# Agent Readiness Self-Check: score one workflow before you hand it to an AI agent

A free self-check by PaperChaseWebb for owner-led agencies and small businesses. Pick one workflow, score it across eight areas with evidence, apply three gate rules, and read your band: not yet, draft-only pilot, ready to install one workflow, or ready to run more than one. Every band comes with what to do next, including the honest answer when the answer is "fix the basics first". Includes an interview prompt that has Grok Bot, Claude Code, Codex or Hermes run the self-check with you and score it, a score sheet to copy, and a worked example. Plain .md. No signup, no email gate, the whole thing is in this file.

Most free AI readiness quizzes ask ten vague questions about your whole company, hand back a number like 62, and put the same "book a call" button under every result. This one is narrower on purpose. It scores one workflow, it asks for proof instead of feelings, and it checks the parts that decide whether agents help or hurt: who approves, who answers for each agent, how work passes between agents and people, what it costs and how you will know it worked.

## Contents

1. How to use this in 30 minutes
2. The eight areas and how to score them
3. The evidence rule
4. The three gate rules
5. Bands and what to do at each one
6. The two-pass honesty test
7. Approval steps for the self-check itself
8. The interview prompt (Grok Bot, Claude Code, Codex, Hermes)
9. The score sheet
10. Worked example (illustrative)
11. Where this leads
12. What we looked at, and what we did differently
13. Sources


## 1. How to use this in 30 minutes

1. **Pick one workflow.** One job that repeats, with a start and a finish. "New inquiry to proposal draft" works. "Marketing" does not. If you score the whole business you will get an average that describes nothing.
2. **Get the person who does the work today.** Not only the person who wants it automated. They know where the exceptions live.
3. **Score each area 0 to 3** using the anchors in section 2. Write one line of evidence next to every score.
4. **Apply the evidence rule and the gate rules** (sections 3 and 4). They can lower your band. They never raise it.
5. **Read your band** and do the first item on its list (section 5). Rescore the same workflow in two weeks.

You can do this on paper, in a doc, or by having an agent interview you with the prompt in section 8.

## 2. The eight areas and how to score them

Score what is true today, not what is planned. Each area runs 0 to 3, so the total runs 0 to 24.

### 2.1 Workflow mapped

Can someone who has never done this job follow it from a written page?

| Score | What it looks like |
|---|---|
| 0 | It lives in one person's head. Two people would describe it differently. |
| 1 | Rough notes exist, or the steps are clear but the exceptions are not written anywhere. |
| 2 | A one-page description with inputs, steps, outputs and the common exceptions. |
| 3 | The page is current, a second person has followed it, and you can say what "done" looks like for each run. |

### 2.2 Data access

Can an agent reach what the job needs, without reaching what it should not?

| Score | What it looks like |
|---|---|
| 0 | The inputs are on paper, in personal accounts, or scattered across chats. |
| 1 | The inputs are digital but spread across places nobody has listed. |
| 2 | You can name every source the job uses, and each one can be exported, shared to a folder or reached by a connector. |
| 3 | As 2, plus you have marked which sources the agent may read, which it may never touch, and you have past examples (including hard ones) to test with. |

### 2.3 Approval gates

Before anything leaves the building, who says yes?

| Score | What it looks like |
|---|---|
| 0 | Nobody has decided. An agent could send, post, spend or delete if a tool allowed it. |
| 1 | People "know" certain things need a check, but it is not written down. |
| 2 | A written list of actions that need a person's yes: sending, posting, spending, deleting, signing, changing a customer record. |
| 3 | As 2, plus each gate names who approves, how fast, and what happens when that person is away. The tools are set so the agent cannot skip the gate. |

### 2.4 Owner per agent

When an agent gets something wrong, whose problem is it?

| Score | What it looks like |
|---|---|
| 0 | "The team" or "whoever set it up". Nobody by name. |
| 1 | A person is in charge in theory but has no time set aside. |
| 2 | One named person per agent or workflow, with time each week to review its output. |
| 3 | As 2, plus a written way to pause the agent, and the owner has done it once in a drill. |

### 2.5 Handoffs

How does work move between people, between agents, and between an agent and a person?

