# Plug & Play - AI-driven operations
## Sample maintenance and handover runbook

An illustrative template by Nikolai Søndergaard. It demonstrates the handover included with a tailored operating setup. Replace each planning field and test every service-specific procedure before relying on it. This is not documentation for a live client system.

### 1. Ownership and inventory

- Name a primary system operator, backup operator and business owner for each workflow.
- Record the service, its purpose, hosting location, dependencies, update method and owner.
- List approved email sources, chat channels, website/social integrations and back-end connections.
- Record what each integration may read or change, its credential-store reference, expiry and revocation procedure. Keep secret values out of this document.
- Record where project knowledge, decisions, task status and evidence live. Identify the authoritative source when several tools show the same item.
- Record provider/support contacts and the escalation route. Do not assume the original builder is the permanent incident contact.

### 2. Routine automation

Agree the automation policy per service. Possible automatic checks include:

- Service reachability and scheduled-job completion signals.
- Backup completion, storage capacity and credential/certificate expiry reminders.
- Agreed low-risk operating-system security updates, subject to service and restart constraints.
- Notifications for meaningful exceptions, including the affected workflow, evidence, owner and next step.

Do not interpret a successful heartbeat as proof that the complete business workflow works. Use a harmless representative workflow check as well. Alert on the absence of expected check-ins, not only on explicit errors. Avoid repeated notifications for the same unresolved event.

### 3. Weekly operator review

1. Review unresolved alerts and jobs that failed, stalled or stopped reporting.
2. Check backup completion and the date of the last successful restore test.
3. Check capacity, upcoming credential expiries and changes to third-party connections.
4. Review work without an owner, blocked agent tasks and unaccepted handovers.
5. Record the action taken and any risk requiring the business owner's decision.

### 4. Controlled updates

Application, agent, workflow and connector updates follow the agreed change process.

1. Identify the affected services, workflow owners and expected impact.
2. Read release notes and check compatibility with the current setup.
3. Preserve the working configuration and record the current version.
4. Verify a recoverable backup and an accessible rollback route before changing anything.
5. Test in an isolated environment where practical. Agree a maintenance window if disruption is possible.
6. Obtain the designated owner's approval for consequential changes.
7. Apply the change and verify a representative workflow, its permissions and its visible outcome.
8. If checks fail, use the documented rollback procedure without overwriting newer business data.
9. Record the version, checks, result and remaining follow-up in the change log.

### 5. Incident response

1. Identify the affected service, last successful operation and business impact.
2. Pause repeated failing or externally acting work if needed. Preserve relevant logs and state.
3. Check the service-specific runbook and recent changes before proposing a repair.
4. Before retrying, confirm which actions already completed to prevent duplicate messages or records.
5. Use the approved recovery or rollback steps. Escalate when the cause or safe recovery path is unclear.
6. Verify recovery through the actual user-facing workflow. A restarted process alone is not enough.
7. Tell the owner what recovered, what remains uncertain and what needs follow-up.

### 6. Knowledge and agent handovers

Every handover should include the intended outcome, current state, evidence, open questions, next action and named owner. An agent's finished run does not mark the task accepted. Separate draft, implemented, tested, reviewed and accepted states.

Maintain reusable skills alongside the process they describe. Custom Model Context Protocol (MCP) servers expose only approved tools. Review permissions when a workflow changes, and revoke access that is no longer needed.

### 7. Backup and recovery drill

Agree what is backed up, where it is held, retention, protection and the acceptable recovery time and data loss. A successful backup job is not a restore test.

A backup operator should periodically restore a sample into an isolated location, verify that the necessary data and configuration are present, run a representative workflow and record the result. Keep recovery credentials available through the approved store without publishing them in the runbook.

### 8. Handover acceptance

Before handover is accepted, a second operator should be able to:

- Find the inventory, owners, support contacts and authoritative knowledge records.
- Access the agreed credential store and explain the permission boundaries.
- Identify a failed job and follow its escalation route.
- Pause an agent or workflow and distinguish pause from completion.
- Explain how an update is tested and rolled back.
- Complete a sample recovery exercise and record the verification.

### Operating expectation

The aim is minimal routine maintenance and minimal reliance on the original builder. Hosting, third-party services, credentials and unusual incidents still need an accountable operator. Define any ongoing support arrangement separately.

More: https://sondergaard.io/organization-in-a-box/
Contact: https://sondergaard.io/contact/
