Skip to content
Agent Mode AI
Tools · Implementation

Agentic AI implementation readiness checklist

The questions to answer before a pilot becomes a procurement decision. Not a bingo card — each item is a real gate, and most enterprise programmes stall because at least one is not yet answered.

Tool status. A PDF version of this checklist circulated during the WordPress era. The content has been folded into the canonical page here so the checklist updates when the underlying evidence updates.

1. Problem definition

  • Can you name the single workflow the agent will own end-to-end?
  • What breaks today when that workflow runs? (Specific failure mode, not vague "inefficiency".)
  • Who owns the outcome the agent is accountable to?

2. Data and identity

  • Is every data source the agent will touch already on the approved list?
  • Does the agent have its own non-human identity, or is it reusing a human's credentials?
  • Can you audit the agent's actions to a specific request-ID chain?
  • What happens to the agent's access when the requesting human leaves the org?

3. Threat model

  • Prompt-injection surface: has the agent been tested against documented in-the-wild attacks?
  • Cross-agent delegation containment: can this agent call other agents, and is that chain logged?
  • Kill-switch: who has the authority to stop the agent, and how fast does it take effect?

4. Governance and compliance

  • Does the agent fall under EU AI Act Title III high-risk obligations for your deployment geography?
  • Is there a named Data Protection Impact Assessment (DPIA) for agents acting on behalf of employees or customers?
  • What's the retention policy for agent-generated artefacts (outputs, logs, intermediate reasoning)?

5. Economics

  • What is the per-action cost, including inference and orchestration overhead?
  • Does the business case survive a 2× price increase from the vendor?
  • At what adoption fraction does the per-seat or per-call economics break?
  • What is the exit cost — how long to detach, and what data do you lose?

Related tracked claims

The questions above reference ongoing tracked claims under our Holding-up index. Pricing-tier claims specifically: AM-003 (GPT-5 Pro economics, reviewed every 30 days).

Vigil · — reviewed