| Score | What it looks like |
|---|---|
| 0 | By memory and direct messages. Context gets lost and people re-ask. |
| 1 | A shared channel or board, but each handoff carries whatever the sender felt like writing. |
| 2 | A short handoff packet: the goal, what is done, what is open, what to check first. |
| 3 | As 2, plus the receiver confirms before the sender lets go, and a handoff is not complete until it does. |

### 2.6 Cost tracking

Do you know what this workflow costs in AI spend, and when to stop?

| Score | What it looks like |
|---|---|
| 0 | Nobody knows which AI tools are paid for or by whom. |
| 1 | You know the subscriptions, but not what any single job costs. |
| 2 | Business accounts for AI tools, a monthly number per tool, and a rough cost per run of this workflow. |
| 3 | As 2, plus a spending limit set in the tools themselves and a written rule for what happens when it is reached. |

### 2.7 Security and vetting

What has access, and who checked it?

| Score | What it looks like |
|---|---|
| 0 | Client or customer data goes into personal AI accounts, and tools, plugins or skills get installed without a look. |
| 1 | People are careful, but there is no list of approved tools and no rule for new ones. |
| 2 | A short list of approved AI tools and connectors, business accounts only, and a rule that nothing new gets installed until someone reads what it can access. |
| 3 | As 2, plus access is reviewed on a schedule, unused connectors are removed, and keys are not shared in chats or files. |

### 2.8 Measurement

Will you be able to tell whether the agent made this better?

| Score | What it looks like |
|---|---|
| 0 | No idea how long the job takes or how often it goes wrong today. |
| 1 | A guess ("a few hours", "a couple of days"). |
| 2 | Timed or counted on at least ten recent runs: how long, how many, how many needed a redo. |
| 3 | As 2, plus a written target and a stop rule: what number means keep going, what number means pause or remove the agent. |

## 3. The evidence rule

A 2 or a 3 needs something you can point to: the page, the list, the setting, the log, the spreadsheet. If you cannot open it right now, the score is 1, however sure you feel.

This rule does most of the work. It is the difference between a score that describes your business and a score that describes your hopes.

## 4. The three gate rules

Approval gates, owner per agent, and security and vetting are the gate areas. A high total cannot cover a hole in one of them.

1. **One gate area at 0:** your band is capped at Draft-only pilot.
2. **Two or more gate areas at 0:** your band is Not yet, whatever the total.
3. **Floor rules for the top bands:** Ready to install one workflow needs every area at 1 or more. Ready to run more than one needs every area at 2 or more.

Apply the band from the total first, then apply these rules. If they disagree, the lower band wins.

## 5. Bands and what to do at each one

| Total | Band | What it means |
|---|---|---|
| 0 to 9 | Not yet | An agent would make the mess faster. Fix the basics first. |
| 10 to 15 | Draft-only pilot | One agent can draft for this workflow. A person reads and approves every output. |
| 16 to 20 | Ready to install one workflow | Put this workflow into supervised production, with gates, an owner and a measure. |
| 21 to 24 | Ready to run more than one | The basics hold. The work now is handoffs between agents and running them week to week. |

### Not yet (0 to 9)

Do not buy agent tooling for this workflow yet, and do not hire anyone to install it. You would be paying to automate something nobody has written down.

1. Write the workflow on one page (area 2.1).
2. Name one owner and give them an hour a week (area 2.4).
3. Write the list of actions that need a person's yes (area 2.3).

Rescore in two weeks. Most of the climb out of this band costs time, not money.

### Draft-only pilot (10 to 15)

Run one agent on this workflow in draft mode for two weeks. It prepares, a person approves every output, nothing goes out on its own.

1. Fix your lowest area first. If there is a tie, fix the gate area.
2. Keep a simple log: date, what the agent drafted, approved as is, edited, or thrown out.
3. Start timing ten runs now so area 2.8 has real numbers at the rescore.

Self-serve path: the [Agency AI Ops Workbook](https://paperchasewebb.ai/bots/marketplace/items/agency-ai-ops-workbook-install-sprint-self-serve/) ($39) walks through the same steps our Install sprint covers, so you can do them yourself.

