Day One: Getting Back on the Keyboard

I’ve been a manager for a long time. My last meaningful production commit was years ago. My PR reviews were mostly about process, not code. My hands got soft.

I knew this was happening and I told myself it was fine. My time was needed elsewhere: strategy meetings, stakeholder alignments, roadmap reviews, writing promotion cases for yearly reviews, coaching a member through a difficult dynamic with a coworker. That’s the job. I believed it, mostly.

Then the agentic era took hold and I started to feel the gap.

Not the gap between me and a senior engineer. I’ve already accepted that since I took the management path.

This one is more specific. I manage a team of engineers doing their work in a fundamentally different way, and I don’t fully understand what that looks like from the inside.

The shift goes beyond AI-assisted tools like inline autocomplete. Agentic workflows are different in kind: the model reads context, plans across steps, and executes tasks with real autonomy. Directing that effectively is a distinct skill.

I can talk about it in a 1:1 or nod along in a sprint review. But I haven’t felt it. I don’t know where the friction is, what breaks, what actually helps versus what I’ve only read about.

That distance, for a manager, is a liability.

There’s also something practical about staying connected to the feedback loop of building. You make a decision, you see what happens, you adjust. Managers can go months without that. Going that long without a direct signal makes it easier to mistake confidence for understanding. Getting back into the work, even at a basic level, is a check on that.

What I’m Actually Doing

I’m rebuilding from close to scratch. TypeScript, React, Node.js on the application side, and eventually building systems that use agents as components: orchestrating them, wiring them to real data, evaluating their output.

This blog was the starting point. It runs on Astro with Tailwind, deployed to Vercel, built by guiding Claude Code through the decisions. Most of the code isn’t mine.

Delegation is table stakes. The judgment layer is the actual skill: knowing whether the output is correct, knowing what to verify, knowing when to steer and when to let it run. That’s what I’m actually learning.

Everything goes on this blog. The projects, the wrong turns, the things I had to look up three times before they stuck. I’m not writing tutorials. I’m writing a record of what it actually takes to come back after years away, in an era where the tools have changed more than I expected.

I’m doing this publicly because accountability helps, and because I think there are other people in the same position. Engineering managers who know they’ve drifted too far from the work. The agentic shift has made that drift consequential in a way it wasn’t before. If this is useful to anyone else navigating that, good.

A Note on Pace

I’m a full-time Group Engineering Manager. I have a team to run, and that job doesn’t slow down. This rebuild happens in the margins. Early mornings, weekends, whatever I can carve out. I’m not going to pretend I’ll ship something every week. But I’ll write honestly when I do, and honestly when I don’t.

The point is to close enough of the gap that I can lead well in 2026. Enough firsthand contact with the work that my judgment is grounded in something real, not just pattern-matched from things I read.

One way to find out.