The Moat Is You

In 2015 I spent six months and a small fortune building an MVP. Four people. Backend, frontend, mobile, and a project manager whose primary contribution was scheduling meetings about meetings. We had standups. Sprint planning. Code reviews. Retrospectives about the retrospectives. We burned through a serious budget before a single user touched the product.

Last month I built something comparable in 12 hours. Alone. With AI agents writing 80% of the code while I focused on architecture and the parts that actually matter.

I'd like to say I'm not bitter about it, but I absolutely am.

Six months to 12 hours. That's not an improvement. That's a different universe wearing the same t-shirt. And if you're building software right now, as a founder, an engineer, or a company betting on a "technical moat," this essay is about what happens next.

---

I remember when this was hard

The year is 2015. You want to build a SaaS product. Here's what that looks like:

• $50,000 to $150,000 in development costs for a basic MVP
• 4 to 6 months with a small team (2 to 5 devs, a designer, a PM)
• US agencies charge $100 to $250/hour per specialist
• A pricing page with Stripe integration? That's 2 to 3 days of developer time.

You raise a seed round just to find out if anyone wants the thing. Most of them didn't. But at least the process felt appropriately painful.

Fast forward to 2026. That same pricing page with Stripe? 45 minutes. The full MVP? A weekend. Cost? Maybe $5,000 if you're being generous with yourself at the coffee shop.

Pieter Levels built a flight simulator MMO in 30 minutes using AI. It scaled to hundreds of thousands of users and generates $50K/month. He runs a $3M/year portfolio with zero employees. Zero. Not "lean team." Zero.

25% of YC's Winter 2025 batch had codebases that were 95%+ AI-generated. Not by non-technical founders fumbling through prompts. By highly technical people who simply didn't need to write the code themselves anymore.

The compression is here. Building is solved. Everything downstream just changed.

---

This has happened before

Here's the thing nobody in tech wants to hear: this isn't new. The exact same pattern has played out at least seven times in the last 500 years. Every single time, the people in the middle of it thought their situation was unprecedented.

They were wrong. The pattern is always the same.

The printing press (1450)

Before Gutenberg, copying a single Bible took a scribe 136 working days. After the press, the Ripoli Press in Florence could produce 1,025 copies for the cost of three hand-copied ones. A 341x cost reduction.

What happened to the scribes? Obsolete within a generation. Their entire "unique skillset" was just... gone by end of century.

What emerged? The Protestant Reformation. Luther's 95 Theses spread from Wittenberg to London in 17 days. The Church's monopoly on knowledge interpretation was broken. Not by theology. By a mechanical device that made copying cheap.

Nobody saw the Reformation coming. They were too busy worrying about scribes.

The new moat after the printing press[+]

Photography vs. painting (1840s)

Portrait painters charged fortunes and took weeks. Then the daguerreotype arrived: 1 to 15 minutes, $2 to $6 by 1853. In 1849 alone, 100,000 photographic portraits were produced in Paris.

Documented casualties: Justus Dalee, miniature portraitist, went into the grocery business. Royall Brewster Smith became a carpenter. Rufus Porter pivoted to house murals and founded Scientific American. At least Porter landed on his feet. The others got groceries.

But here's the part nobody tells you. Impressionism was born because of photography. The first Impressionist exhibition in 1874 was held in the studio of Nadar, one of Paris's most famous photographers.

When cameras made reproduction trivial, painters were freed to do what cameras couldn't: capture light, emotion, movement, subjectivity. Art moved up the stack. It always does.

The bedroom Grammy (2020)

Michael Jackson's Invincible cost $30 to $40 million to produce. Billie Eilish's When We All Fall Asleep, Where Do We Go? swept the four biggest Grammy awards. Recorded in her brother's bedroom on under $3,000 of equipment. A pair of $200 Yamaha monitors, a $100 mic, and Logic Pro X.

Bon Iver's For Emma, Forever Ago went platinum. Recorded alone in a cabin with two guitars and a single SM57 microphone.

The result? 120,000 new tracks uploaded to streaming services every single day. More music released in one day in 2024 than in the entire year of 1989. But only 14,700 DIY artists earned over $10,000 from streaming in 2022.

