Marcus Povey

I have been noticing a shift in the kind of work my clients have been asking of me. Not long ago, the value proposition was easy to describe. Coding. Someone needed a product, platform, system of feature set, and they needed someone who could design and build it. Engineering itself was a visible thing that they wanted to buy.

These days, not so much…

Once a platform is in place, many clients increasingly feel they can use AI to manage routine updates, generate code, patch content, and keep things moving without paying for ongoing engineering support. From their point of view, that can look completely rational, and I don’t begrudge them. If AI can produce something that seems good enough, quickly and cheaply, why keep paying for the person who used to do that work directly?

I don’t necessarily think this is a bad idea, and in many cases it makes perfect economic sense. It works, at least for a while. Regardless, it is getting much harder to sell “writing code” alone as a premium service, when the market can now see a machine produce plausible code on demand.

More interesting, perhaps, is the question of what remains valuable once the implementation becomes a commodity?

For me, writing code was never the hardest part. The hard part is deciding what should exist at all. It is understanding what my client is actually trying to achieve, and spotting the gap between what my client is asking for and what would genuinely solve the problem, accounting for tradeoffs between speed, complexity, risk and maintainability. Something which technically works, may still not be the right direction, and on numerous occasions I have directed clients as such, pointing out that what they were asking for may not best serve their goals. This layer has not disappeared, and if anything, it matters even more now that implementation (thanks to our new AI overlords) is getting increasingly easier.

Kids with dynamite.

AI is great at creating momentum, and can help with execution. It helps people feel less blocked by a blank page or an empty repository, but it does not reliably tell you which problem is worth solving and which small decision will become expensive later on. My experience is, as I described to my team, that the AI is an enthusiastic undergrad. Eager to please, and will do what you ask of it to the best of its ability… even if what you ask it to do is really really stupid.

I think that is where the real distinction is these days… the difference is not engineering, it is judgement.

Indeed, I have already had a number of former clients reengage after having come a little unstuck, whether by the AI enthusiastically obeying dangerous orders, or by a lack of understanding of what needs to be in place. One startup founder recently reached out to me after the agent enthusiastically performed a software change which unfortunately broke their payment system, this wasn’t picked up as the agent’s “truthy” looking code passed the tests it had written for it. Increasingly that is forming the basis of my offer, less: “I build software for you” and more “I help you decide what to build, what to change, and what to trust”

Building systems was, for me at least, a lot easier to sell as a value proposition, and as AI rapidly eats that world, it requires some painful growth. However, it’s a useful forcing function, pushing me to describe the value I provide more realistically.

I don’t just make things, I make sense of things. The code is getting cheaper. Judgment is not.

Don’t hesitate to reach out if you need help!