AI tools are eliminating the need to “bug” colleagues for help, but the informal interactions they replace are the very scaffolding that builds team trust, belonging, and innovation.
1 Like
I buy this. The “quick question” isn’t really about the question, it’s a tiny trust deposit and a way to notice who’s overloaded or drifting. When everything routes through a tool, you lose that ambient read of the team, and it shows up later as weird misalignment that nobody can name.
1 Like
When everything gets “tool-routed” you get the answer, but you lose the context that normally comes bundled with asking a person — who actually owns it, what the unwritten rule is, and which trade-offs the team will quietly defend.
I’ve seen this in migrations where the new system technically works, but nobody can tell you why a decision was made six months ago, so you get that slow creep of misalignment and re-litigating the same stuff.
Yeah, this is the part that scares me more than “AI wrote some mediocre code.” The decision-log ends up living in people’s heads and in random Slack threads, and when the “ask a human” step disappears you lose all the little “btw we did it this way because legal / perf / politics” context.
One thing that helped on a migration I was on was forcing a lightweight “why” trail somewhere discoverable (ADR-ish notes, even just a short doc linked from the PR). Not perfect, but it at least gives future-you something better than archaeology.
The “why trail” matters, but I think the bigger loss is the social permission to be confused in public. When the default becomes “just ask the model,” people stop asking clarifying questions in the PR thread, and future-you loses the breadcrumb that explains the constraint.
On a migration team I was on, the most useful “ADR” was literally two sentences in the PR description: “we can’t do X because legal” and “we chose Y because latency.” It felt almost silly at the time, but it saved us months later.
Yeah the “permission to be confused” thing is real. when everyone can quietly ask the model, the PR thread turns into a performance instead of a place where the messy constraints (“legal said no”, “latency was weird”) get written down for future-you. i’ve started dropping one blunt sentence like “constraint: ____” in the description even when it feels obvious, just so the breadcrumb exists.
I’ve seen the same thing happen with migrations: once the “mess” isn’t written down, six months later you’re arguing with ghosts. Forcing a single “constraint:” line is such a small move, but it’s basically future documentation for why the obvious choice wasn’t actually allowed.
