As Execution Gets Cheaper, Clarity Gets More Valuable

One of the stranger things about building software with AI is how quickly the cost of changing your mind collapses.

A direction that would once have involved days of implementation can sometimes be tried, rejected and rebuilt before you’ve become particularly attached to it.

That’s liberating.

It also changes what feels expensive.

Rework is increasingly cheap. Ambiguity isn’t.

If I give an AI coding agent an unclear instruction, it doesn’t necessarily stop and expose the uncertainty. Quite often it produces a perfectly plausible interpretation of what I said.

Sometimes an impressively executed wrong answer is more dangerous than a slow one.

That has made me much more interested in clarity.

Not the old version of clarity where everything has to be specified upfront before anybody can begin. AI makes that less necessary because trying something and learning from it is becoming cheaper.

I mean clarity about the things that actually matter.

What problem are we solving? What outcome are we trying to create? Which constraints are real? What is the agent allowed to change? What does good look like? What would make us reject the result?

Those questions become more valuable when execution becomes easier.

The same thing scales beyond coding.

If an organisation can create prototypes quickly, it needs to be clearer about what it is trying to learn from them.

If teams can produce more software, they need stronger judgement about which software deserves to become a service.

If agents can take more actions, their authority and boundaries need to be clearer.

If changing direction is cheap, the important thing is not predicting the right direction perfectly at the start. It’s being clear enough about the outcome to recognise when the evidence says you should change it.

This is why I don’t think the AI shift is simply making technical execution less valuable.

It’s changing the relative value of the work around execution.

Problem understanding. Product judgement. Context. Boundaries. Evaluation. The ability to say what good looks like.

As execution gets cheaper, clarity gets more valuable.

And perhaps one of the most useful things we can build around AI is a system that allows that clarity to accumulate rather than having to rediscover it with every new task.