Map
Record the current workflow and baseline effort.
GUIDES / SUPERVISED AGENTS
What an always-on agent needs beyond a VPS.
An AI agent that runs unattended is an operational commitment, not a feature flag. Before one touches real work, it needs boundaries, supervision, measurement and a way to stop it — this guide covers what that means in practice.
Check Your Digital Readiness01 / SUITABLE WORK
Workflows that suit a supervised agent.
- Prepare a weekly website-owner briefing
- Classify enquiries and draft internal routing notes
- Turn approved evidence into content briefs
- Watch public journeys and create actionable alerts
- Maintain a non-secret dependency register
- Assemble evidence for a Readiness review
- Coordinate an incident timeline
02 / THE METHOD
A supervised pilot, step by step.
Bound
Define permitted data, prohibited data and tools.
Approve
Mark human approvals and irreversible actions.
Measure
Define quality, cost, latency and failure measures.
Test
Probe injection, leakage and shutdown controls.
Decide
Conclude fit, expand or not yet.
03 / BEFORE IT RUNS
What the business should own first.
- A map of the workflow the agent will touch, with its baseline effort
- A permission and authority matrix — what it may read, recommend and do
- A data and privacy map covering permitted and prohibited data
- An operating charter with human-approval points and prohibited actions
- Evaluation fixtures for quality, cost, latency and failure
- An incident and shutdown process that works without the agent's cooperation
An agent does not replace accountable people, and no agent's accuracy is guaranteed. Bounded, supervised and reversible is the standard — anything less is not ready for real work.
YOUR NEXT STEP
Start with the problem - not the supplier.
Identify the priorities, understand the responsibility and then continue through the route that fits the work.