Most of the work a product team does is not the roadmap. It is the steady run of bugs from support: a customer hit something, an agent wrote it up, and now it has to find its way to the person who can fix it and back to the person who reported it. Every team has a process for this. Almost none of them like it.
Here is the path one ticket took last week, with the three moments a person was involved marked. Everything else happened because the record moved.
9:12 · Support files it
Marcus is on the support desk. Ticket #8841: exports over a few thousand rows time out, two customers this week. He is already in Claude with the sleds connector, so he says what he has.
The connector reads the product map and names what this touches (exports, jobs). It looks for the error and finds the Sentry group: 41 events, 2 users, first seen Tuesday, the stack attached. It asks him the one thing a bug report from support needs that he can answer: how many customers, and are any of them on a renewal. Then it files the request with the ticket number, the stack, the counts and his words, sized S · $2–6 · 10–30 min.
Person one: Marcus, for about ninety seconds.
9:40 · Product says how much it is worth
The row is on the board, under the exports area, waiting on one thing: worth. Dana owns that part of the product. The menu bar app shows one item needing her; she opens it, sees two customers and one renewal, taps high. Approval is on auto for bugs under a medium band in this workspace, so that tap is the yes.
Person two: Dana, one tap.
9:41 · The seat that fits takes it
Routing sends it to a seat with headroom that owns exports. This morning that is Priya's Claude Code, which has the row, the stack, the ticket text and the acceptance line (an export of 10k rows completes under a minute) before she looks. She starts it, reviews what comes back, and opens the PR. It carries the lines sleds wrote: source, estimate, seat.
11:05 · The fix lands, and the receipt goes back
Priya merges. The GitHub App reads the merge, writes the actual (22 min · $3.80 · in band) to the row, and resolves the Sentry group with the PR link. The row tells Marcus in the Claude thread where he filed it, in the words the org wrote in AI Guidance for support: what changed, which release, what to tell the customer. Marcus replies to both tickets.
Person three: Priya, who was always going to be the one to fix it.
What did not happen
- No triage meeting. The worth question went to the one person who could answer it, and nobody else.
- No status thread. Support could see the row move, and was told when it mattered.
- No estimate argument. The band came from the last forty bugs in that area, and the actual landed in it.
- No re-explaining. The stack, the counts and the customer's words travelled with the request.
Three people touched it, each once, each for the thing only they could do. That is what "automated" means here. The judgement stayed with people. The carrying did not.