibrant.

If you tell a standard AI you’re sad, it lists three local therapists. Efficient? Yes. Cold? Absolutely.

If you tell Macaron you’re sad, it remembers that you like hiking. It proactively suggests a trail. It checks in on you the next morning. It acts like a friend, not a search engine.

This is what Chen calls “Product Taste.” It’s the difference between a tool and a partner.

The Takeaway for Us:

In your professional life, your “Deliverables” are likely commodities.

The report you write? AI can do it. The code you ship? AI can generate it. The strategy deck? AI can outline it.

If you compete on utility, you lose.

You have to compete on Taste and Proactivity.

Don’t be the freelancer who waits for the client to ask for the status update. Be the partner who anticipates the anxiety and sends the update before they ask.

Don’t just deliver the spreadsheet; deliver the insight about what the numbers mean for their specific goals.

Your “Agency of One” needs a personality. It needs to be opinionated. It needs to care.

Intelligence is cheap. “Giving a damn” is the premium feature.

3. Don’t Just Do Work; Build Systems That Do Work

This is the part that blew my mind.

Macaron isn’t just a chatbot. It’s a factory.

Users don’t just chat with it; they tell it to build “Mini-Apps.”

A user says, “I want to track my calories by taking photos of my food.” Boom. Macaron writes the code and generates a mini-app for that specific purpose.

A user says, “I need to analyze my spending to save for a house.” Boom. A finance tracking app appears.

Chen isn’t trying to build every feature himself. He built a system that allows non-technical users to build their own tools.

He has 300,000 users, but they’ve created 200,000 apps. That is a staggering ratio of creation.

He is moving from “Creation” (him building the app) to “Usage” (users building the value).

The Takeaway for Us:

We need to stop treating AI as a typewriter that writes faster, and start treating it as a developer that builds tools.

If you find yourself doing the same task three times, you are failing as a CEO.

Don’t write the email response.

Build the prompt chain that analyzes the incoming email and drafts the response in your voice.

Don’t manually summarize the meeting notes.

Build the workflow that records, transcribes, summarizes, and emails the action items to the client automatically.

You are not the worker bee. You are the architect of the hive.

Chen’s users are building “Mini-Apps” to solve their life problems. You need to build “Mini-Workflows” to solve your business problems.

Build the asset once. Let it run forever.

To Recap:

1. Stop watching tennis. The only data that matters comes from swinging the racquet and hitting the net. Ship the work.

2. Vibe check yourself. Your utility is replaceable. Your “Taste” and proactivity are not. Be a partner, not a tool.

3. Build the factory. Don’t just do the task. Build the mini-app (workflow) that does the task for you next time.

Kaijie Chen is betting that AI’s future isn’t about bigger brains, but about better experience.

I think he’s right.

Now I’m going to go ask an AI to build me a mini-app that reminds me to stop doom-scrolling and go to bed.

Links:

1. https://x.com/KaijieChen12236

2. https://macaron.im

3. https://36kr.com/p/2610259787456641

4. https://36kr.com/p/3559477201484678

5. https://mp.weixin.qq.com/s/mGYTK8s4fjBm46zh703acg

Evening, CEO!

Let’s talk about ElevenLabs.

You know them. They are the reason your Instagram feed is full of motivational quotes read by a voice that sounds suspiciously like Morgan Freeman but definitely isn’t.

In about three years, they went from “two guys in a London apartment” to a multi-billion dollar unicorn that basically owns the voice AI market.

Meanwhile, in that same three-year timeframe, I have successfully:

1. Managed to keep a single fiddle-leaf fig tree alive (barely).

2. Overthought 4,000 messages.

We are not the same.

But that’s fine. Because today, we are going to crack open the brain of their CEO, Mati Staniszewski, and steal the operating system he used to build this empire.

We are going to look at how he turned specific constraints into massive leverage. Because if a couple of guys can revolutionize audio while sitting in a flat in London, you can probably optimize your Q4 deliverables without having a nervous breakdown.

Here is the playbook.

1. Your “Crappy” Background is Your Secret Weapon

There is a temptation, especially in tech, to assume that if you aren’t in a coffee shop in San Francisco eavesdropping on OpenAI engineers, you have already lost.

Mati proves this is nonsense.

In fact, ElevenLabs exists because Mati isn’t from San Francisco. He’s from Poland.

Now, here is a fun fact about Poland that I just learned: Voiceovers there are insane.

In the US, if you watch a dubbed movie, you get a full cast. The hero sounds like a hero. The villain sounds like a villain.

In Poland, traditionally, they used a “Lector.”

This is one guy. Just one dude. usually speaking in a monotone, flat voice.

He reads the hero’s lines. He reads the grandma’s lines. He reads the alien’s lines.

