Modern advances in generative tooling seem to think they’re pretty cool. As someone who is enjoying the more strategic nature of agentic coding (with a sensible amount of cynicism and reticence of course), I’ll attest that certain parts behind the hype are indeed neat to see operating in the real world. However.
It’s a movement that considers itself lofty, with chat of “10x” and “your job is solved”, but outside of sensationalist social media marketers dressed as programmers, there’s an inherent speed limit on continuous development that’s required to keep things on track. We’ve now learned that human involvement is still needed when agentically engineering, to the surprise of nobody who was competently writing code before LLMs started trying, and humans are relatively slow. It turns out that slow is actually a good thing for thinking and consideration and boring stuff like that.
A 15x developer stealing your girlfriend, yesterday.
As is tradition in the development world, optimising without losing fidelity means shifting left, so now we’re front- the planning phase and having the agent extract our knowledge as a starting point. The idea isn’t that you spend the same time planning as you did coding, but the idea is that you do spend some time planning. Please.
Just as there’s always been folk who think they can go fast by avoiding solid test coverage, so shall we also get happy ignoramuses of proper planning when attempting to get their work done for them by the Magic Words Box®️. I haven’t checked if Magic Words Box®️ is a real thing, so please don’t sue me Magic Words Box®️.
Going too fast is nothing new - when I was a lad, we called what you got when you did that a bug. We all made bugs sometimes by trying our luck with the laws of coding physics, and it was on those bugs we cut our debugging teeth by nursing the system back to health. Not always great for shareholders, but there’s no learning experience that sticks with you quite like a live incident.
Senior developers ready to tackle a P1
Sadly, I’m skeptical about whether this scales into the new world. I can’t say with certainty that 80x AI-speed developers of the future will cause a similar epidemic of bugs that will benefit the 20x developers. I suspect instead, that the LLM oligarchy will just sell a whole bunch more tokens.
So I say, slow it down and be the sensible one in the room. On the roads we’re told that the speed limit isn’t a suggestion; it’s there for everyone’s safety, and that applies to engineering too. You can, and always could, go very fast and be a danger to yourself and others. Nothing’s changed - it’s still on you to apply the brakes.