PolicyLayer Connect your agent
PolicyLayer for Claude Managed Agents

Give it a way to ask and keep going.

When the playbook has no answer, PolicyLayer asks the decision owner in Slack and the rest of the run continues. Repeated answers become playbook rules.

01 / During the run

One open question does not stop the run.

PolicyLayer sets aside only the piece of work the playbook cannot answer. The agent completes everything else on this run.

Scheduled run 08:00
  1. PL-0049 Statement request The playbook answers it ✓ Done
  2. PL-0050 Address update The playbook answers it ✓ Done
  3. PL-0051 Late fee during outage No answer in the playbook Waiting on answer
  4. PL-0052 Card delivery update The playbook answers it ✓ Done
  5. PL-0053 Account closure confirmation The playbook answers it ✓ Done
The result Four of five done. One question waits for its owner.
Slack · #policy-questions 09:41
PL-0051 · Consumer support

Should we waive this £15 late fee?

The customer attempted payment during the confirmed outage. The playbook does not cover this.

Answer View case
Meanwhile The run has already moved on.
02 / The answer returns

The agent gets what it needed.

The answer went to the agent that needed it, not into a thread nobody will find again.

One case across two scheduled runs PL-0051
01
Tuesday, 09:52

The owner answers in Slack.

“Yes, waive it. The outage was ours.”
02
Wednesday, 08:00

The next run finishes the work.

The agent collects the answer and completes what it set aside.

03 / The compounding loop

Answers compound into authority.

01 The question repeats
PolicyLayer spots the pattern.

Third late-fee question this month, all answered the same way.

02 One click to approve
A rule is drafted for the owner.

Waive late fees up to £15 during confirmed outages. Approve, edit or reject in Slack.

03 The playbook grows
Every future run starts smarter.

The next signed playbook carries the rule. The agent settles matching work itself.

04 / Setup

Add one decision path. Keep the agent you built.

Add it once, then run: the next scheduled session starts from the approved playbook.

  1. 01

    Import your playbook

    The rules your agent follows today become playbook version one.

  2. 02

    Connect Slack

    Choose where questions and approvals land.

  3. 03

    Add the credential

    Reveal your agent key once and store it in Anthropic's credential vault.

  4. 04

    Add PolicyLayer to the agent

    Paste the generated MCP server and instructions into your Claude Managed Agent.

Model · unchanged Agent code · unchanged Tools · unchanged Sandbox · unchanged Schedule · unchanged
05 / Evidence

Proved on real scheduled runs.

After seven approved rules
40% → 20%
Cases needing a person
6 → 0
Repeat questions
See the full evidence →
06 / Fit

Beside your agent. Not inside its execution path.

Does it replace my agent's existing tools?

No. The agent keeps its model, tools, sandbox and schedule. PolicyLayer supplies company answers and rules.

Does it write to customer systems?

No. PolicyLayer returns words and rules only. Your agent remains responsible for every action.

What if PolicyLayer is unavailable?

The agent falls back to its last signed playbook and gains no new authority.

Machine authority compounds
The run doesn’t stop. The agent gets what it needs.
Connect your agent