Execution is becoming cheap. Judgment is not.

If building is no longer the hard part, deciding what deserves to be built still is.

The trigger

I still hear these quite a lot from people I interact with:

“A.I. is a bubble, I’m not impressed.”

“A.I. is too expensive for what it does; humans are much more efficient.”

“I’m only using A.I. because my employer wants me to.”

“Damn clankers stole my job.” 🤖

“🦆 A.I. !!!”

These are all genuine, sincere reactions, so how can it be that, for some, A.I. is portrayed so badly, while for others it’s an enabler?

Building products: the new order

Some say that building a successful business is about creating added value. Others say it’s about removing as much risk as possible: market risk, financial risk, product risk, execution risk. The more uncertainty you can remove, the clearer the path towards making something viable becomes.

It was a late evening in the winter of 2025 when I asked Gemini what it thought about a business plan I had put together. Gemini said that the biggest risk was execution: writing all the software and wiring the hardware. It seemed reasonable. After all, there’s an unfathomable amount of code, logistics and organizing required to make baseworks.tech reality.

Fast-forward about a year later and it turned out that what the A.I. did was to dramatically reduce that exact execution risk -  to the point where it was no longer the determining constraint. Building a product was fundamentally different from how it would have been before A.I. And I attribute that largely to A.I. covering so much of the “execution” part.

The practical consequences

Q: Can we actually build this? How long will it take? How many people do we need? How much experimentation can we afford?

A: Those questions increasingly become less about raw execution capacity and more about how much work we can plan, delegate and organize while keeping the whole thing coherent. Read “coherent” as “not vibe-coded”.

Q: Are you saying I need to throw a lot of money at A.I. to make this work?

A: Not necessarily. Budget still matters, but I’ve found that the real constraint is often not how much A.I. you can afford to use, but how much of its output you can meaningfully direct, review and integrate. You can get a surprising amount done on a relatively modest budget if the problem is well defined and the work stays coherent. Spending more can increase throughput, but it doesn’t remove the need to make sense of what is being produced. Implementation is wildly faster; our time and attention aren’t expanding with it.

Q: OK, so what you’re saying is that A.I. can execute. I just need to somehow make it do what I want, and do it well. What do I need for that?

A: What becomes valuable is the ability to define the right problem, find the right audience - those who share the worldview that the problem needs solving - and then manage the immense complexity while keeping everything coherent and converging towards the solution. 

Analytical thinking. Creativity. Being able to articulate why something matters. Understanding users. Turning ambiguity into wire-frames, diagrams, flows and interactions that people as well as A.I. can leverage. And on the technical side - everything that you know about your domain - in the form of patterns and designs, not in the form of finished work.

A.I. may be good at exploration - breadth - but I’ve largely found that its biggest advantage is depth: give it a poorly defined problem and you’re almost guaranteed to get spectacular results 🤡. Give it a precisely articulated problem and it can go remarkably deep, turning clarity into genuinely useful execution.

Perhaps that is the real shift: not that we can do everything, but that we now have to decide more carefully what is worth doing.

Thank you for reading!