For fifty years, the hardest part of building a software company was building the software.
That era is ending.
I am not talking about a marginal improvement. I am talking about a structural shift in what is difficult and what is not. The thing that used to take six engineers three months now takes one person a weekend. The thing that required a $2M seed round to prototype now requires a $20/month API subscription and a clear idea.
This changes everything. Not because software stops mattering - it matters more than ever. But because the bottleneck moves. And most people are still optimizing for the old bottleneck.
The old world: shipping was the wall
Think about what "building a startup" meant in 2015. You had an idea. You needed to hire engineers. You needed to manage sprints, argue about architecture, debug production issues, maintain infrastructure, handle deployments, write tests, review pull requests. The gap between "I know what to build" and "it exists and works" was enormous.
This gap created entire industries. Project management tools. CI/CD platforms. Code review systems. Agile consultants. Boot camps that promised to turn you into a developer in twelve weeks. Venture capital that existed primarily to fund the engineering labor required to cross the gap.
The gap also created a power dynamic. Engineers were the bottleneck, so engineers had leverage. Technical cofounders were the most sought-after people in startup ecosystems. "I have a great idea but I need a technical cofounder" was the defining sentence of a decade of entrepreneurship.
The gap was real. It was expensive. And it is collapsing.
What actually happened
AI did not replace programmers. It did something more interesting: it made the act of programming dramatically less expensive in time, effort, and skill required.
I build software every day. The difference between how I worked two years ago and how I work now is not incremental. It is categorical. Things I used to plan for a week, I now ship in an afternoon. Entire features - database schema, API routes, frontend components, tests - materialize in a single session. Not because I got smarter. Because the cost of translating intention into working code dropped by an order of magnitude.
This is not theoretical. I currently run five products simultaneously. Not because I have a team of fifty engineers. Because the cost of building and maintaining software fell far enough that one person with good judgment and the right AI setup can operate at a scale that used to require a small company.
The bottleneck moved. And it moved fast.
So what is the new bottleneck?
If shipping is no longer the hard part, what is? I think there are four things that become dramatically more valuable in a world where code is cheap.
1. Knowing what to build
This was always important. But when building was expensive, you could differentiate just by being able to build at all. "We have engineers who can execute" was a competitive advantage. That advantage is evaporating.
What remains is the question that was always underneath: is this the right thing to build?
Most software that gets built should not exist. It solves problems nobody has, or solves real problems in ways nobody wants, or solves the right problem for the wrong audience. The hit rate has always been low. But when building was expensive, bad ideas died early because they ran out of funding before they could ship. Natural selection via capital constraints.
Now? Bad ideas can ship in a weekend. The filter is gone. Which means the world is about to be flooded with software that nobody needs - built fast, deployed instantly, abandoned quickly. The people who win are not the ones who ship fastest. They are the ones who have the judgment to know what is worth shipping at all.
Problem discovery becomes the rare skill. Not "I can build an app" but "I understand a specific group of people well enough to know what they actually need, even when they cannot articulate it themselves."
2. Taste
When everyone can build, quality of judgment about how to build becomes the differentiator.
I do not mean pixel-perfect design, though that matters. I mean the thousand micro-decisions that determine whether a product feels right or feels off. Where to put the button. What to leave out. When to add complexity and when to resist it. How the error state should feel. What the onboarding sequence communicates about who this product is for.
These are taste decisions. They cannot be automated because they require understanding human psychology, cultural context, and the specific emotional register of the audience you are building for. AI can generate a hundred variations of a landing page. It cannot tell you which one will make a compliance officer in Frankfurt feel like this product was built specifically for their world.
Taste was always valuable. But in the old world, it was often subordinate to execution speed. "We do not have time to get the details right, we need to ship." That excuse is gone. If shipping is fast, there is no reason not to get the details right. And the companies that get the details right will destroy the ones that do not.
3. Distribution
This is the one most people miss.
Building software is approaching zero marginal cost. But getting that software in front of the right people is as hard as it has ever been - arguably harder, because there is more noise.
