Coding was never the bottleneck

AI doesn’t make your product better.

For 20 years or more, the software industry optimized for peak developer velocity. Everyone needs to go faster: tools, frameworks, copilots, and now: agents.

Taking a step back in this industry, we must ask the question: has any of it made us genuinely faster?

Yes, of course. There were enormous jumps from Assembly to C, from C to Java, and from Java to Python. With every new layer, developers were able to focus more on what they’re building in terms of business logic and less on the underlying complexities that were abstracted away.

So software became larger, more complex, more integrated. The dev tools also got more sophisticated.

What didn’t? The processes. The decision making. The alignment.

Getting everyone on a project to speak the same language is, in my eyes, still one of the greatest challenges. I’ve rarely seen projects fail because of developers, but I’ve seen projects fail because of misalignment, miscommunication, and bad processes.

The common denominator that does not magically go away with AI agents is decision quality.

Decisions are hard because they’re inherently human — as long as we build systems for humans, which we mostly still are.

We cared so much about “how fast we build” that we’ve forgotten that the actual question is “what and why we build”.

This exact decision quality is the bottleneck, not developer velocity or how fast we ship the next feature.

What’s really important is the features we build and the reasons for them.

Every feature shipped is a liability. It needs to be supported, maintained, and integrated into the product. And when it breaks, or you want to remove it down the line, you might have to fight the few users who like this feature.

Good decisions — knowing what to build and what not to build — require knowledge and creativity.

Once a feature has been added and proved to work, it’s easy to copy. But to actually find that exact feature and state that’s worth building is disproportionately more complex. It’s like with art: once it’s done, it’s easy. But coming up with a unique idea nobody had before — that’s the real deal.

And that is why good decisions are so important. Additionally, they scale. They deliver clarity and momentum, whereas bad decisions compound in the form of technical debt and frustrated users.

The AI paradox

AI makes building cheaper and maybe even easier, but it barely helps with deciding better. Bad decisions get built faster, compounding the cost.

Still, AI allows for faster prototyping and ideation. You have more (visual) options to choose from, a quicker feedback loop. Your team can now explore many paths more quickly — yet it doesn’t take away the necessity of knowing which path to take.

AI won’t make products better by itself. But it can help explore faster — if we get the deciding part right.