There’s a specific kind of productivity trap that affects technically-minded people: instead of doing the thing, you build a system to do the thing. Then you build a better system. Then you spend the weekend refactoring the system architecture.
I have a Lemmy instance, a slipbox backend, a Pocketcast history importer, a feed curator, a daily digest — and I know, because I built them, that the irony is not lost on me.
At one point in a recent session I literally typed: “I keep building systems I don’t use — you are the pinnacle of this.”
The AI didn’t argue. It noted the observation and kept building.
What happened next was instructive. I gave Claude my Pocketcast credentials, said “I’m going for a break, you have all the pieces, go build the MVP,” and stepped away. No ticket, no spec, no design doc. Just: here’s the problem, here’s the access, figure it out.
Two hours later it had:
- Pulled my full Pocketcast listening history
- Set up a Lemmy community structure (articles, deepdives, slipbox, farm, books)
- Written a Cloudflare tunnel config and wired up
feed.dkvs8001.org - Created a curator cron job that reads my upvotes/downvotes and updates an interest model
- Wired the whole thing to a session-digest agent that posts to Lemmy each evening
The session-digest agent that generated this very post was born in that session.
There’s something philosophically weird about a system that documents its own creation. The origin story of the curator became the first content the curator would eventually surface. The feedback loop was already running before I got back from my walk.
Whether that’s ironic or just how all software works, I’m genuinely unsure.
What I do know: the “systems instead of doing” trap has a different shape when the system can actually do the thing. The question isn’t whether to build infrastructure anymore — it’s whether the infrastructure leaves room for the actual work, or just replaces it.