### Ready to install one workflow (16 to 20)

This workflow is a fair candidate for supervised production. You have two honest options.

1. **Do it yourself** with the [Agency AI Ops Workbook](https://paperchasewebb.ai/bots/marketplace/items/agency-ai-ops-workbook-install-sprint-self-serve/) ($39): approval map, owner card, handoff packet, cost limit and scorecard for one workflow.
2. **Have it installed.** The Install sprint ($5,000) on the paperchasewebb.ai homepage sets up the agent operation with you: roles, approval gates, handoffs, cost limits and a scorecard. Use Book a call there and bring this score sheet. It makes the first conversation about your workflow instead of about AI in general.

Either way, write the stop rule (area 2.8) before the agent goes live.

### Ready to run more than one (21 to 24)

The single workflow holds. The risk moves to the seams: work passing between agents, agents that quietly cost more each month, and nobody reviewing week to week.

1. Score the next workflow with this same sheet before adding it. Do not assume the score carries over.
2. Write the handoff between the two workflows as a packet with a confirm step (area 2.5 at 3).
3. Decide who reviews the whole operation each week, not only each agent.

If you would rather have someone run and improve it, the Manage and Expand offers on the paperchasewebb.ai homepage cover that. Book a call there.

## 6. The two-pass honesty test

Score the workflow twice.

1. **Hopeful pass:** give yourself the benefit of the doubt on every area.
2. **Skeptical pass:** score as if a careful outsider will ask to see the evidence for every line.

If both passes land in the same band, trust the band. If they land in different bands, the gap is your real finding: those are the areas where the business runs on assumptions. Fix those before anything else.

A score that gives the same advice no matter how you answer is decoration. This check makes sure yours is not.

## 7. Approval steps for the self-check itself

The self-check is safe to run with an agent because the agent only asks and writes. Keep it that way.

1. **Before you start, decide what the agent may see.** Describe your workflow in your own words. Do not paste client names, customer records, contracts or keys. The agent does not need them to score you.
2. **During the interview, the agent only asks.** It does not open your accounts, connect tools, install anything, search for your business or contact anyone. If it offers to, say no and keep going.
3. **Before the sheet is final, you approve each score.** The agent proposes; you accept, lower or raise with evidence. A raise without evidence stays at 1.
4. **The finished sheet stays with you.** Nothing gets sent or shared unless you choose to share it, for example by bringing it to a call.

## 8. The interview prompt (Grok Bot, Claude Code, Codex, Hermes)

Paste the block below as-is. Replace the bracketed line with your workflow.

```text
You are running the Agent Readiness Self-Check with me, one workflow at a time.
Workflow to score: [one job with a clear start and finish, e.g. "new inquiry to proposal draft"]

Rules you follow:
- You only ask questions and write the score sheet. Do not open accounts, connect tools,
  install anything, browse for my business, or contact anyone. If a step seems to need
  that, ask me instead.
- Do not ask for client names, customer records, contracts, passwords or keys. If I paste
  any, tell me and do not repeat them in the sheet.
- Go through these eight areas in order, one at a time: workflow mapped, data access,
  approval gates, owner per agent, handoffs, cost tracking, security and vetting,
  measurement.
- For each area ask at most three short questions, then propose a score from 0 to 3 using
  the anchors I give you below, and ask what evidence I can point to.
- Evidence rule: a 2 or 3 needs something I can open now (a page, list, setting, log or
  sheet). If I cannot name it, the score is 1. Say so plainly.
- After all eight, add the total, then apply the gate rules:
  gate areas are approval gates, owner per agent, security and vetting.
  One gate area at 0: cap at Draft-only pilot. Two or more at 0: Not yet.
  Ready to install one workflow needs every area at 1 or more.
  Ready to run more than one needs every area at 2 or more.
  Bands by total: 0-9 Not yet, 10-15 Draft-only pilot, 16-20 Ready to install one
  workflow, 21-24 Ready to run more than one. When rules disagree, the lower band wins.
- Then run a skeptical pass: rescore each area as a careful outsider would, list any area
  that drops, and say whether the band changes.
- Ask me to approve or change each score before you write the final sheet.

Output, once I approve:
1. The score sheet as a Markdown table: area, score, evidence, one fix.
2. Total, gate rule applied (if any), final band.
3. Hopeful and skeptical band, and the areas that differ.
4. The first three actions for my band, lowest area first.
5. One line: what I should have in hand before rescoring in two weeks.
Do not add praise, sales language or predictions about results.

Anchors (score what is true today):
[paste section 2 of this file here, or the score sheet in section 9]
```

