Every big software engineering team right now is racing to out-do themselves on their adoption of agentic coding practices, and ship faster. There is something more insidious going on with many of the software engineers I talk to1 though. A lot of pressure to build “more! faster!” comes from themselves.
This shows up all over: the “you only have 2 years to escape the permanent underclass” meme2, or the various breathless LinkedIn or Twitter posts of 996’ing startups, labs, or particularly obsessive interns.
Things that used to require teams can now be done by a sufficiently keen solo engineer with a gang of Claudes, or Codexes, or a K2 agentic-swarms. That is thrilling, and it opens up the door to projects that you wouldn’t normally have bothered building. But it also open the door to thinking you need to build those things, and that’s not quite the same.
One of the observations of most people that take an extended leave from a large corporation is that much of the work they were doing wasn’t all that important. Either no one did it while they were out, or how they left it was… fine. Yet, much of that work somehow regains urgency as they come back to the role.
It’s very hard to tease apart how much of your output actually matters. Coordinating a large group of people inevitably takes overhead, and so many annoying aspects of work are genuinely important. But, much like Wanamaker’s famous quote about advertising, half of the work you do doesn’t matter, the trouble is you don’t know which half.
Adding a helpful and harmless model to the mix can certainly accelerate the rate of output, but it doesn’t do much about determining which bucket the work goes into. In fact, I’d say that the problems you take on when given a Max subscription are mildly more likely to to be things that haven’t been done because they are not worth doing. The combination of increased capacity and a pervasive sense of urgency is not a great recipe for quality decision making, or for a healthy relationship with your work.
It can be helpful to take the outsider perspective, at work or with personal projects. Would ask you someone else to do whatever you are considering, even with the expectation they would leverage agents to help them?
It’s often easier to see the value in something, or lack thereof, if you have to convince someone else of it. That can save you from some rabbit-holes filled with a sense of obligation to “extract value” from the time you already sunk into a misguided project.
This doesn’t mean you should ignore all of the ideas you have: you really can just do things, and you sometimes should! Just be clear about whether you want to spend your time3 that way, regardless of what the agent is doing.
Footnote:
-
Including myself! ↩
-
I appreciate Scott Alexander’s contribution on this topic: “You only have X years to escape permanent moon ownership” ↩
-
I didn’t actually quote him but everything about this article feels like a poor software engineer’s Oliver Burkeman, so you should just read him. ↩