Two people, two decisions that clash
Small teamsTeam where two leads own overlapping areas
Anil recorded 'soft delete everywhere' in March. Ritu recorded 'hard delete user data on request' in October for privacy requests. An agent building account deletion got both and stopped to ask which one wins.
What do I do when my AI agent finds two team decisions that contradict each other?
With ArcRoom, settle the clash once: the leads agree, then one records a decision that replaces the other, or marks the old one with arcroom wrong. From then on agents get one answer. ArcRoom doesn't detect conflicting decisions by itself. It relies on the agent noticing the clash and asking, and on people settling it.
Also asked as: Our AI agent got two conflicting rules from different leads. Which one wins? · How do I retire an old team decision so agents stop using it? · Can an AI agent ask instead of guessing when two team decisions conflict?
Recall can hand an agent two decisions that both fit the prompt.
- Decision
- Recall
- Wrong-memory report
Decision (Mar 4, anil): soft delete everywhere.
Step by step
- Recall can hand an agent two decisions that both fit the prompt.
- A good agent spots the clash and asks instead of guessing.
- The leads agree, and one of them records a decision that replaces the other, or marks the old one with arcroom wrong.
- From then on agents get one answer.
What the agent sees
Decision (Mar 4, anil): soft delete everywhere.
Decision (Oct 9, ritu): hard delete user data on an erasure request.
Agent: these conflict for account deletion. Which applies?Compared with today's tools
Architecture decision records (ADRs) in the repo
Each decision is a markdown file; a new ADR marks the old one as superseded in review.
- Setup
- 1 hour
- Ongoing
- write and review an ADR per decision, ~20 min
ArcRoom
Every agent reads the team playbook before it works and writes to it when it finishes. Nothing to update by hand.
- Setup
- 2 min per person (npx arcroom init)
- Ongoing
- None
ADR review catches the clash before anyone builds, which is better; ArcRoom gets the current decision in front of every agent without being asked.
Where it stops
ArcRoom doesn't detect conflicting decisions by itself. It relies on the agent noticing and people settling it.
Try it in your project
Works with the agents your team already uses, and any agent over MCP or the HTTP API.