Every channel is saturated. Google is full of SEO content. Social media algorithms change monthly. Cold email is drowning in AI-generated outreach. App stores are flooded. The attention economy was already brutal. In a world where anyone can ship software, it becomes apocalyptic.
The people who win the next decade are not the best builders. They are the best distributors. The ones who have audiences, communities, trust networks, brand equity, channel partnerships. The ones who can put a product in front of the right thousand people on day one.
This is why I think the most undervalued skill in tech right now is marketing. Not growth hacking. Not viral loops. Actual marketing - understanding who your customer is, what they care about, where they pay attention, and how to earn their trust over time.
Distribution has always been the real moat. Software being cheap just makes that moat deeper.
4. Integration and context
Here is something counterintuitive: as individual pieces of software become easier to build, connecting them becomes harder. Not technically harder - APIs and protocols handle that. Harder in the sense that understanding what should connect to what, and why, requires deep context about how an organization actually works.
The CTO who knows that the compliance team's spreadsheet feeds into the finance team's quarterly report, which drives the board's investment decisions, which determines whether the engineering team gets headcount - that person cannot be replaced by AI. Because they hold the organizational context that makes technical decisions meaningful.
As software becomes commoditized, the value shifts to the people who understand the systems that software operates within. The business context. The regulatory environment. The human workflows. The political dynamics of who needs to approve what.
This is why domain expertise becomes more valuable, not less. The person who has spent fifteen years in healthcare compliance and can now build their own tools is far more dangerous than a general-purpose developer who can ship fast but does not understand the domain.
What this means for builders
If you are a developer who defines yourself by your ability to write code, you have maybe two years to redefine yourself. The skill of translating requirements into working software is being commoditized. It is not worthless - plumbers are not worthless because everyone has pipes - but it is no longer sufficient to build a career on.
The developers who thrive will be the ones who move up the stack. From "I can build it" to "I know what to build and why." From execution to judgment. From code to strategy.
If you are a founder, this is the best time in history to start a company. The cost of building a first version of almost anything has collapsed. But do not mistake low cost of building for low cost of winning. The bar for execution has dropped. The bar for everything else - taste, distribution, domain knowledge, strategic clarity - has risen.
If you are an investor, the old signal of "this team can ship" is becoming meaningless. Everyone can ship. The question is whether this team understands something about their market that nobody else does. The edge moves from execution to insight.
The meta-point
We are entering a world where the ability to create software is no longer a distinguishing capability. It is table stakes. Like literacy. Like spreadsheet skills. Like being able to send an email.
The interesting question is what humans do when one of their primary bottlenecks disappears. History suggests: they find the next bottleneck, and the people who identify it first have a massive advantage.
I think the next bottleneck is judgment. Not intelligence - AI handles that increasingly well. Not knowledge - that is searchable. Not speed - that is automated.
Judgment. The ability to look at a situation, understand what matters, and make a decision that accounts for context that cannot be fully articulated. The ability to say "this is worth building" or "this is not" and be right more often than you are wrong.
That is a human skill. It comes from experience, from taste, from understanding people, from having been wrong enough times to develop intuition about what right looks like.
Software is free now. Judgment never will be.
What I am doing about it
I run my companies with this mental model. I spend very little time thinking about whether something can be built. Almost everything can be built. I spend almost all of my time thinking about whether it should be built, for whom, and how to get it in front of them.
My AI setup is designed around this. I do not use AI primarily to write code faster - though it does that. I use it to think more clearly about what to build. To explore problem spaces. To stress-test assumptions. To understand markets. The code is a side effect of good thinking. It was always supposed to be.
The founders who figure this out first will look, from the outside, like they have some unfair technical advantage. They do not. They have an unfair thinking advantage. They spend their time on the right problems because they understand that the right problems are the only bottleneck left.
Shipping is not the hard part anymore. Knowing what to ship is. And that has always been the real game - we just could not see it clearly until the noise of execution got out of the way.