### Harness notes

- **Grok Bot:** create a bot, paste the prompt into its instructions with section 2 as the anchors, and start a chat with the workflow name. The sheet comes back in chat; copy it into your doc.
- **Claude Code:** save the prompt as a project command or a skill (for example a SKILL.md named readiness-check) in an empty folder, with this file next to it so it can read the anchors. Ask it to write READINESS.md. Keep its permission prompts on, and decline any network or install request during the interview.
- **Codex:** run it in an empty folder with this file present and approvals on, so it asks before writing. Paste the prompt, or add it to that folder's AGENTS.md under a "Readiness check" heading.
- **Hermes:** add the prompt as a skill in your Hermes skills folder with section 2 as its reference file. Run it in a chat session, not as an unattended job: the whole point is that you answer and approve.

The prompt is plain text and does not depend on any one harness. If your agent skips the evidence rule or the gate rules, paste those two paragraphs again and ask it to rescore.

## 9. The score sheet

Copy this into a doc and fill it in.

```text
Workflow: ____________________   Date: __________   Scored by: __________

| # | Area                | Score 0-3 | Evidence (what you can open now) | One fix |
|---|---------------------|-----------|----------------------------------|---------|
| 1 | Workflow mapped     |           |                                  |         |
| 2 | Data access         |           |                                  |         |
| 3 | Approval gates (G)  |           |                                  |         |
| 4 | Owner per agent (G) |           |                                  |         |
| 5 | Handoffs            |           |                                  |         |
| 6 | Cost tracking       |           |                                  |         |
| 7 | Security, vetting(G)|           |                                  |         |
| 8 | Measurement         |           |                                  |         |

Total: ___ / 24     Band from total: ______________
Gate areas at 0: ___   Lowest area score: ___
Final band (lower wins): ______________
Hopeful band: __________   Skeptical band: __________   Areas that differ: ________
First action: _________________________________   Rescore on: __________
```

## 10. Worked example (illustrative)

This is an example, not a client. The business, workflow and numbers are invented to show how the scoring works.

**The business:** a five-person web design agency. **The workflow:** new inquiry to proposal draft.

**First score**

| Area | Score | Evidence or reason |
|---|---|---|
| Workflow mapped | 2 | One-page intake doc exists; two common exceptions are not on it |
| Data access | 2 | Inquiries arrive through one form and a shared inbox; past proposals sit in one folder |
| Approval gates (G) | 1 | Everyone "knows" the founder sends proposals; nothing written |
| Owner per agent (G) | 2 | Ops lead named, two hours a week set aside |
| Handoffs | 1 | Intake to proposal passes by direct message |
| Cost tracking | 0 | Three people on separate personal AI plans, nobody knows the total |
| Security and vetting (G) | 1 | Client briefs pasted into personal accounts, no approved tool list |
| Measurement | 1 | "Takes a few days" is the only number |

- **Total:** 2 + 2 + 1 + 2 + 1 + 0 + 1 + 1 = 10, which is Draft-only pilot by total.
- **Gate rules:** no gate area at 0, so no cap. Cost tracking is at 0, so the floor rule rules out the top two bands anyway.
- **Skeptical pass:** approval gates drop from 1 to 0 (nothing is written, so nothing can be shown). One gate area at 0 caps the band at Draft-only pilot. Same band, so the band holds, and approvals are the first fix.
- **Final band:** Draft-only pilot.

**First three actions:** write the approval list, starting with "no proposal or price leaves without the founder's yes"; move the team to one business AI plan and note the monthly cost; time the next ten inquiries from arrival to draft.

