Guide · 8 min read
AI Agent Readiness Checklist
Not every workflow is ready for an AI agent, and that is not a judgment on the business. Readiness depends on how clearly the work is defined, whether the agent can reach the information it needs, and whether someone can supervise it.
Work through the six sections below for one specific workflow. If most answers are 'yes', it is a strong candidate. If several are 'no', the gaps tell you what to address first — often in days, not months.
1. Process clarity
- Can someone describe the workflow step by step in under ten minutes?
- Do experienced staff largely agree on how it should be done?
- Are the rules for common decisions written down, or could they be?
- Can you collect twenty or more real past examples, including awkward ones?
2. Volume and value
- Does the workflow happen many times a week?
- Does it consume meaningful time from skilled people?
- Would faster turnaround matter to customers or colleagues?
- Is there a clear owner who wants it improved?
3. Data and knowledge
- Are the inputs digital (email, forms, documents, records)?
- Is the information needed to complete the task accessible in a system?
- Is the relevant documentation reasonably current?
- Do you know which data is sensitive and who may see it?
4. Systems and integration
- Do the systems involved have APIs, webhooks, or another reliable access method?
- Can you create a separate account or credentials for an agent?
- Is there a test or sandbox environment?
- Does someone own each system and can approve access?
5. Risk and control
- Can mistakes be detected and corrected?
- Can you name the actions that must always need human approval?
- Are there legal, contractual, or regulatory constraints you know of?
- Would you be comfortable explaining the agent's role to customers or auditors?
6. Supervision and measurement
- Is someone available to review the agent's work during the first weeks?
- Can you measure how the workflow performs today?
- Have you agreed what 'good enough' looks like?
- Is there a plan for who maintains the agent after launch?
Reading your results
Mostly 'yes': the workflow is a strong candidate. Next, scope the agent's role in detail and plan a supervised pilot.
Gaps in sections 1 or 3: document the process and fix the information sources first. An agent cannot be more consistent than the inputs it relies on.
Gaps in section 4: an integration assessment is the next step. Many access problems have practical solutions.
Gaps in sections 5 or 6: start with a copilot that assists people rather than an agent that acts. It builds evidence and confidence with less risk.