Judgment record
CASE PL-0051 No rule covers this Question routed
Should we waive this £15 late fee?
Charged during a confirmed outage. PolicyLayer routes the question, parks the case and lets the agent keep working.
The system of record for AI agent authority
Never answer your agent twice.
When your agent hits a company decision its instructions don’t cover, PolicyLayer asks the owner and lets the agent keep working. The owner’s answer lets the agent finish that case. Human approval turns that decision into a rule every agent can reuse.
Connect your agent ↗
Prefer to talk first? Ask about a design partnership
- 01Your team answers once
- 02The agent handles more on its own
- 03Every run starts smarter
Ask once. Approve once. Reuse the rule.
The judgment loop
PLAYBOOK v2 · 05 STEPS The result The next time the same case comes up, the answer is already there.
01 / 05 · Policy checked
Seven rules. Half as many cases needed a person.
Across more than 50 real support cases, human decisions became seven approved rules. The agent stopped pulling the same people into the same questions.
In the next 20 cases
Every approved rule expanded what the agent could handle on its own and freed the team from repeating work.
Cases requiring human judgment
40% Before
20% After seven rules
Repeat questions6 → 0
New rules7
Less time answering the same question. More cases closed without waiting on a human.
Only human approval changes company policy.
PolicyLayer can capture answers and draft rules. Agents receive new authority only after a person approves it.
An answer applies only to the case that asked for it.
Until a rule is approved, the answer cannot grant authority in another case.
Rules exist only after a named human approves them.
PolicyLayer drafts from repeated decisions. A person approves, edits or rejects every rule.
Every agent runs the same signed playbook version.
Approved rules go into the next signed playbook version and reach every agent that uses it.
Unreachable means less autonomy, never more.
If PolicyLayer is unavailable, the agent falls back to its last signed playbook and gains no new authority.
Your company owns every playbook version.
Export every version and keep the same company judgment when you change model or provider.
Three MCP tools.
Keep your agent’s model, tools and schedule. Add three PolicyLayer tools; handle questions and approvals in Slack.
Running a Claude Managed Agent? See the exact setup →get_playbook Load the latest signed company playbook.
ask_policy Park the unresolved case and route one question to its decision owner.
get_answer Return the decision with its owner and context.
Existing agent → PolicyLayer MCP → Decision owners in Slack
Any model. Any agent framework. One source of company authority.
PolicyLayer uses standard MCP. Your company’s rules stay intact across models, frameworks and MCP clients.
Agents are going to run the world’s companies. PolicyLayer is how company judgment becomes machine authority.
PolicyLayer is the system of record for AI agent authority.
Connect your agent ↗
Prefer to talk first? Ask about a design partnership