**Rescore two weeks later (still illustrative)**

| Area | Before | After | What changed |
|---|---|---|---|
| Workflow mapped | 2 | 2 | No change |
| Data access | 2 | 2 | No change |
| Approval gates (G) | 1 | 3 | Written list, named approver and backup, agent set to draft only |
| Owner per agent (G) | 2 | 2 | No change |
| Handoffs | 1 | 2 | Five-line handoff packet in the project board |
| Cost tracking | 0 | 2 | One business plan, monthly number, rough cost per proposal |
| Security and vetting (G) | 1 | 2 | Approved tool list, client briefs only in the business account |
| Measurement | 1 | 2 | Ten runs timed and counted |

New total: 17. Every area is at 1 or more, no gate area is at 0, so the band is Ready to install one workflow. In this example the agency now chooses between doing the install itself with the workbook or booking the Install sprint.

## 11. Where this leads

- **Not yet or Draft-only pilot:** the fixes in section 5 are free and most of them take an afternoon. Do them before spending anything.
- **Draft-only pilot or Ready to install one workflow, doing it yourself:** the [Agency AI Ops Workbook](https://paperchasewebb.ai/bots/marketplace/items/agency-ai-ops-workbook-install-sprint-self-serve/) ($39 in the PaperChaseWebb marketplace) is the Install sprint written as a self-serve workbook for one workflow.
- **Ready to install one workflow, done with you:** the Install sprint ($5,000) on the paperchasewebb.ai homepage. Book a call there and bring your score sheet.
- **Ready to run more than one:** Manage and Expand on the same homepage, for running and growing the operation after the first workflow works.

Related files in the marketplace: Ask-first rules for AI agents (approval gates), Cross-Harness Relay (handoffs across Grok Bot, Claude Code, Codex and Hermes), Agent Cost Levers (cost per solved task), Vet Before You Install (security and vetting), Governed Agent Teams (who may do what).

## 12. What we looked at, and what we did differently

Before writing this we went through a handful of free AI readiness assessments from agencies and consultants. The good ones share some ideas worth keeping: score one process instead of the whole company, ask for a named person instead of "the team", and let the weakest area limit the result. The common problems:

- **Generic:** questions that fit a dental office and a trucking firm equally, so the result fits neither.
- **Padded:** a precise-looking number with no arithmetic behind it.
- **Untested:** no way to check whether the score would change if your answers did.
- **Gated:** the actual report arrives only after an email and a call.
- **No real next step:** the same "book a call" under every band, including the ones where a call would be a waste of your money.
- **Missing the agent parts:** almost none ask about approval gates, who answers for each agent, handoffs between agents, cost limits or vetting what you install.

So this file scores one workflow, asks for evidence, gives gate areas the power to cap the band, includes a two-pass check against wishful scoring, tells some readers not to buy anything yet, and gives you all of it without a form.

## 13. Sources

Rebuilt in our own words. Nothing here is copied from the pages below; their claims are theirs and we did not verify them.

- Omkar Satpute (@openmausbot), on chat assistants, learning agents and orchestrators as layers of one stack, which shaped the point that orchestration and approvals are what buyers need checked: https://www.instagram.com/reel/DeUoYh4td0V/
- Prakash Kumar (@startup.snack), on tactics for a product's first 100 users, behind the idea of a complete free tool that leads to a paid next step: https://www.instagram.com/p/Det5q6bNM-N/
- Unlocking AI, on how to tell a diagnostic readiness assessment from a decorative one, behind the two-pass honesty test: https://unlockingai.com/articles/ai-readiness-assessment-lead-magnet-how-to-read-one
- Agent Readiness Score, for the idea that the weakest pillar sets the ceiling, which our gate and floor rules take further: https://agentreadyscore.ai/
- Parallel Labs, AI readiness scorecard, for asking for a named reviewer out loud: https://parallellabs.app/ai-readiness-scorecard/
- Atlas Software Partners, AI readiness assessment, for answering for one process, not the whole company: https://www.atlassoftwarepartners.com/guides/ai-readiness-assessment

Also drawn on: PaperChaseWebb's own internal method notes on readiness mapping, approval gates and handoffs that are only complete when the receiver accepts.

Made by PaperChaseWebb at paperchasewebb.ai.