It is a terrible user experience. It strips all emotion out of the art.

But if Mati had grown up in California, he never would have noticed this problem. He would have taken high-quality dubbing for granted. Because he suffered through the “One Guy” problem, he understood deep in his bones that the world needed AI that could capture emotion, not just words.

The Lesson for Us:

Stop trying to copy the “cool kids” in your industry.

Your leverage comes from the weird, specific, annoying things unique to your situation.

That manual spreadsheet process you hate? That client who asks the same question 50 times? That is your “Polish Lector.” That is the pain point you understand better than anyone else. Solve that, and you become irreplaceable.

2. The “Slider of Shame” (Or: How to Stop Overthinking)

I am a recovering perfectionist.

If I need to post a 15-second Instagram Reel, I will spend 30 minutes debating whether the font “Courier New” conveys enough irony, or if I need to download a new font pack. I end up posting nothing.

ElevenLabs had this problem too.

Early on, users kept asking for a simple feature: A slider to control how fast the voice speaks.

Simple, right? Just give them a speed knob.

But the ElevenLabs team was full of purist researchers. They said, “No! That is crude! The AI model should be smart enough to intuit the correct speed based on the emotional context of the text!”

So they spent nine months trying to teach the AI to guess the speed.

Nine months.

The users were annoyed. The researchers were stressed. And the AI still wasn’t getting it quite right.

Finally, they gave up. They stopped trying to be geniuses and just built the slider into the product interface.

It took a fraction of the time. The users were thrilled.

Mati turned this failure into a heuristic I am now taping to my forehead: The 3-Month Rule.

If the “Research” (the deep, perfect, automated solution) can’t solve the problem in 3 months, the “Product” (the manual, slightly hacky, human intervention) has to take over.

The Lesson for Us:

There is a battle inside your brain between the Architect and the Builder.

The Architect wants to spend six months designing a perfect, fully automated, AI-powered content machine.

The Builder just needs to get a post out by 5:00 PM today.

Don’t let your inner Architect hold the project hostage. If you can’t automate it perfectly today, just build the “Slider.” Do it manually. Use the hacky fix. Ship the B+ work so you can live to fight another day.

3. Paying Commission on the Deal You Killed

This is my favorite part of the story because it hurts my brain in a good way.

ElevenLabs has a sales team. Salespeople like money. They get money through commissions.

One day, a massive competitor—a huge Foundation Model company—came knocking. They wanted to license ElevenLabs’ voices for their demos.

For the sales team, this looked like a jackpot. Huge contract. Massive commission check.

But Mati realized something: Strategy is a Leading Indicator. Commission is a Lagging Indicator.

If they sold their voices to a competitor, they were arming their enemy. They would get a pile of cash today, but they might lose their business tomorrow.

So Mati killed the deal. He said “No.”

Normally, this is the part where the sales team riots and burns down the office because they lost their bonus.

But Mati did something genius: He paid them the commission anyway.

He realized that the sales team did their job. They found a customer. It wasn’t their fault that the strategy required saying no. If he stiffed them, they would stop bringing him opportunities. They would start hiding things.

The Lesson for Us:

In your “Me Inc,” you are constantly negotiating between Short-Term You (who wants the easy win/money/dopamine) and Long-Term You (who wants a sustainable career).

Sometimes, you have to turn down a project that pays well but ruins your reputation.

Sometimes, you have to say no to a “promotion” that is actually just 50% more work for 0% more pay.

When you make those hard strategic choices, you have to find a way to reward yourself anyway.

Don’t punish yourself for making the hard, right choice. If you skip the “easy win” to protect your long-term value, buy yourself the fancy coffee. Take the Friday afternoon off.

Pay yourself the commission, even if you kill the deal.

To Recap:

1. Embrace your “Outsider” status. Your specific frustrations are your roadmap.

2. Stop researching, start shipping. If you can’t perfect it in 3 months, use the “Slider.”

3. Protect the Asset. Don’t sell your future for a quick win today (but reward yourself for the discipline).

I am off to go water my fiddle-leaf fig. It’s looking at me with judgment.

Links:

1. https://uk.linkedin.com/in/matiii

2. https://elevenlabs.io

3. ElevenLabs CEO/Co-Founder, Mati Staniszewski:The Untold Story of Europe’s Fastest Growing AI Startup

4. ElevenLabs CEO: Why Voice is the Next AI Interface

5. Mati Staniszewski at ElevenLabs Summit 2025 in San Francisco

Evening, CEO!

While I was debating if changing from “day pajamas” into “night pajamas” counts as productivity, Eline van der Velden was busy dismantling the entire business model of Hollywood.

She’s the founder of Particle6.

