CHANGELOG · 7th September 2026
Rank the work against the bets you have made.
Name the few outcomes you are betting on, and every priority groups and ranks under the bet it serves.
Bets name the outcomes you are betting on, and the priority board groups and ranks the work under them.
A ranked list answers which item is loudest. On its own it does not answer whether the loudest item serves what you set out to do this quarter, and evidence that is not a customer quote has nowhere to count.
You type a bet in place on the board, in your own words, with the one number that says whether it is happening. Priorities group under the bet they serve, the board says which bet is first, and a bet with nothing under it still shows, so what you have not started is as visible as what you have. Bets are part of the welcome, live under Settings and are reachable from your coding agent.
How to use it
A ranked list tells you what is loudest. A bet tells you what you are trying to make true, so the board can show whether the loudest work serves it.
- Type a bet on the board in your own words, with the one number that says whether it is happening.
- Place priorities on the bet they serve. A bet with nothing under it stays visible.
- Rank the bets. The order sets what gets worked on first.
Also in this release
Record a decision, and what would show it was wrong
A priority closes on one of 4 exits, signed by a person, with the thing that would show the call was wrong and the date to check it.
A spec reads as a document, at any width
The priority sheet reads top to bottom like a document on desktop, tablet and phone, with the evidence first and each fact said once.
Ask keeps the thread
A follow-up question sees the turns before it, and the number Ask gives you matches the list it shows.
One message per person, however much shipped
Someone whose feedback sits on 4 shipped priorities hears once, with every item keeping its own answer.
Your product's own words ground the work
Annsa reads your published pages for the words your product uses, and grounds naming and themes on them.
Improvements
- Priorities and the board
- Bets and progress read as grouped bands in one table. Priority titles are editable in place. Splitting a priority asks for a name and ranks the new one on its own evidence. A priority with one signal reads as one, and the board filters by status and by bet.
- Transcripts
- A quote carries the speaker's whole turn. Names show in quotes, the speaker gate reads role labels as well as names, and an import with no signal in it says why.
- Coding agents
- Every MCP reply names the environment, workspace and caller, and refuses another workspace's id in words. A spec is named by the priority's number, the same number on the board. An agent can place a priority on a bet, and a capability Annsa does not carry is answered as unknown.
- Roadmap
- A bet's priorities sit in that bet's row whether or not they have a spec. Columns collapse from the header, and the agent holding a card is marked on it.
- Welcome and reset
- Enter advances every field in the welcome. A workspace reset reaches every open tab, clears bets with the rest, and a tab opened later picks it up.
- Billing and delivery
- A retried request is recognized as the same request, so it is charged once. Links in a shipped-work email reach their destination and replies route to your workspace. A destructive action reports its outcome once that outcome is known.