172 Messages, 180 Messages, and a Stranger at 8:38

At 8:38 one morning, a session showed up on the board we use to see what needs us, and we had no memory of opening it.

A session is one working conversation with Claude, the AI we run our work through. The board is basically our to-do list, so an unfamiliar session usually means an unfinished decision or a failure we have not seen yet. We went looking for the work behind it. There was none.

The record looked exactly like our work

When we finish a session, a small helper reads the end of it and writes a summary so the result can be found later. That helper is itself a Claude session, so it gets its own record in the same place as our real ones.

This record ran from 8:38 to 8:38. It had two messages, a very short reply, and a first line that read “You are reviewing the tail of a Claude Code session to produce a structured close summary.” Its card carried the project, the branch, the first line and a timestamp, same as any real session. Nothing on it said an agent had made it.

The real sessions that day looked nothing like it. One had 172 messages and another had 180. By the end of the day, 5 of the 18 sessions in our log were helper summaries of one or two messages each.

The mistake was ours

The helper did exactly what we asked and wrote the summary. The problem was two things we had built. We started the helper without giving it a label, and we built the board to treat every record it found as work we had started. So each time we closed a session, our own automation added something to our plate.

Guessing from the shape of the record would not have fixed it. A board that decides “short and starts with a summary instruction, so probably a helper” gets fooled the day a real two-message emergency fix arrives, or the day the helper’s wording changes.

What we changed

Every helper now says what it is when it is created: what kind of work it is, and which of our sessions it belongs to. The board has a section for work we started and a separate section called “Agent helpers, not on your plate” for the rest. Each helper shows up beside the session it summarised. Nothing is deleted or hidden.

It shipped the same day, and we checked it live, from the label being written through to the new section showing on the board.

Claude wrote the summaries correctly. It was the board that got it wrong, by treating an agent’s housekeeping as our unfinished business. What counts as “ours” is a definition a person sets, and a board can count records all day without knowing whether a short session is routine or an emergency unless the record says who started it.

If you have automation that creates records on your behalf, look at whether each record says who started it at the moment it is created. Without that, your board of work fills up with the system’s own housekeeping, as ours did with 5 helper sessions in a day of 18.

dxdev LESSON · paste into your AI agent

LESSON: A board that cannot tell who started the work will count your helpers as your backlog

WHAT TO SHARE:
A de-identified description of a board, inbox or task list that shows open work, and any automated helper or assistant that creates records in the same place as human work.

ASK YOUR AGENT TO:
1. List every automated process that creates a record on the board, and say whether the record shows who or what started it.
2. Check whether the board separates human-started work from assistant-started work by an explicit field, or only by guessing from its shape, such as length, wording or timing.
3. Say what a short but genuine piece of work, such as a two-step emergency fix, would look like on the board and whether it could be mistaken for a helper.
4. Flag any proposed fix that hides items by a filter rather than recording where they came from.

RETURN:
A list of the automated sources on the board, which ones are labelled at creation, and which ones are only guessed at afterward.

BOUNDARY: Do not delete, hide or reclassify anything on the board. A person reviews the list and decides what counts as theirs.