What is an AI activity and decisions ledger?

Every team loses the same thing: the reasoning behind its own decisions. Six months later someone asks why the service is built this way, and the answer is buried in a Slack thread nobody can find, or it left with the person who made the call. The code records what changed. It does not record why.

The usual fixes ask for discipline that never holds. Architecture decision records are great in theory and abandoned in practice, because writing them is a separate chore from doing the work. Changelogs capture events, not reasoning. The honest question is: what if the AI that was already in the room, doing the work, wrote the record itself? That is an AI activity and decisions ledger.

The two streams: activity and decisions

The ledger has two streams because they answer different questions.

• Activity is the high-frequency record of what happened: what shipped, what changed, what ran. It is the team's work journal, written continuously.
• Decisions are the lower-frequency record of what got settled, with the reasoning attached. These are the calls that stick, and the why is the valuable part.

Keeping them apart is deliberate. It means your AI can answer "what did we decide about pricing" cleanly, without making you wade through a feed of every commit. A third stream, tasks, points forward: who owns what, by when. Past, settled, and next, in one place. The ledger sits inside the broader shared context every tool reads.

What does "AI-witnessed" mean?

This is the part that makes the ledger work where ADRs fail. The record is written by the AI that was already doing the work, not gathered from elsewhere and not logged by hand. Three things follow from that, and together they are the differentiator:

• No scraping. The ledger is not stitched together from your other systems. It is the deliberate record of the work itself.
• No manual logging. Nobody has to remember to write it down. When the assistant finishes a task or recognizes a decision in the conversation, it writes the entry.
• AI-witnessed. The record comes from the tool that watched the work, so it captures the reasoning while it is fresh, not reconstructed weeks later.

The reason ADRs get abandoned is that logging is a separate task from working. An AI-witnessed ledger removes the separation: writing the record is part of doing the work. We make the full case in manual context-logging is dead.

Why a decision log beats threads and changelogs

Compare the ways teams capture what they decided. Most leak in the same place: the why.

Approach	Who writes it	Stays current	Captures the why	Readable by your AI
Slack threads	People, scattered	No	Sometimes	No
Manual ADRs	People, if disciplined	No	Yes	No
Changelog / commits	People	Partly	No	Partly
AI-witnessed ledger	Your AI, as work happens	Yes	Yes	Yes

The last row is the only one that gets all four right, and the reason is the same throughout: the record is a by-product of the work, written by a tool that is already there, in a form other tools can read. For engineers specifically, this is auto-generated ADRs without the discipline tax. For the team, it is part of the context problem most tools never solved.

What happens when teammates' AIs log overlapping things?

In a real team, several people's tools will write about the same work. A naive log would fill with duplicates and contradictions. The ledger reconciles them, a step called harmonization, and it handles three cases:

• Corroboration. Three people independently log the same decision. Instead of three duplicate rows, they merge into one canonical decision with a count, which is a strong signal the whole team agrees.
• Supersession. A newer decision replaces an older one. The old entry is kept and linked, marked superseded, so the history of why is intact rather than overwritten.
• Conflict. Two decisions contradict each other. The ledger flags them, links them, surfaces the conflict to the people involved, and leaves it for humans to settle rather than silently picking one.

That reconciliation is what turns many people's individual logs into one trustworthy team record. It is also something a personal memory feature cannot do, because it never sees more than one person.

How the ledger stays current

The ledger is part of the same loop as the rest of shared context: tools read the relevant slice at the start of a session, and write activity and decisions back when they finish. Because writing is automatic and tied to the work, the record does not rot the way a wiki or an ADR folder does. The next teammate, and the next session, start already caught up. The compare page shows how this differs from the memory and wiki approaches that have no such record at all.

Related reading

What is shared context for AI tools? (2026 guide)

Shared context for AI tools is the company, project, and decision background every AI reads automatically, so your whole team's tools stop guessing.

Manual context-logging is dead: your AI should witness the work

Manual context-logging never holds, because writing it is a chore separate from the work. The fix: let the AI that did the work witness and record it.

The team-context problem nobody has solved yet

Every AI tool solves context for one person. The team-context problem, one shared, current context across every tool and teammate, is the gap nobody filled.

Bigger context windows won't fix team knowledge

Every model ships a bigger context window, but team knowledge is not a capacity problem. Here is why more tokens won't fix what shared context solves.