How it works
How Backbrief works
One long chat cannot hold a project. And it cannot check itself.
You have felt both halves of this. Somewhere past the hundredth message, the thread starts to unravel: earlier decisions get dropped, the same ground gets re-covered, and you end up holding all the coordination in your own head. The quieter failure is worse. A single agent that writes the work also signs off on the work. It rubber-stamps itself, and whatever it missed rides along into what you ship.
The fix is not a longer chat and not a smarter prompt. It is a team with a separation of duties. One agent owns one job. The plan lives in a file instead of your memory. And a separate verifier, with fresh context and no attachment to the draft, reads the result before it goes anywhere. That is what the kit installs.
Goal in, verified work out.
01
Goal. You state what you want and what done looks like.
02
Plan. The orchestrator reads your context and writes the plan to a file.
03
Route. Each piece goes to a specialist agent, or to a cheap runner for mechanical work.
04
Verify. A fresh-context verifier checks the result before it is called done.
Before you trust it, watch it say no
We tested it on the maker's business plan.
C, 5.9 of 10. Below the 8.5 pass bar. Two independent panels a month apart graded the same material C, 6.4 then 5.9, and put the same two dimensions on the floor. Financial viability, with the disagreement kept in the record, not averaged away:
“contrarian 2 (‘a margin floor with no reconciled cost stack is a wish’) vs. expansionist 5.”
A recorded review of NestPet, the maker’s pet memorial business. This demonstrates the review process, not a customer outcome.
This is Backbrief DIY working, in the $79 bundle.
The commands
Two commands you will feel on day one.
/critique puts one artifact in front of one adversarial reader and asks it to find what is wrong, with citations, before anything ships.
/council is the fastest way to feel the whole system: paste a decision you are stuck on and five advisors with distinct lenses debate it, then a chairman gives the call and names where they clashed hardest.
You saw the council run for real in the 6.4 scorecard above: a 3 and a 6 sitting side by side in the record, because that gap is the signal.
Disagreement gets reported, not averaged away. That is the point.