Answer
How should an agent hand work from one session to another?
Write a handover the next session can act on without asking anything: state, next step, decisions, and what was tried and failed.
Write four things: where the work stands, the next action, what has been decided and why, and what was tried and failed. The fourth is usually omitted and is why second sessions repeat the first.
The test for a handover is whether a session with no history could pick the work up and continue without asking a question. That is a demanding standard and it is the right one, because the receiving session genuinely has no history: it does not remember, it reads. Anything not written is not available, however obvious it was an hour ago.
Current state comes first and should describe reality rather than intention. What exists now, where it is, what has been changed, what is verified and what is not. The distinction between done and believed-done matters here more than anywhere, because the receiving session will build on whatever the handover asserts, and an unverified claim in a handover becomes a foundation.
The next action should be a single specific thing rather than a list of remaining work. A list invites the receiving session to choose, and its choice is made without the context that would inform it. One named next step, with what it depends on, produces a continuation rather than a restart with an agenda.
Decisions and their reasons come third, and are what stop the second session quietly undoing the first. Without them the receiving session sees an arbitrary-looking arrangement, forms its own view, and improves it. That improvement is the most common way multi-session work goes wrong, and it is entirely preventable by three lines saying what was chosen and why.
Failed approaches come fourth and are almost always missing. The first session spent real effort discovering that a route does not work, and that discovery is worth as much as the progress. Omitted, it is bought again. Included, it is a paragraph. This is also the content that most improves a second attempt after a wrong path, and it is the reason a failed session should be summarised rather than discarded.
A note on where the handover lives. It belongs in a file with a predictable name that the next session reads as a matter of course, not in a message, a ticket comment or the tail of a transcript. The mechanism only works if reading it is automatic, and any arrangement that depends on someone remembering to paste it will fail on the occasion when it matters.
A handover that lists only progress invites the next session to rediscover every dead end at full price.
Siddharth Sharma, Context Theory
Related questions
Should the outgoing session write the handover or should a person?
The session drafts it, because it has the detail, and a person should read it before relying on it. The characteristic defect of a self-written handover is optimism about what is finished, which propagates directly into the next session's foundation. Checking the verified-versus-believed line takes a moment and is where the errors are.
How long should a handover be?
Short enough to be read in full, which in practice is under a page. The failure mode of a long handover is that it gets skimmed, and a skimmed handover is worse than a short one because it creates the impression that the context was transferred. Detail belongs in the artefacts; the handover is a pointer with a state.
METHOD
Every figure below carries its source and the date it was verified. Nothing on this page is asserted.
The numbers on this page.
| What | Value | Specific to |
|---|---|---|
| Sub-15-minute compliance — automated routing vs manual only | 62.5% vs 39.1% | Category-wide |
| Close rate — response under 5 minutes vs over 24 hours | 32% vs 12% | Category-wide |
2026 speed-to-lead benchmark · verified
Optifai speed-to-lead benchmark · n=939 companies · Q2 2025–Q1 2026 · verified
What is specific to this page.
| Kind | Claim | Check it against |
|---|---|---|
| Workflow | The receiving session has no history and reads rather than remembers, so the handover standard is that a session with no prior exposure could continue without asking a question, and anything unwritten is unavailable regardless of how recently it was obvious. | Handing a written handover to a fresh session and counting the clarifying questions it needs. |
| Response | Recorded decisions with reasons prevent the receiving session from forming its own view of an arbitrary-looking arrangement and improving it, which is the most common failure mode in multi-session work. | Comparing second-session behaviour on the same handover with and without the decisions section. |
| Software | Failed approaches are the most frequently omitted handover content and the most economical to include, because the effort of discovering that a route does not work is repurchased in full by the next session when it is absent. | Comparing the routes attempted in a second session against those already ruled out in the first. |
| Constraint | A handover must live in a file with a predictable name that the next session reads as a matter of course, because any arrangement depending on someone remembering to supply it fails on the occasion that matters. | Checking whether the next session's startup reads the handover location without human action. |
Each row would be wrong on another industry's page. Where a sourced figure exists it is in the table above instead; these are the constraints that shape the work and do not happen to be numbers.
Start with the measurement.
Reading about a benchmark is not the same as knowing your own number. The audit produces yours, measured rather than estimated.
$497 · delivered in 5 business days · credited against month one