Software production costs are falling. Not slowly, not gradually – they're plummeting. What took a team two weeks in 2023 can often be done in two days today. What used to require three developers often needs one who can use AI well.
This is real. I see it in our own projects. In client projects. In the speed at which prototypes become products.
But what does it mean when the production cost of software approaches zero?
The Photography Analogy
When digital cameras became affordable, everyone could take photos. The production cost dropped from several dollars per shot (film, development) to practically zero. The result: not better photography. More photography. Billions of images, most of them mediocre.
The photographers who survived weren't those who took the most photos. They were those who knew which photos to take. Composition, timing, narrative – none of that got cheaper.
Software is heading the same way. When everyone can build software quickly and cheaply, the software itself loses differentiation value. What remains is the question: Are you building the right thing?
The Real Cost Was Never the Code
Most failed software projects didn't fail because the code was expensive. They failed because:
– The wrong problem was solved.
– The architecture didn't fit the reality of operations.
– Nobody talked to the actual users.
– Priorities were set politically, not technically.
– The team built what was requested, not what was needed.
None of these problems get cheaper through AI. They get faster. And faster in the wrong direction is worse, not better.
Speed Without Direction
I observe a pattern: Companies integrate AI into their development process, and initially, throughput increases. More features, more deployments, more PRs. The dashboards glow green.
Then, after a few months, a different picture: More features, but more conflicts between them. More deployments, but more hotfixes. More PRs, but less architectural coherence.
The problem isn't the speed. The problem is that speed without direction creates chaos at a higher frequency.
When everything is cheap to build, the most expensive thing becomes deciding what not to build.
The Price Paradox
Clients notice that software is getting cheaper to produce. They expect lower prices. That's logical.
But what they actually need – strategic advice, architectural assessment, prioritization, domain understanding – doesn't get cheaper. It gets more expensive. Because the people who can do this are rarer than ever.
The industry needs to learn to separate these two things: the cost of code, and the value of direction. A line of code generated by AI is worth nearly nothing. The decision about which line of code needs to exist is worth more than ever.
What Remains as Differentiation
When everyone can build fast, speed is no longer a competitive advantage. What remains:
Domain knowledge. Understanding the problem deeply enough to know which solution actually works in practice.
Architectural judgment. The ability to build systems that don't just work today, but can still be evolved in three years.
Saying no. The courage and experience to reject features, simplify requirements, reduce scope. The hardest skill in software development.
Operational reality. Knowing what works in production, under real load, with real users, in real regulatory environments. No AI generates this experience.
Software will soon cost almost nothing to produce. And that's precisely what will make the expensive part even more important: knowing what to build, why, and for whom.
The future doesn't belong to those who can write the most code. It belongs to those who can make the best decisions about code.