In May 2026 I got flattened, M→IC, after being an engineering manager long enough to write a book about it. I assumed the hard part would be the craft — shipping diffs again, being personally on the hook. That’s not the hard part. The hard part is that my value prop changed, and I’d spent a decade building an identity on the old one.
I’ve changed companies before. Microsoft to Amazon to Meta — each one a real transition, each one hard in its own way. But those were changes of context. This is a change of what I’m for. There’s so much more that has to change, and so much more to say about it, that it needs its own post.
- Your value prop changed
- The two things you don’t get to be responsible for
- You have to trust someone else will solve it
- This takes time — work it through
- What’s different in the agentic era
- The upside, which is real
- Open questions
Your value prop changed
As a manager, my value was the team. What they shipped, how they grew, whether the system around them worked. On a good day I touched nothing directly and everything got better.
As an IC, my value is what I personally decide, build, and unblock. Nobody is going to look at my calendar and conclude I had a productive week.
The trap isn’t learning the new scorecard. The trap is running the old scorecard against the new job without noticing. I’d finish a day where I shipped something genuinely good and feel vaguely useless, because no one grew. And I’d finish a day of meetings feeling productive, because that used to be the job. It took a few weeks before I could name what was happening.
The two things you don’t get to be responsible for
There’s a symmetry here I didn’t see until I’d lived both sides.
As a manager, you can’t be responsible for all the technical decisions. You’re not in the code. By construction you know the least about the problem — I used to tell my teams this out loud, that my words should be discounted because I had positional authority and the least context. You have to let the team decide, and you have to let them be wrong sometimes, because a team that can’t be wrong can’t own anything.
As an IC, you can’t be responsible for the people. You see the person who’s drowning. You see the process that’s quietly grinding someone down, the reorg that’s going to land badly, the new hire nobody has picked up. And it is not yours.
That second one is the one I’m still working through. I believe people are the most important thing — that’s not a position I hold, it’s how I’m wired. Believing that and having no ownership of it is a strange place to stand.
You have to trust someone else will solve it
The only way through is trusting that someone else will handle it — either in the short term or the long term.
Short term, that’s their manager. Who may be less good at it than I would be. That’s fine, and noticing how much it stings tells you something unflattering and useful about your own ego.
Long term, sometimes nobody solves it. The team absorbs it. It works out anyway, or it doesn’t, and the org keeps moving. Watching that happen without intervening is a skill I did not have.
There are two failure modes and I’ve done both:
Shadow managing. You still hold the 1:1s, still coach, still fix the process — with none of the authority and none of the accountability. It feels like generosity. Mostly it’s an inability to put the hat down, and it makes life harder for the person whose job you’re doing a worse version of.
Pretending you don’t care. Also fake, and everyone can tell.
Where I’ve landed, for now: keep the caring, drop the ownership. I’ll coach anyone who asks, and I ask questions instead of pulling levers. I don’t run the loop anymore.
This takes time — work it through
This isn’t a decision you make, it’s a thing you metabolize. It took me longer than I expected, and I’m not done. If you’re doing this transition, give yourself a couple of review cycles before you conclude anything about whether you like it. Your first month is grief wearing a productivity costume.
Writing it down helps. That’s most of why this post exists.
What’s different in the agentic era
Here’s the twist: the IC job I came back to is not the IC job I left.
I gave up a team of humans and got a team of agents. Wally — my work claw — isn’t a single assistant, he’s an org: an orchestrator, persistent staff, ephemeral workers. What used to be an IC role now looks a lot like a manager-of-managers role.
Which means almost everything I learned as a manager ports:
| Manager skill | How it shows up as an agentic IC |
|---|---|
| Writing a crisp spec so someone else can execute | Writing a crisp prompt/plan so a fleet can execute |
| Deciding what to delegate vs. do yourself | Deciding what to hand an agent vs. hand-write |
| Code review as coaching, not gatekeeping | Review as the only quality gate you have left |
| “Never have a human do what a computer can do” | 100x more true than when I wrote it |
| Sustainable execution — ship fast, pay down debt | The whole game, now that output volume is unbounded |
| Agency — seeing what needs doing, doing it | Still the thing that separates people, just more visible |
And the part that doesn’t port is exactly the part that made management worth it.
In the manager book I described the trade this way: as a manager the day-to-day sucks and the moments are superb — you catch someone years later and they tell you the team was the most fun they ever had. As an IC the day-to-day is fun and the artifact is meh — you look up and the software you built is deprecated.
The agentic era changes one of those terms. The day-to-day is fun and I have leverage I never had as an IC before. What it doesn’t do is give me the moments. Agents don’t have career growth plans. They don’t get demoralized, and they don’t come back in three years to tell you it mattered. I got the manager’s leverage back without the manager’s meaning.
I don’t have that resolved. Naming it honestly is where I am.
The upside, which is real
- No calibration season. I hadn’t appreciated how much of the year I spent carrying that until it was gone. It’s a big part of why this summer’s time off is landing so differently.
- The day-to-day joy came back. Coding, designing, debugging a live outage — I forgot how good that feels in the moment. I underrated it for a decade.
- My side project became my job. CHOP, vibe coding, agent orchestration — the stuff I’d been doing nights and weekends is now what I get paid for, with a much bigger sandbox.
- The learning curve is the point. The role at Meta’s AI Lab is wildly undefined. Scary, ambiguous, chaotic — and I’m leaning in.
Open questions
- Do I go back to managing? Do I want to? Ask me in a year.
- If I keep coaching without authority, am I being useful or just failing to let go?
- How does multi-player agentic engineering work — a team of humans, each running their own fleet?