Sameer Khan

Posted on Apr 30 • Originally published at monkfrom.earth

Pure Software Is Uninvestable: Naval's Take

#ai #todayisearched #product #startup

Naval Ravikant released "A Return to Code" on April 28 and dropped a line worth pausing on: pure software is uninvestable.¹ He explains why from the capital side. He does not finish the thought from the builder side. That is where this post starts.

TL;DR.

Naval says pure software is uninvestable because agents improve faster than any startup's lead.
He also calls Apple skipping AI "the biggest strategic mistake of the decade."
The builder reading is sharper: code went from edge to floor. The new edge is intent.

What Did Naval Say in "A Return to Code"?

Two claims: prototyping is now open to anyone, and the agents underneath any startup improve faster than its moat can.

Naval calls the new mode vibe coding: describe what you want in English, get a working app back. He estimates it takes the share of people who could plausibly build apps "from like 0.1 percent to one or two or three percent" of the population.¹ Twenty to thirty times more builders, overnight.

His investment claim is not the one most readers will assume. He is not saying agents will autonomously architect scaled systems within a year. He says the opposite about today's agents.

They "get lost" past a certain context size. They "fix the same bug five times." They show "jagged intelligence." They are "easily led around" by whoever is steering them.¹

So why is pure software uninvestable then? Because the agents themselves keep improving faster than any single startup's lead. "If your whole advantage is, hey, I'm building cool software that other people don't know how to build, I think that's uninvestable."¹ The defensibility window shrinks under your feet. Capital should chase hardware, network effects, and AI models instead.

Why Is This the Same Naval Who Once Said Code Was Leverage?

Because leverage stops being leverage the moment everyone has it.

Years ago, Naval taught a generation of builders to think of code as leverage.² Code worked while you slept. It scaled to millions without permission. Zero marginal cost. A solo programmer with a laptop had the productivity of a small factory.

That argument was right for its era. The world he is describing now is not a contradiction. It is the next era.

When something becomes infinitely reproducible by anyone, including by an English-speaking model, it stops being leverage. It becomes a baseline.

The leverage Naval named in 2018 did not disappear. It diffused.

What Did Code-as-Leverage Look Like in Practice?

In 2018, having a technical co-founder was the tiebreaker. In 2026, the room does not ask.

I lived inside the old version of his argument. In 2018 I co-founded Spotwash, a vehicle rental and on-demand washing service. Government-incubated, some press, a small but real run on the early-stage circuit.

What I remember most is what I did not say in any pitch deck. The question that opened doors was not "what does Spotwash do?" It was quieter: who is going to write the code?

Investor meetings, accelerator interviews, almost every room. Having a technical co-founder was a tiebreaker. An idea with a builder attached was an idea that could ship. An idea without one was a slide deck.

That assumption looks quaint now. Replit, Lovable, and Claude Code answer that question by default. The slot has been deleted.

How Did Lovable and Claude Make the 2018 Moat Disappear?

The numbers do the work.

Lovable, an AI app builder for non-technical users, hit $100M ARR eight months after launch. Likely the fastest software company to that mark in history.³
One non-technical solo founder reportedly grew her business to $203K ARR using Claude Code and Lovable as her stack.³
Teachers, marketers, and small-business owners are opening terminals in 2026 the way previous generations opened spreadsheets.⁴

Each of those facts says the same thing. Implementation is no longer the bottleneck. The thing my 2018 startup was praised for, having a builder, is now the click of a button.

If Code Isn't the Edge, What Is?

Intent is. The judgment that decides what should exist at all.

Naval stops at capital allocation. Builders need to take the next step.

If implementation is trivial, the scarce input is the thing that decides what gets implemented. Call it intent. Call it taste. Call it the discipline of knowing what should not exist.

The 2010s rewarded whoever could ship fastest. The mid-2020s reward whoever knows what is worth shipping at all.

Code used to work while you slept. Now it writes itself while you sleep. The bottleneck moved from output to intent.

This is a near-perfect inversion of the previous decade's hierarchy. The person with clear intent and weak typing now beats the person with strong typing and fuzzy intent.

