The AI handoff note for long-running work
When AI becomes part of a real project, the hard part is not one impressive answer. It is keeping the work coherent after the third session, the fourth decision, and the first messy exception.
The useful signal in recent AI updates is that context over time matters. OpenAI described safety work that helps ChatGPT recognize risk from earlier context in rare high-risk situations. Anthropic and PwC are training tens of thousands of professionals on Claude as part of enterprise workflow redesign. OpenAI's workplace adoption research also frames AI value around practical guidance, writing, research, and decision support rather than one-off novelty.
For everyday knowledge work, the practical takeaway is simple: build a handoff note. Do not rely on the model to remember the shape of a project. Give each future AI session a compact briefing that says what happened, what matters, what is off-limits, and what comes next.
The skill
An AI handoff note is a living project summary you update at the end of each useful session. It is short enough to paste into a new chat, but structured enough to prevent the model from rediscovering old decisions or crossing boundaries you already set.
AI handoff note
Project:
{one-line description}
Goal:
{what successful completion means}
Current state:
{what has been completed, drafted, decided, or rejected}
Key decisions:
- {decision and reason}
- {decision and reason}
Constraints:
- {brand, legal, privacy, budget, deadline, tooling, audience}
Source links or files:
- {source and why it matters}
Open questions:
- {question blocking progress}
Risks to watch:
- {what could go wrong if the AI assumes too much}
Next action:
{the single next task the AI should help with}A worked example: planning a team training page
Imagine you are building a short internal AI training page. You have already discussed the audience, tone, and examples with an AI assistant. Tomorrow you want another session to draft the first section.
A weak restart prompt is: "Continue helping me with the AI training page." A better handoff looks like this:
Project:
Internal AI training page for operations managers.
Goal:
Create a practical page that helps managers use AI for weekly reporting without exposing customer data.
Current state:
- We chose a practical, non-technical tone.
- The page will teach three habits: source-first prompting, approval gates, and review checklists.
- We rejected a tool-comparison angle because the audience needs workflow habits.
Key decisions:
- Use examples from reporting and meeting prep, not engineering.
- Avoid promising full automation.
- Keep every prompt copy-pasteable.
Constraints:
- No customer names, private metrics, or personal details.
- Keep reading time under 6 minutes.
- Use plain English.
Source links or files:
- Draft outline: training-outline-v2
- Current policy note: data-handling-guidelines
Open questions:
- Should the first example use weekly reporting or customer-risk review?
Risks to watch:
- Do not imply AI may send updates without manager approval.
Next action:
Draft the opening section and one copy-paste prompt for weekly reporting.The prompt
Use this at the end of any AI session that produced decisions, drafts, or useful context:
Create an AI handoff note for the next session.
Include:
1. Project
2. Goal
3. Current state
4. Key decisions and reasons
5. Constraints
6. Source links or files mentioned
7. Open questions
8. Risks to watch
9. Recommended next action
Keep it concise enough to paste into a new chat.
Do not add new facts. If something is uncertain, label it as uncertain.How to use it
- Save it where the work lives. Put the note in the project doc, task card, or shared folder.
- Paste it before asking for new output. Treat it like the first page of a project brief.
- Update it after meaningful progress. New decision, new source, new risk, new note.
- Separate facts from preferences. "The legal team requires this" and "we prefer this tone" are different kinds of context.
- Prune aggressively. The handoff note should reduce noise, not become a second project archive.
Why it works
Most AI productivity losses happen at the edges: restarting context, re-explaining constraints, accepting a draft that ignores an earlier decision, or letting the model invent continuity that was never established. A handoff note makes continuity explicit.
It also changes your role. Instead of treating AI as a magic memory, you become the editor of the project state. That is a durable skill: useful across models, tools, agents, and whatever the next interface looks like.