She managed to cut production costs by 60%, turned an “actress” who doesn’t physically exist into a global news sensation, and is currently running circles around traditional studios who are still trying to figure out how to convert a PDF.

Here is how she stopped worrying and learned to love the bomb (AI), and how we can steal her playbook.

1. Stop Waiting for the Bus (It broke down in 2012)

Here is a visual representation of my early career strategy:

Me, sitting on a bench. Waiting for a fancy person in a suit to walk by, hand me a bag of money, and say, “Hannah, we have been waiting for your genius.”

Spoiler: The bench is cold. The suit never comes. And I just get older and hungrier.

Eline calls this “Waiting for the Bus.”

In the TV world, producers spend their lives writing pitches and praying that a commissioner at the BBC or Netflix picks them. They are gatekeeper-dependent. They are playing the lottery, but the ticket costs six months of unpaid work.

Eline looked at the timetable and realized the bus wasn’t coming.

The budgets are shrinking. The risk appetite is zero.

So, she built a car.

She realized that if she used AI to handle the heavy lifting—pre-visualization, editing, background generation, lighting planning—she didn’t need the gatekeeper’s massive budget.

She could go directly to the client (or the audience) and say, “I can do this for half the price, twice as fast, and it will look like a million bucks.”

The Playbook:

Stop asking for permission to do your job.

If you are waiting for a “boss” or a “client” to give you the resources to do your best work, you are going to die on that bench.

Use the tech to lower your own cost of delivery so drastically that you don’t need their permission to start.

2. It’s Not a Terminator, It’s a Calculator

I often have this sinking feeling that I arrived at the tech party just as everyone else is packing up, and now the robots are here to take the snacks. It’s a heavy feeling.

It’s a valid fear.

But Eline has a different perspective. She compares AI in 2025 to Excel in the 1980s.

When spreadsheets arrived, accountants panicked. They thought, “Well, that’s it. The machine can do the math. I guess I’ll go live in the woods and eat berries.”

But accountants didn’t go extinct. The ones who insisted on using an abacus did. The ones who learned Excel stopped doing basic arithmetic and started doing financial strategy.

Eline treats AI as the “Calculator for Creatives.”

She doesn’t let it replace the human “desire” to tell a story (because AI has the emotional depth of a teaspoon). She uses it to handle the drudgery.

• Need a location scout? AI does it.

• Need to visualize a shot list? AI generates the storyboards.

• Need to fix a lighting error? AI patches it in post.

This moves her up the food chain. She isn’t a “camera operator” or an “editor” anymore. She is the Architect of the Vision.

The Playbook:

If you are still billing for “doing the math” (the manual execution), you are in the danger zone.

Shift your value proposition. Be the person who knows what to put into the spreadsheet, not the person manually adding up the columns.

3. The Onion Theory of Infinite Leverage

This is where my brain actually melted a little bit.

Eline created an AI actress named Tilly Norwood.

Tilly has been interviewed, featured in global news, and offered roles in major movies.

Tilly is data.

Eline didn’t just generate a pretty face on Midjourney and call it a day. She built an asset using what she calls the Onion Theory.

• Layer 1 (The Skin): They generated 2,000 images to get the perfect look. Freckles, skin texture, imperfections.

• Layer 2 (The Meat): Movement and voice. How she walks, how she laughs.

• Layer 3 (The Soul): This is the kicker. They built a backend database of her personality, humor, and backstory.

Tilly isn’t a drawing. She is a Digital Twin.

Here is why this is genius:

A human actor can be on one set, at one time. If they get the flu, production stops. If they have a meltdown in a trailer, you lose money.

Tilly can be in 50 movies simultaneously. She can work 24 hours a day. She never needs a latte. She scales infinitely.

I, personally, cannot scale. I need a nap after a Zoom call.

Eline has moved from selling labor (hours on set) to selling assets (IP that works while she sleeps).

The Playbook:

What is the “Tilly Norwood” of your business?

What can you build once—a framework, a software tool, a content library, a custom GPT—that can “perform” for fifty clients at once while you are eating a sandwich?

To recap:

The bus isn’t coming, so stop waiting on the bench.

The AI isn’t here to kill you; it’s here to do the heavy lifting so you don’t throw out your back.

And if you want true leverage, stop selling your sweat and start building onions.

I’m going to let the AI organize my file structure while I go stare at a tree. It’s called delegation.

Links:

1. https://uk.linkedin.com/in/eline-van-der-velden-2ab8285a

2. https://www.particle6.com

3. Eline van der Velden on How AI Is Changing TV Production Forever

4. Talent Agents Circle AI Actress Tilly Norwood As Studios Quietly Embrace AI Technology

5. Eline van der Velden on Tilly Norwood, AI Actors & The Future of Creativity