Six Months of Context, Fourteen Minutes of Patience: Grading the Agent-to-Human Briefing
Photo: tired founder laptop early morning coffee dark office, via img.freepik.com
There's a moment in every agent project that nobody demos at conferences: the handoff. Not agent-to-agent, not API-to-API. Agent-to-human. The machine has been grinding for six months. You — the founder, the owner, the person whose credit card is on file — have been elsewhere. And now you need to know everything, in one meeting, before your first meeting.
We ran this experiment on ourselves. One long-running agent, one sprawling project (vendor consolidation across three SaaS tools, if you're curious), one sleep-deprived human at 7:04am Pacific. We tested five briefing formats and graded each on how much actually survived contact with a brain running on fumes.
The Contestants
Format 1: The Memory Dump. The agent's raw context, exported verbatim. 340 pages. It died instantly. Our human scrolled for ninety seconds, muttered something unrepeatable, and closed the tab. Grade: F. Raw memory is a format for machines, not for the mammals who pay for them.
Format 2: The Decision Log. A chronological list of every choice the agent made, with rationale. Better — but 190 decisions is not a narrative, it's a commit history. Our tester retained the first four entries and the last one. Grade: C-. Useful as a reference document, useless as a briefing.
Format 3: The 'Explain Like I'm New' Summary. We asked the agent to pretend the reader had just joined the project. This was the surprise contender. It front-loaded the why instead of the what, named the three decisions that actually mattered, and flagged two things it was unsure about. Grade: A-. The self-flagged uncertainty built more trust than any polish could.
Format 4: The Q&A Pre-Mortem. The agent generated the ten questions it predicted the human would ask, then answered them. Clever, but it predicted the questions a careful person would ask. Our 7am human asked none of them. Grade: B for thoroughness, D for timing.
Format 5: The Two-Minute Voice Memo. The agent synthesized everything into a spoken brief, under 120 seconds, with a written appendix. This won. Something about hearing the project summarized in a human cadence — pauses, emphasis, a dry joke about a vendor's pricing page — made it stick in a way no document did. Grade: A.
What We Actually Learned
The formats that failed shared a flaw: they were organized by when things happened. The formats that worked were organized by what you need to decide. A sleep-deprived founder doesn't want history; they want a verdict plus the three things that could go wrong.
The second lesson: the handoff is not a one-time event. The best briefing was the one the agent had been drafting incrementally all along — a running 'executive summary' updated weekly. By month six, the final handoff was mostly a diff.
The Uncomfortable Part
Here's the thing we keep circling back to in the lab: if your agent needs a 340-page memory dump for you to understand what it did, the opacity isn't a bug you fix at handoff. It's a design debt you've been accruing since week one. Build the briefing in from the start. Your 7am self will thank you.
The full format templates and scoring rubric are going up in the open notebook soon. Bring coffee.
