Spec is the New Code

There is a before and after moment in most technical careers — a specific project where you realise that the thing you thought was "overhead" was actually the work.

For me, that moment came not from a success, but from watching an algorithmic trading bot drift silently away from everything I intended it to do. Not a bug. Not a crash. Just a long, quiet separation between my idea and what the system actually became. The absence of a spec.

We are entering a world where AI can execute almost anything — given the right instructions. The execution is becoming abundant. Cheap. Fast. Reliable enough.

What is not abundant is the clarity of what should be built. The decisions made before the first line of code. The constraints, the intent, the reasoning. The spec.

The shift nobody is talking about

For decades, the scarce resource in software was execution. You had ideas but not enough engineers. You had product sense but not enough time to build. So we optimised for building fast — Agile, sprints, MVPs, "move fast."

AI flips this. For the first time in history, you can describe something reasonably well and have a working version in hours. The execution bottleneck is cracking open.

Which means the new bottleneck is the description. The spec. What exactly are you asking for? How precisely can you articulate the intent? What happens when the system encounters an edge case — what should it do, and why?

Spec is the new code. Not metaphorically. Structurally.

What this means for Project Managers

The PM role — the real one, not the meeting-scheduling version — has always been about translating intent into executable direction. What changes now is that the audience for that translation is increasingly an AI system, not (only) a human team.

And AI systems are not forgiving. They will execute what you say, not what you mean. They cannot ask clarifying questions (not reliably). They will not flag when the spec contradicts itself three sections in. They will not tell you when they are drifting.

That is where the Human still matters. Not for execution. For judgment. For holding the intent when the system loses it.

The PM who learns to write precise, honest, complete specifications is not being replaced. They are becoming the load-bearing structure of the entire AI-assisted delivery process.