Day 986 - The AI Productivity Revolution Is Real. The Question Is Who It’s For.
Last year, I worked on a contracting engagement migrating a large retail loyalty platform from Angular to React.
Standard stuff. Senior developer, market rate, fixed term.
Except it wasn’t standard.
I was running AI coding agents throughout — Claude, Gemini — and they were doing a substantial share of the mechanical work. Code generation, boilerplate, test scaffolding, repetitive transformations.
My honest estimate: I was producing roughly twice the output I would have without them.
The project is finished. The client was happy. I was paid the agreed rate.
Nobody renegotiated. Nobody got a bonus.
The productivity gain simply disappeared into the engagement.
I’ve been thinking about that ever since.
Where did the value go?
The client got a shorter project at market rates.
They captured the gain without asking for it. The contract was for a deliverable, not for hours of human effort. When the deliverable arrived faster, they paid less in total.
That’s it. That’s the whole story.
“Labour-saving devices don’t save labour. They transfer it — usually upward.”
— broadly attributed to economist Joan Robinson
This isn’t a complaint about one client. It’s a pattern.
Every knowledge worker who adopts AI tooling is, right now, doing more and finding that the additional value flows directly to the buyer of their output. The market rate hasn’t caught up. Their leverage hasn’t increased.
The tools improved their output and, at the same time, made their premium rate harder to justify.
The rate pressure that’s coming
Here’s the part worth stating plainly.
I delivered double the output. A developer with five fewer years of experience and AI tooling could probably deliver 80% of what I delivered — for 20% less. A developer with ten fewer years and better prompt skills might deliver 60% for 40% less.
None of those transactions feels like a crisis.
Each one is just the market clearing.
But the cumulative effect is a rate floor that drops steadily — driven not by any explicit devaluation of skill, but by a quiet expansion of supply. More output per developer means the market needs fewer developers, or can pay them less, or both.
The maths is not complicated. The implications are.
This isn’t a prediction about the distant future. At current rates of AI capability improvement, it’s a description of the next two or three years — for developers first, then analysts, writers, paralegals, and financial modellers close behind.
The Last Safe Harbour
There’s a second thing I noticed on that project. It’s more uncomfortable.
AI coding tools are genuinely impressive at execution. Clean functions, edge cases handled, tests generated. Where they’re weaker — at least today — is in software architecture. The decisions that determine how a system is structured so that future changes are cheap and localised.
That’s actually the whole point of architecture. Not elegance for its own sake. Not intellectual satisfaction.
Good architecture exists for one reason: to make future changes fast and cheap.
A senior developer knows how to make one change in the right place. That skill takes years to develop.
But then I thought: does it matter?
If an AI can make 57 tricky changes to a poorly-structured codebase — flawlessly, instantly, at near-zero cost — then the value of elegant architecture starts to collapse.
The entire point of good structure is to make future change cheaper.
If future change is already cheap, you’ve removed the problem that good architecture was solving.
I’ve spent thirty years getting good at placing one change in the right place.
That skill may be obsolete within the next three years.
Not a decade. Not some abstract future.
Within the timeframe of this newsletter.
What this means for all of us
I want to be careful not to overclaim. One developer, one project, one data point.
But the pattern generalises.
The standard counter-argument is that productivity gains always create new jobs and new value categories over time. The loom, the assembly line, the internet — all displaced workers, all eventually created new types of work.
But here’s what’s different this time.
Nobody can name what those new jobs might be. Not even roughly. Not even as a category.
With every previous wave, you could point at the new thing. This time, the pointing hand is empty.
It is not a useful frame for the individual knowledge worker deciding what to do in the next eighteen months.
“It is not the strongest of the species that survives. It is the one most adaptable to change.” — commonly attributed to Charles Darwin
At the individual level, the question is simpler and harder:
The value you’re creating is increasing. The share of it you’re capturing is shrinking.
What are you going to do about that?
I don’t have a complete answer.
But I do see one viable path: escape the dynamic entirely. Build financial independence fast enough that your income is no longer hostage to a rate floor that only moves in one direction.
Don’t win the race to the bottom. Get off the track.
That’s what this newsletter is really about.
The gain was real.
The beneficiary wasn’t you.