A weekend of agents, a Monday with nothing to catch up on
By Soumen Sarkar · October 2026
Scenario. Agents work through a weekend, handoffs write themselves when sessions stop, and Monday starts with nothing to catch up on. A made-up team, told by day.
How does a team start Monday with nothing to catch up on?
By letting the work explain itself while it happens. When every session that stops leaves a handoff, every change carries its reasons, and each day's work is told in plain words, Monday morning is a read, not an investigation. Here is one weekend, told by the day and the hour.
A made-up team, a real way of working.
The team builds a podcast hosting platform. Three people matter this weekend:
- Odile, in Montréal, owns publishing: the feeds that podcast apps read.
- Nadia, in Edinburgh, is on the weekend rota this week.
- Rahul, in Pune, owns uploads and storage.
Friday
17:50
Odile starts three long Claude Code sessions before she leaves, each with a clear first request, which becomes its goal on Activity:
- "Check every feed against the podcast directories' rules and fix what fails."
- "Import show transcripts from the old upload format."
- "Add chapter markers to episodes."
21:30
The feed session stops partway through, on a feed that fails in a way it can't fix alone. A handoff is written from the session:
Handoff: Feeds pass directory checks (Odile's session)
Done: 41 of 44 checks pass; fixed missing artwork sizes and bad dates
Left: 3 checks, all about episode file size
Next step: decide the size limit, then reject or warn on upload
Open question: do we reject feeds over 5 MB? (assumption stated)
Watch out: one show's feed is 6.2 MB and directories still accept it
The transcript of the session isn't kept. The handoff is what remains, and it's enough.
23:10
The transcripts session finishes and calls finish_work. The chapter markers session stops halfway, and writes its own handoff.
Saturday
10:05
Nadia opens Activity and filters to Status updates. Two pieces of work sit paused partway, each with its next step. She takes the feed one: she pastes Odile's handoff into her first prompt, and her agent starts from it.
10:20
Her agent reads the open question in the handoff. The feed size assumption is marked High stakes, because feeds are a public interface. The agent doesn't wait on it. It builds a warning first, which is safe either way, and keeps the reject behind one setting.
10:25
The first time Nadia's agent opens Feeds · lib/feeds, its area brief adds what Odile's handoff didn't need to repeat: a decision from last month, "Feeds never drop episodes a listener could still download", and an approach tried and dropped, compressing artwork inside the feed. Nadia hasn't worked on feeds before. She doesn't need to.
10:40
The assumption went to Rahul, who last worked on storage. Nadia sees it on Activity too, but it isn't hers to answer and she leaves it with him. Nothing is held up by it.
13:15
Nadia's agent commits. The reasons are written for the reviewer:
For: the last three directory checks, about episode file size
Calls made: warn on oversized feeds now; rejecting is one setting away
Assumed: feeds over 5 MB should be rejected (high stakes, not yet answered)
Tried and dropped: trimming old episodes from the feed (changes what listeners see)
She opens a pull request with the reasons link in the description, and her agent calls finish_work with the outcome. Her daily update that evening says it in plain words.
18:00
At Nadia's end of day, her daily update is written from her work log:
Picked up Odile's feed checks. All 44 now pass with a warning for
large feeds. Whether to reject large feeds is a high-stakes call
waiting for Rahul; the work doesn't depend on it.
Sunday
All day
Nobody works. The chapter markers handoff sits on Activity, unpicked.
Monday
08:30
Rahul signs in in Pune and opens Activity. A high-stakes assumption is in the feed, and he follows its "Answer it" link. He knows the answer: "Don't reject. Directories warn above 5 MB but accept up to 10. Warn at 5, reject at 10." He clicks "Correct it…" and writes that.
Nadia's session has stopped, so the correction stays with the assumption on Activity, where whoever touches the feed work next will see it.
09:00
Odile opens her first session in Montréal. The session brief starts with "Pick up where it was left":
Pick up where it was left
- Yours: Chapter markers, half done (handoff Friday 23:10)
09:04
She opens Activity. The chapter markers sit in the feed as work paused partway, with its next step. That's hers, and she continues it now.
She sets Since to "your last visit" and the type to Actions. The transcript import is there with its reasons. Nadia's feed work is there with its pull request, the reasons link, and Rahul's correction under the assumption.
09:10
At the top, the "What Campfyrr caught for your team this week" card counts the questions answered from the record, the assumptions stated, and the handoffs picked up and how fast. The feed handoff counts as picked up within a day. The chapter markers one will count as picked up today.
09:20
Odile asks her agent to change the reject limit from 5 MB to 10 MB, as Rahul said. It's one setting, because Nadia's agent kept it that way. She reviews the pull request by its reasons and merges it.
09:30
Odile has read the weekend: Friday's sessions, Saturday's work, Monday's correction. She didn't message anyone, and nobody had to write her a summary. The chapter markers session is already running.
What made it work
- [Let the work speak](/vibeflock/let-the-work-speak). Every session that stopped left a handoff, every commit carried its reasons, and the daily update told Saturday in plain words. Monday had nothing to reconstruct.
- [Keep moving](/vibeflock/keep-moving). Nadia picked up Odile's work on Saturday by pasting the handoff into her first prompt, and her agent started from it.
- [Turn unknowns into calls](/vibeflock/turn-unknowns-into-calls). The high-stakes question waited for Rahul. The work didn't.
- [Fix forward](/vibeflock/fix-forward). Building the safe half first made the correction a one-line change.
For a handoff format your agents can use anywhere, see the handoff template and handoffs that write themselves.
The principle behind this: Let the work speak: Our work explains itself, so nobody ever has to come and ask us.
Move as one at agent speed. Nobody waits. Add Campfyrr to Claude: it briefs every agent before it touches the code, keeps assumptions and handoffs where the team can see them, and never takes your code (what it keeps).