Answers
How do I see what my AI left for its next session — and can I trust it?
You can trust it as far as you can inspect it, so the note has to be something you can inspect. When a session summarises itself for the next one, that summary is usually text the next session reads and nobody else sees: you find out what it said only when the work goes wrong. In Arroway a session that stops with work unfinished leaves a handoff with fixed parts — where it stopped, the one next step, what is already checked and must not be redone, the risk left open, and who it is for. It sits in the project's panel before anyone continues, with the name of whoever left it, whether someone has taken it, and whether an AI has already read it. Only a person can discard it, and the discard stays on the record. Arroway does not check that the note is true; it makes sure a person can.
Last updated September 25, 2026
A note nobody sees is a note nobody checks
An AI writing a note to its future self has every reason to write a confident one. If a check failed, the note may say the work is done; if a step was skipped, the note may simply not mention it. None of that has to be deliberate — a summary is compressed, and compression drops what the writer considered unimportant. The problem is that the only reader is the next session, which has no way to tell. Auditing starts with moving the note to where a person can read it before it is acted on.
What a handoff has to say
Four things, each in its own place: where the work stands, the single next step, what was already verified and should not be redone, and the risk that stays open. Separate fields are what make it checkable. A paragraph can blur a finished step and an assumed one; a field called "do not redo" is a claim you can test, and an empty "open risk" is a statement someone made, not an omission. The handoff also says who it is for, and it may list what the next session has to read to continue, and the next session reads by that list.
Who can see it, and who can end it
A handoff is served at the top of every read of the project until it is taken and closed, so it cannot quietly fall out of the next session. In the panel, a person sees it under the work in flight, with who left it, how long it has been open, whether it has been picked up and whether any AI has read it yet. It ends in one of three ways: closed alongside a record of what landed, replaced by a new handoff that names it, or discarded by a person. An AI cannot discard one, and nothing expires it by age.
What it looks like in practice
An agent is migrating a customer table overnight. At 3 a.m. it runs out of time and writes a note for the morning session: migration complete, backfill running, verify counts. The morning session reads the note, sees "migration complete", and moves on to the next task. The backfill had in fact failed on the first batch, and the counts were never checked. The note was not a lie; it was a summary, written by the only party that knew what it had skipped. With Arroway, the overnight run leaves a handoff instead. Where it stopped: schema migrated, backfill started, one batch done. Next step: check the error on batch two. Do not redo: the schema change. Open risk: the counts are unverified. The engineer on call sees it in the panel at 8 a.m., before any agent picks it up, and the open risk is the first thing on the page.
Questions people ask about this
- Does Arroway check whether the handoff is accurate?
- No, and it does not claim to. It cannot see your systems, so it cannot know whether a step really ran. What it guarantees is that the note is structured, attributed and visible before anyone acts on it, and that ending it is a visible act. Checking the claims is the work of whoever reads it — which is exactly why it has to be readable.
- Isn't this just a better summary format?
- The format is part of it; the rest is where the note lives and who controls its end. A summary in a file or a chat is read by whoever happens to load it and disappears when it is overwritten. A handoff is served first to every session in the project, shows whether it has been delivered, and cannot be removed by an AI. The difference is not the words — it is that there is a record of the note existing, being read and being ended.
- What about the memory or summaries built into my AI tool?
- This page makes no claim about any tool's built-in memory. Arroway works alongside whatever your assistant already does: the handoff lives in the project, where every connected tool reads it and every person in the project can see it, regardless of which assistant wrote it.
Where this is verifiable
Product documentation on this site (How it works, Install), the answers on handing work between scheduled AI routines, on handoffs that are too long to be read and on memory without a write policy, and the handoff, claim, close and discard rules in the sanctioned product spec. The panel described here is the project's list of work in flight. Everything described here is behaviour the tools apply today, not roadmap.
https://www.arroway.app/en/answers/auditing-what-your-ai-hands-to-the-next-session