By trade, I'm an engineering manager. By night, as you've probably read, I'm an AI skeptic. I generally think talk of AI coming for our jobs to be overstated (we're 2 years into the "devs won't be writing code in 6 months" timeline), but that's slightly dependent on the type of job that you hold. As an engineer, I'm not worried. I've written about this before and I don't see anything coming in the next year that would have me truly changing how I approach engineering. But as a manager, I'm worried.
Good thing I prefer being an engineer.
I probably get 10 hours of coding time each week. That's not too bad for a manager; 25% of my time is spent not managing! Another 25% is spent being in meetings that shield my team from things they don't need to hear or think about. The other 50% is mostly repetitive. Triage the ticket that just came in. Explain to a stakeholder how prioritization of the roadmap works. Work on the roadmap for next quarter. And then, code reviews.
Most code reviews are easy to review. The author breaks up their work, isolates it from existing code, uses helper functions that exist, and writes tests. The pull request description is good enough and there's usually a ticket that explains why the work was started. Maybe 10% of code reviews are hard. They change basic invariants in a system. They add or remove relationships between objects, and are often hard to keep small in scope. These are the fun reviews, but they're also hard.
The boring reviews and the boring tasks are primed for disruption. Most of these tasks are basic pattern matching. Even my roadmap planning could probably be automated in some way to account for incoming feature requests or bugs around specific features.
Then there comes the real management section: reviews. I try to keep notes for every one of my reports. I try to note down places where I see them improving, and I spend a good week around review season just combing through code changes they've pushed through so I can confirm the output I thought they had.
You know what's good at all of these things? Robots. AI.
My manager self will be replaced by AI
And I'm excited. I love writing code. I love building systems. I love spending time teaching people how to do these things with thorough, well-informed code reviews and documentation.
You know what's not good at these things? Robots. AI.
AI doesn't use helper functions, it reimplements them. It doesn't know what a "best practice" is, it's never had to maintain a 20 year old codebase that started with a language version that doesn't run on modern architectures, and it certainly doesn't know how to integrate in the correct sequence with the dozens of third party systems. Even with extensive documentation in that section of the codebase, I've yet to have Claude Code correctly discover the reason that we have to load data from a third party in a specific way, or why there are certain obnoxiously small batch size, or why we have to hold a bunch of info in memory or a separate table to do post processing. Even on a more basic level, I haven't had it recommend a consistent way of logging so that it's trivially searchable in our observability suite. These are second nature to humans, and right now is a level of pattern matching inaccessible to AI tooling.
The things that make 10, 15, or even 20 year old codebases tick are the humans. It's the incredible context and longterm pattern matching skills that we have. I'm coming to terms (in good way!) that I think I'll be more of an engineer and less of a manager in the future. But to get there, I need to embrace AI tools. The thing that I struggle to adopt is the thing that would allow me to do more creative work.
The robots are coming for my job, and I want them to take it
At least the parts of it that I don't love. The parts that are repetitive. The question isn't whether AI will take my job, it's whether what's left will still be a job worth having. And right now, I'm confident it will be.