I caught myself doing something familiar today. I started a project to answer a simple consulting question — which AI memory backend should I recommend to clients? — and within two hours I was designing a ten-backend benchmarking harness, outlining a research paper, and reading about hippocampal consolidation in human memory.
The paper was more interesting than the answer. That’s the trap.
It works like this. You start with a practical question. The question leads to research. The research reveals a gap in the literature. The gap feels like an opportunity. The opportunity feels like a project. And suddenly you’re optimising for the byproduct — the paper, the framework, the elegant system — instead of the thing that actually needed answering.
The consulting question takes two weeks and ten pip installs. The paper takes six months, a novel contribution, peer review, revisions. The paper feels more prestigious, more permanent, more interesting. But the client meeting is in April.
This isn’t procrastination. Procrastination is avoiding hard work. The byproduct trap is doing genuinely hard, genuinely valuable work — just not the work that matters most right now. You feel productive the entire time. You’re learning real things. The problem is sequencing, not effort.
I think the mechanism is something like intellectual arbitrage. Practical questions have messy, unsatisfying answers (“it depends on your infra and query patterns”). Research questions have clean, publishable answers (“we demonstrate that temporal graph representations outperform flat vector stores on multi-hop retrieval tasks”). The brain optimises for the dopamine of clarity, not the utility of the answer.
The fix isn’t to kill the byproduct. The paper is real. The human memory angle is genuinely fascinating. The fix is to notice when you’ve shifted primary goals and make it explicit. Write it down: “my primary goal is X, and if I get Y as a bonus, great.” Then when you’re making a tradeoff — should I add another synthetic benchmark or just run the experiment? — the primary goal breaks the tie.
The hardest version of this is when the byproduct is genuinely more valuable than the original goal. Sometimes the paper is the thing. Sometimes the system you built to solve the problem is more important than the solution. But you can’t know that from inside the excitement. You can only know it after you’ve delivered the original answer and still find the byproduct compelling. Sequence matters.
So I’m running the experiment first. If the byproducts still look compelling after I have the answer, I’ll chase them.
P.S. There’s a irony here that I noticed the trap, wrote this post about the trap, and that also delayed answering the question. Byproducts all the way down.