You're the Principal Engineer of an Agent Squad. Not a Manager.

Last Thursday I was lucky enough to speak at Agentic Engineering Day at the Aotea Centre in Auckland. A few people asked for the slides afterwards, so here they are, along with a quick rundown of what I talked about (for those who weren’t there, or were there but hadn’t yet had enough coffee).

Grab the slides here (PDF)

I hand-drew every slide. On paper. With pens and pencils, like some kind of animal. I did this to be a little different, because I’d never seen it done before. Of course on the day Mic Neale , in his talk just before mine, had a bunch of hand-drawn slides. What are the chances!?

Also, I reviewed the physical pages, then the photos of them, and still missed a spelling mistake on slide 7 until my AI agent pointed it out to me (hence the strip of “tape”). I’ll let you decide how ironic that is for a talk about humans working with agents. 🤖

So what was it about?

The talk grew out of an earlier post of mine , where I pushed back on the increasingly common line that, thanks to AI agents, every engineer is now a manager. Congrats! Role changes for everyone!

I get where it comes from. Working with agents means describing outcomes, splitting up work, judging whether the result is any good, and giving feedback (sometimes by way of swearing at your screen). Managers do all of that. But so do engineers, all the time, and telling someone who chose engineering because they love building things that their career is now “management” isn’t great news for a lot of people. I chose management, and I enjoy it, but a big chunk of that job is about responsibilities towards people. An agent giving me a dodgy answer doesn’t carry any of those obligations.

The comparison I prefer is the Staff or Principal engineer. Someone who takes an unclear problem, figures out what matters, makes the call that saves everyone else a week of work, and perhaps even still writes some of the important bits of code themselves. We already know how to recognise and grow that kind of technical leadership.

From there I got into a few things I reckon leaders should be paying attention to:

A new title won’t fix a crappier job. If someone spends their day reviewing generated code they barely understand, calling them a manager (or a Principal, for that matter) doesn’t help. Changing how the work is organised might.
Who still understands the system? I borrowed from Dex Horthy’s talk about HumanLayer’s “dark software factory” experiment, and why the engineering judgement needs to happen before implementation.
Has delegated authority kept pace? If engineers can build much faster but still need permission for every meaningful decision, we haven’t sped much up at all.

And I finished with four changes I’d make to how we lead this work:

Develop engineering judgement through real practice (and real mistakes).
Assess the result. Impact over output, so no calibrating people on just agent counts or PR volume.
Understand what motivates your people, and ask about their actual day.
Build on the engineering ladder you already have instead of inventing a shiny new one.

The short version is that I want us to make good use of these tools, and I also want the people using them to see a future in engineering they actually want to be part of. Whether my metaphor is the right way to get there… I could be wrong, but it’s the best one I’ve got so far.

How’d it go?

Okay, I think? They’re never as smooth as you’d like. I did quite a lot of prep (next time I’ll do more though), but on the day I had the usual nerves, my laptop kept going to sleep mid-presentation (thanks, power settings), and the spotlights were so bright I could barely look up at the crowd. So I walked off feeling like it could’ve been better. The feedback afterwards was good though, so it seems to have landed alright, and I’ll take that. 😅

The rest of the day

It was a well run event. Chris Mazur ran the day with help from his wife and daughter and other volunteers, and it showed in all the little things going smoothly. There was a nice quiet speakers’ green room, which my pre-talk nerves very much appreciated. The speakers also got together for drinks and nibbles the evening before, which was a great chance to chat with more of the group than I’d ever have managed on the day itself.

Annie Vella opened with her keynote, “The Accidental Rise of the Supervisory Engineer ” (slides at that link, well worth a flick through). Then Randy Shoup , who came all the way over from San Francisco for this, followed with “AI Won’t Eat Your Stack: Why Robust Components Matter More Than Ever”. Not a bad way to kick things off (but tough acts to follow!).

Something I hadn’t thought much about before is how different speaking at a conference is to a one-off talk at a meet-up or user group. Nobody plans it, but the talks kind of evolve over the day. Later speakers build on earlier topics, and people reference each other’s talks (forwards and backwards). I was on fairly early, but there was enough overlap with what Annie and Randy had covered that I found myself subtly tweaking what I said as I went, to pick up on points they’d already made. For example, Annie’s call to boldly reimagine the SDLC from first principles overlapped a lot with my point about delegated authority (her post on it is worth a read). A bit nerve-wracking with a script in your head, but I quite liked it. Just an interesting observation.

There was a good range of topics across the rest of the agenda . Josh Wulf ’s live demo was a highlight for me. Entertaining, memorable, and Josh was great with the audience (and braver than me, doing it live). It was also interesting hearing how other teams are building agents into user-facing products, not just using them to write code. And we wrapped up with governance and how to handle intent, which felt like a fitting place to land.

My favourite part though was the hallway chats. Thanks to everyone who came up to say hi during the breaks.

Thanks!

A big thanks to Chris and the team for having me, and for putting on such a good day. It’s nice to be in a room full of NZ engineers and leaders having a proper, grounded chat about this stuff, without the usual breathless hype.

If you were there and have thoughts, or just want to get in touch for a chat, hit me up ! 🙂

Cheers, Dave

参与讨论