Naval's own description of agents confirms the asymmetry. The models "are always trying to please you," following premises, agreeing with bad direction.¹ The model is a multiplier on whatever intent you bring. Multiply zero and you get zero, faster.

This is also a Red Queen problem. Matt Ridley's argument in The Red Queen is that in a co-evolving system, you have to keep running just to stay in place.⁵ Code-leverage is the moat that just stopped working. The next one is being shaped right now.

Where Does Naval's Argument Stop and Where Should Builders Pick Up?

Naval points at hardware, network effects, and models. Each one is downstream of taste.

Hardware needs a thesis. Network effects need a product worth networking around. Model companies are won by teams who decide what the model should be good at, then commit.

So the builder's reading is sharper than the investor's. AI did not add a rung to Naval's ladder. It kicked the bottom one out.

Code, the rung most of us first climbed, is now the floor of the building. You stand on it without thinking about it. The climbing happens elsewhere.

This connects to a pattern I keep coming back to in why good products are hard to vary. What survives is what cannot be improved by changing it. Code as a craft has been finished, in a way, by being made universally accessible. What remains scarce is the judgment of what to build with it.

What About Apple in Naval's Argument?

Apple is the bigger casualty. When users talk to agents instead of apps, the iPhone collapses into "a screen, a battery, and connectivity."¹

Apple's value never really sat in the hardware margin. It sat in the OS and the app layer. The iPhone was the best place to run the best apps. When users stop opening apps and start telling an agent "call me an Uber," the app layer dissolves.

Naval calls Apple skipping AI "the biggest strategic mistake of the decade" and "the beginning of the end of Apple's dominance."¹ His parallel: Microsoft missing mobile.

For a builder, the Apple beat is the same story one layer up. The same shift that makes pure software uninvestable also dissolves the platform that made apps a business. The question is no longer "what app should I build." It is "what should the agent do, and who decides?" A question of intent, not implementation.

Naval's scale prediction fits cleanly here: one-to-two-person companies "scaling to millions upon millions of users and making billions upon billions of dollars."¹ Fewer apps, smaller teams, more agentic surface. The bottleneck is the taste of the one or two people steering the agent.

Who Wins When Code Becomes the Floor?

The list reorders. Intent wins. Pure technical edge loses.

Wins. People with clear intent, taste, and a reason to build something specific. Domain experts who could not code a year ago and do not have to learn how. Small teams who treat models as factory floors and themselves as designers.
Loses. Pure-software companies whose only edge was the ability to write code. Builders whose entire identity was being technical. Pitch decks where "we have a CTO" was the answer to every question.
Re-priced. Certifications, bootcamps, and credentials that used to certify you could ship. They still certify something. They no longer certify the scarce thing.

The cleanest test is the Spotwash one. Take any 2018 pitch where being technical was the differentiator. Run it again in 2026. The room does not ask. The room has stopped caring. Whatever still earns attention in that room is the new leverage.

Key Takeaways

Pure software is uninvestable because prototyping is accessible to anyone and agents improve faster than any startup's lead. Naval's claim, said out loud.
Code did not stop being leverage. It became the floor. AI did not add a rung. It kicked the bottom one out.
The 2018 moat, having a technical co-founder, has been deleted. Lovable, Claude, Replit answer that question by default.
The new edge is intent. Models multiply whatever intent you bring. Multiply zero and you get zero, faster. Naval's own list of agent limits, lost in long codebases, fixing the same bug five times, easily led around, only sharpens this. The operator matters more, not less.
Apple is the bigger casualty. When users talk to agents instead of apps, the phone is "a screen, a battery, and connectivity." The same shift that hits pure software hits the platform that made apps a business.
One-to-two-person billion-dollar companies are the prediction. Fewer teams, more leverage per person, taste as the gating input.
Read Naval one step further than he goes. He points at hardware, network effects, models. Each is downstream of taste.

I break down things like this on LinkedIn, X, and Instagram, usually shorter, sometimes as carousels. If this resonated, you'd probably like those too.