Production became free. Finding an audience became the hard part. Sound familiar?

The pattern

Every time creation costs collapse, the same seven phases play out:

1. Cost collapse. 10 to 1,000x cheaper overnight.
2. Incumbent destruction. Craftsmen and gatekeepers wiped out.
3. Volume explosion. Supply goes parabolic.
4. Quality democratization. Amateurs match pros.
5. Discovery crisis. Too much supply, attention becomes scarce.
6. New art forms emerge. Things nobody predicted.
7. Value migrates up the stack. From execution to taste.

Scribes to authors. Painters to artists. Studios to bedroom producers. Typesetters to graphic designers. In every case, the people who survived weren't the fastest at the old thing. They were the first to understand the new thing.

---

But this time...

I know. "This time it's different." The four most dangerous words in investing. But hear me out, because some things are different.

Previous compressions commoditized physical creation. Copying books. Capturing light. Recording sound. This one commoditizes the tool of commoditization itself. Software is the thing we use to make other things cheaper. And now software itself is becoming free.

That's a recursive loop. And the data backs it up.

The SAASpocalypse

On February 3, 2026, Anthropic released Claude Cowork with 11 enterprise plugins. The market erased $285 billion in market cap in a single day. Thomson Reuters fell 16%. LegalZoom fell 20%. Salesforce, ServiceNow, Adobe, each down 7%.

A quarter trillion dollars vanished because an AI company shipped a product update. Let that sink in.

Morgan Stanley identified three fears driving the crash:

1. "Do it yourself" software. Companies will build their own rather than buy.
2. Model providers rendering apps obsolete. AI agents as intelligent interfaces that make distinct "apps" disappear.
3. The seat model dying. If AI agents do the work, you don't need 100 Salesforce seats. You need 10.

Jason Lemkin put it plainly: "If 10 AI agents can do the work of 100 sales reps, you don't need 100 Salesforce seats anymore. That's a 90% reduction in seat revenue for the same work output."

At SaaStr itself, 20 AI agents managed by 1.2 humans now do the work previously handled by 10 SDRs and AEs. I love the 0.2 of a human. Presumably that's someone who checks Slack every other Tuesday.

Steve Yegge's dead companies walking[+]

The recursive compression

Here's what makes software compression different from the printing press:

• AI makes AI better. Each generation of models produces better code, which builds better tools, which trains better models. The compression compounds.
• The "seat" dies. Previous compressions killed products. This one kills the billing model. When agents do the work, per-seat pricing collapses.
• Software eats itself. Marc Andreessen said software would eat the world. It did. Now AI is eating the software that ate the world. It's cannibalism all the way down.

Sam Altman frames it sharply: "Build things where you're hoping the model gets better, not where you're praying it doesn't."

If your moat depends on AI staying dumb, you don't have a moat. You have a prayer.

---

The abundance problem

So everyone can build. What happens next?

Look at music. 120,000 tracks a day. More music than any human could listen to in a lifetime, uploaded every single day. And only 14,700 independent artists earn more than $10,000 a year. That's not a career. That's a side hustle with Spotify characteristics.

Look at YouTube. 500+ hours of video uploaded every minute. Somewhere between 21% and 33% of the feed is now AI-generated content. Billions of views, captured by mass-produced slop. Meanwhile, original creators struggle to be discovered under the avalanche of "10 SHOCKING AI Facts" videos narrated by a voice that has never been shocked by anything.

The same thing is coming for software. The app stores and SaaS directories are already flooding with AI-generated wrapper apps. Product Hunt launches that took months now take afternoons. The number of products went up. The number of products anyone cares about stayed the same.

This is the abundance paradox: more software than ever, but harder to find software that matters. A Cambrian explosion of decent products, none of them remarkable. A thousand B+ apps drowning in each other's mediocrity.

Ben Thompson's Aggregation Theory predicted this. When distribution costs go to zero, value shifts from controlling supply to controlling demand. The aggregators (Apple, Google, Spotify, YouTube) win by owning the relationship with consumers. The producers become commoditized.

It happened to musicians. It happened to publishers. It's happening to software builders now. We're not special. We just thought we were.

