The handoff note: let the outgoing session write it for you

Overview

When you move a piece of work from one AI tool to another — a thread you started in ChatGPT, carried on in your company’s Copilot, finished in a separate writing tool — nothing crosses the gap but what you bring. The usual move is to bring it the hard way: re-explain it from memory, or paste the whole old transcript across. There is a better move. The session you are about to leave is holding the full context right now, so make it write the handoff note for you.

Why now. Writing on 26 June 2026 about how work moves between AI tools, Nate B Jones reduced the problem to one line: “Right now, the integration layer is you.” Every time a job crosses from one tool to the next, he notes, “you carry the state: what was decided, what source mattered, what changed, what the next tool is allowed to touch.” You are the wiring — but carrying the state and authoring it from scratch are two different jobs, and only one of them has to be yours.

What you’ll be able to do. Get the session you’re about to leave to produce a short, accurate handoff note, check it, and hand it to the next session so it picks up where the last one left off.

The content

The obvious read is that this is solving itself — every tool has memory, projects and connectors now, so surely the context follows you around. It doesn’t. Memory is per-tool and mostly per-thread; the moment you cross a real boundary — a fresh chat because the old one slowed down, a switch from your research tool to your drafting tool, a change of model, or simply hitting the context limit and starting again — the state does not come with you. What carries over is whatever you bring.

So most people bring it badly. They re-explain the task to the new session from memory — vaguer the second time, quietly dropping whatever they forgot — or they paste the entire old transcript in so nothing is lost. Both are the wrong instinct. The session you are leaving already has the whole thing loaded: the decisions, the source that mattered, the dead ends you don’t want repeated. It can write a cleaner handoff in ten seconds than you can reconstruct in five minutes. Ask it to.

You carry the note. You don’t have to write it.

The one instruction that matters is short. Tell it to distil, not dump — and there is hard evidence for why. In an October 2025 study across five open- and closed-source models, accuracy fell by between 13.9% and 85% as the input grew longer, even when every relevant fact was present and perfectly retrieved, and even when the surrounding filler was masked so the model was forced to attend only to what mattered. Length by itself costs you. A pasted transcript doesn’t carry your context to the next session; it buries the part that counts under the part that doesn’t. A tight handoff the outgoing model wrote for you is the thing that travels well.

Your job doesn’t disappear — it moves to the right place. You stop being the typist and become the checkpoint. Read the handoff before you carry it: did it capture the decision you actually made, or a softer version of it? Did it carry forward a number or a clause that has since changed? A confident snapshot of a stale fact is worse than no note at all, because the next session will build on it without a second thought. And when the next step needs the real thing — the actual contract, the actual dataset — carry the document itself, not a description of it. The note moves the state around the work; it is not a substitute for the work.

Try it

Next time you’re about to cross a boundary on a real task — closing a slow chat, moving to another tool, starting fresh — don’t re-explain anything. Before you leave, paste this into the session you are leaving:

I'm about to continue this work in a fresh session that will have none of
this context. Write me a short handoff note — under 200 words — for that
next session. Include only:

- Goal: the outcome this work is driving toward
- Decided: what's settled and should not be reopened
- Hinges on: the one source, number, or fact that matters most
- Changed: anything that shifted since we started
- Off-limits: what the next step must not touch or assume
- Next: the immediate next step

Write it for an assistant that knows nothing about this. Leave out the
back-and-forth; keep only what the next session needs to continue.

Then read what it gives you before you carry it anywhere. If it softened a decision or dropped the constraint that matters, fix those lines — you have the context to catch that; the next session won’t. Paste the corrected note as the first message in the new session and keep going.

Where it breaks: the outgoing model can misjudge what mattered, or restate a fact that was true an hour ago with full confidence — so the read-through isn’t optional, it’s the whole of your job here. And don’t lean on a note where the next step needs the real document; paste the document. A snapshot is only as good as the moment, and the eye, that took it.

Additional reading

• Make Your AI Agents Hand Off Work Without You — Nate B Jones (Jun 2026) — source of “the integration layer is you” and the framing that you carry the state (decisions, sources, changes, what the next tool may touch) across every tool boundary.
• Context Length Alone Hurts LLM Performance Despite Perfect Retrieval — Du et al. (Oct 2025) — across five open- and closed-source models, accuracy fell 13.9%–85% as input length grew even when all relevant information was present and perfectly retrieved, and even when the filler was masked; the case for a short handoff over a pasted transcript.
• Why Claude Skills Don’t Travel to Codex (and How to Fix It) — Nate B Jones (Jun 2026) — the companion argument for keeping your way of working in a portable form you own rather than trapped inside one vendor’s platform.

Editor’s note

It’s an easy habit to develop, and if you aren’t already doing it, it will make your life so much easier. I used to be reluctant to end some of my working sessions because the loss of grounding in whatever I was doing was a pain. Now, it’s a comfortable and easy process. There’s a place for /clear and /compact, but more valuable is just knowing how to transition to a fresh session without losing momentum. Make it a reflex, and you won’t regret it.