The Quiet Displacement
The Thing You Forgot You Knew
There's a specific moment that happens to a lot of engineers, usually around month 8-12 of heavy AI use.
Someone asks them a question — a technical one, something in their wheelhouse — and they feel it before they think it.
The feeling: I know this. I know this. But I can't quite reach it.
Not "I don't know this." That's fine. That's normal. You learn new things, you forget others.
This is different. This is: I used to know this. I used to know this cold. I could have answered it in my sleep six months ago. Now I can feel the knowledge but it's behind fog.
That's the moment.
And the thing is — it's not dramatic. Nobody announces it. You don't suddenly become unable to do your job. The code still ships. The deploys still go green. The reviews still get done.
You just feel, sometimes, like you're reading from a script someone else wrote.
What Hasn't Changed
The parts of the job that are still fully yours — the judgment calls, the reading of requirements, the reading of people, the architectural tradeoffs that require knowing your specific system — those haven't gone anywhere.
AI is very good at generating code from clear requirements. It's less good at knowing what problem to solve in the first place, or whether this particular client's needs actually mean you should take a different approach than the one they asked for, or why this specific architecture decision matters more than that one in the context of this team's trajectory.
That second thing — the contextual, system-specific judgment — doesn't get automated.
It gets developed. It deepens. It compounds.
The most valuable engineers I know aren't the ones who know the most frameworks. They're the ones who know the most about their specific system — who can walk into a production incident and immediately know where to start looking because they've seen something like this before, in this specific arrangement of code and infrastructure.
The Skills That Compound
The interesting thing about software engineering — the thing that doesn't show up in job descriptions — is that the knowledge that matters most is highly local.
What you know about this codebase, this team, this problem domain, this business — that is extremely hard to automate away because it's accumulated over hundreds of hours of context.
AI can generate code. It can't generate that.
The question is: are you still building it?
Not in the sense of "are you still learning new things" — the knowledge is growing, the domain understanding deepens, the patterns accumulate.
But are you still building the skill of knowing? The muscle of working through a problem without something else doing the middle of it?
What would you lose if you couldn't use AI for a week?
Not hypothetically. Just: sit with that question for a minute. What would you actually not be able to do anymore?
The Follow-Up
Once you have that list — the things you'd genuinely struggle with — ask the follow-up:
Is that a problem?
Sometimes the answer is genuinely no. Some things genuinely don't need to be maintained at depth. AI is the right tool for them.
But sometimes the answer is: I'd lose something I actually want to keep. Something that, if I keep delegating it, I'll look up one day and it will be behind fog I put there myself.
The gap between those two answers — the one AI can't do vs. the one I'd miss — that's your dependency gradient. That's what the velocity signal doesn't show you.
This Week's Practice
Once this week, take a problem you would normally hand to AI and work on it yourself for 20 minutes.
Not because productivity. Not as a challenge. Just: to feel where the line is.
The 20 minutes is not the point. The 20 minutes is instrumentation.
What you find out about the line is the point.
What I'm Reading: The Manager's Handbook for Remote Leadership — If you're managing engineers right now, this is worth your time. Not a framework. More like: a collection of the conversations you need to be having with your team about what AI is actually doing to their relationship with their own work. Written for the manager who's watching their best engineer seem slightly less engaged than they were a year ago.