Your agent works in the demo. Is it ready for real money?

The Agent Readiness Kit is a ten-chapter field guide and 14 editable templates for deciding what an AI agent may do on its own, proving it with evals and drills, and knowing exactly what to do when it breaks.

Read a free sample

PDF plus Markdown, YAML and CSV templates. Instant download. Secure checkout by Stripe.

agent-readiness-kit.zip$19
  • agent-readiness-kit.pdf
  • templates/01-architecture-contract.md
  • templates/02-permission-matrix.csv
  • templates/02-tool-risk-tiers.md
  • templates/03-approval-and-dry-run-policy.yaml
  • templates/04-golden-task-set.yaml
  • templates/05-eval-scoring-rubric.md
  • templates/06-failure-drill-cards.md
  • templates/07-telemetry-retention-budget.csv
  • templates/08-incident-and-quarantine-runbook.md
  • templates/09-agent-roi-ledger.csv
  • templates/10-readiness-checklist.csv
  • templates/README.md

Built for the week after the demo

The prototype answers questions well. Now someone wants it to issue refunds, merge pull requests, or update customer records without a person clicking approve every time.

That's where most teams stall. The model isn't the problem. Nobody has written down what the agent is allowed to do, how you'd know it's misbehaving, or who gets paged when it does.

This kit gives you those answers as documents you fill in, not essays you agree with. Each template comes with a complete worked example, so you can see what a finished one looks like before you start yours.

Four levels of trust

The kit is organized around a readiness ladder. Every template helps you earn the next rung, and tells you what has to be true before you take it.

  1. Shadow

    Drafts only

    The agent drafts. People do everything. You collect evidence for free.

  2. Supervised

    Acts on approval

    The agent acts only after a person approves each action.

  3. Bounded

    Acts within limits

    Alone inside explicit limits on amount, tool and time. Everything else goes to a person.

  4. Autonomous

    Acts, then reports

    Alone across its scope. People review samples and exceptions.

What's inside

Ten chapters, each ending in a template you fill in that week.

  1. 01

    How to use this kit

    The readiness ladder, who fills in what, and a two-week plan.

  2. 02

    Write the architecture contract

    One page that says what the agent is for, what it may touch, and who owns it. Template 01

  3. 03

    Map tool permissions and blast radius

    Rate every tool action, then give each one the narrowest credential that works. Template 02

  4. 04

    Put approvals and dry runs where they matter

    Approval requests a busy person can judge in ten seconds, and previews of exact side effects. Template 03

  5. 05

    Build an eval suite before you need one

    A golden task set with injection and must-escalate cases, plus a scoring rubric. Templates 04, 05

  6. 06

    Run failure drills

    Twelve drill cards for the failures demos never show you. Template 06

  7. 07

    Observe without hoarding

    What to log for every run, what to throw away, and how long to keep the rest. Template 07

  8. 08

    When it breaks

    Quarantine mode, severity levels, and a first-hour runbook. Template 08

  9. 09

    Prove it pays

    A weekly ledger that counts review time and rework, not just deflection. Template 09

  10. 10

    The readiness review

    A 60-point scored checklist and a go or no-go meeting you can run in an hour. Template 10

A look at the readiness checklist

Sixty statements you can verify, scored 0 to 2, with the ones that block each level marked. These are six of them.

Download the free sample
  • AC-01The architecture contract (template 01) states the agent's purpose in one sentence and lists the actions it must never take.
  • PC-01Every tool action has a row in the permission matrix (template 02) with a risk tier and a maximum blast radius.
  • AD-01The approval policy (template 03) lists every action that needs human approval and the threshold that triggers it.
  • EV-01A golden task set (template 04) built from real traffic covers normal, edge and adversarial cases for every in-scope task type.
  • FD-01Every failure drill card (template 06) that applies to the actions under review was run in the last 90 days.
  • OB-01Every run has a trace id that links the input, each model call, each tool call with parameters, and the final outcome.

Questions

Anything else, email guide@alchemictechnology.com.

What do I actually get?

A zip with the PDF field guide and the editable templates in Markdown, YAML and CSV. Copy them into your repo, wiki or spreadsheet and fill them in.

Which agent frameworks does it work with?

Any of them. The kit is about decisions, limits and evidence, not a particular framework or model provider. If your agent calls tools that change real systems, it applies.

I'm a solo developer. Is this overkill?

Use the parts that match your risk. Chapter 1 tells you which templates a one-person team needs before letting an agent act unattended, and which can wait.

Is the worked example a real company?

No. Northwind Supply and its Refund Desk agent are fictional, so every template has a complete filled-in example you can compare against.

How is it delivered?

Right after checkout you get a download link on the confirmation page and by email. The link works five times over 30 days.

Can I share it with my team?

Yes, inside your organization. Please don't resell or publish the templates.

Give your agent a job description.

Fourteen templates, one worked example, and a review you can run in an hour. $19, delivered instantly.

Free sample