---

The moat is you

So if code is free, features are copyable, and your product can be cloned in a week with a good prompt, what's left?

You.

Not metaphorically. Literally. The research is clear, and it echoes 500 years of compression history. Every single time creation got commoditized, value migrated to the same place: human judgment.

Taste

When everyone can build, knowing what to build is the differentiator.

Billie Eilish didn't win Grammys because bedroom recording was new. She won because she had taste. A vision for what the music should feel like that no amount of studio equipment (or lack thereof) could produce.

When cameras killed portrait painting, the painters who survived weren't the ones who tried to out-photograph the camera. They were the Impressionists. The ones who understood that art's job had changed.

In software, taste means: which problem is worth solving? What should the product feel like? Where is AI output "good enough" and where does it need a human hand? The engineers who can answer these questions are worth more than they've ever been. The engineers who can only write code are worth less than they've ever been. The irony is not lost on me.

The 80/20 that defines everything[+]

Distribution

If your product can be replicated in a week, the moat is who already trusts you. Who knows your name. Who will choose you over the 47 identical alternatives in the search results.

Cursor beat GitHub Copilot, a product backed by Microsoft's distribution machine, through pure product-led growth. $1B ARR, 2.1 million users, zero ad spend. They won on taste and speed, not on distribution. But that's the exception that proves the rule. Most of us aren't building developer tools where the product is the distribution.

For everyone else: your email list, your Twitter following, your community, your reputation in a specific domain. That's the moat. Not the code. The code is a prompt away from existing. Your audience is not.

Domain expertise

Paul Graham: "The durable moats are human: trust, community, brand, proprietary data. These take time to build and can't be replicated overnight."

The non-technical founder with 15 years in construction who builds an AI-powered estimating tool has a moat that no vibe-coding teenager can replicate. Not because the code is complex, but because the knowledge is deep. The teenager can build the UI in an afternoon. The teenager cannot know which line items subcontractors always lie about.

The most successful AI companies emerging right now are vertical, not horizontal. They don't build "AI for everything." They build AI for radiology, for contract review, for supply chain logistics. The code is commodity. The domain understanding is the moat.

Vertical AI startups captured over $1 billion in combined funding in early 2025 alone. AI funding accounted for 50 to 53% of all venture capital in 2025. The money is flowing to domain depth, not technical breadth. VCs finally figured out that "we built it with AI" is not a business model. "We know things you don't" is.

Speed of iteration

Not just building fast. Learning fast. Adapting fast. Shipping, measuring, adjusting, shipping again.

The 10x engineer becomes the 100x engineer not by writing more code, but by having better judgment about what code should exist. By orchestrating agents rather than typing characters. By running five experiments where a competitor runs one.

When building is free, the bottleneck becomes the feedback loop. How fast can you go from idea to user to insight to next idea? The companies that compress that cycle, not just the build cycle, are the ones that win.

Trust

This is the sleeper moat. When AI can generate anything (websites, products, content, legal documents) how do you know what to trust?

The SAASpocalypse recovery happened when Anthropic partnered with trusted incumbents like FactSet and Thomson Reuters. The market stabilized when it saw that AI would augment established trust, not replace it.

In a world of infinite AI-generated content and AI-built products, a trusted name, a person, a brand, a community, becomes the filter. The signal in the noise. Being trustworthy is now a competitive advantage. What a time to be alive.

---

What this means for you

Here's what I'd tell 2015-me, the one spending six months and a serious budget on that MVP:

The tools got cheaper. The taste didn't.

Anyone can build a SaaS app now. Just like anyone can record an album, shoot a film, publish a book. The tools are free or nearly free. The gap between "has an idea" and "has a working product" has collapsed to hours.

But the gap between "has a working product" and "has something people choose, trust, and pay for"? That gap is wider than ever. Because the competition just went from "other people who can code" to "everyone."

So build the thing. Use AI. Ship it in a weekend. But know that the weekend build is table stakes now. Everyone's doing it.

The moat is the thing AI can't generate: your taste, your domain expertise, your distribution, your reputation, your speed of learning, your judgment about what should exist in the world.

The moat is you.

And honestly, it's the only moat that's ever been real.