Free checklist
AI Agent Readiness Checklist
Use this before you trust an agent with real workflows, real data, or meaningful execution. The point is not to sound sophisticated. The point is to decide whether your current setup is safe enough to own useful work without creating hidden risk, hidden cost, or hidden cleanup.
Authority
Inputs
Review gates
Proof
How to use this in 20 minutes
- Pick one agent or one workflow you already use.
- Score each category from 0 to 2.
- Circle the first red-flag issue you find.
- Write the single safest workflow this stack should own next.
- Decide whether you need cleanup, a stronger operator baseline, or a real implementation project.
Scoring rule: `0 = unclear or unsafe`, `1 = partly defined`, `2 = explicit and reliable`.
1. Authority
/ 2
- Can the agent's job be explained in one sentence?
- Does it know what it cannot do without review?
- Is there a human owner for the workflow?
Write it: “This agent owns __________________ and must escalate when __________________.”
2. Inputs
/ 2
- Are model names, links, paths, and IDs verified?
- Are secrets kept outside template files?
- Are example files sanitized?
Write it: “This workflow reads truth from __________________, not from guessed values or memory.”
3. Review gates
/ 2
- Which actions always require human approval?
- Which actions are only safe in non-production?
- What is the rollback path for meaningful changes?
| Always review | billing, publishing, auth, destructive actions, new external integrations |
|---|
| Usually safe | drafting, formatting, local summaries, deterministic internal prep |
|---|
4. Proof
/ 2
- What exact artifact proves the task worked?
- Where does the output land?
- How do you tell service health from workflow health?
Write it: “Success means the artifact __________________ exists at __________________ and looks like __________________.”
5. Spend control
/ 2
- What is the default cheap model tier?
- What documented failure justifies escalation?
- Who approves premium model usage?
| Good enough default | cheap or free model for drafts, routing, and repetitive low-stakes work |
|---|
| Escalation trigger | quality failure, rate limit, or clearly documented stability problem |
|---|
| Bad habit | using expensive models because nobody wrote a default policy |
|---|
Red flags that mean “do not scale this yet”
Stop
The agent can publish, spend, or modify production without a review gate.
Stop
No one can point to the exact file, queue item, report, or artifact that proves success.
Stop
The workflow still depends on remembered links, hidden prompts, or manual secret copying.
Readiness score
| 0-3 | Not ready Your stack is still experimentation. Keep the workflow narrow and add review gates first. |
|---|
| 4-7 | Partly ready You have signal, but one or two missing controls will cause avoidable cleanup. |
|---|
| 8-10 | Operationally usable One internal workflow is probably ready to own useful work with supervision. |
|---|
Decision output
Fill this before you move on:
Current workflow audited: __________________
Biggest control gap: __________________
First workflow safe enough to trust next: __________________
Best next step: cleanup / operator template pack / systems diagnosis / custom build
Next step
If the checklist exposed missing authority, missing review gates, or weak proof standards, the next useful step is the AI Operator Template Pack. If it exposed broader systems debt across multiple workflows, move into a Systems Diagnosis.