Agent readiness assessment. Free, scored, in your browser.
Agent readiness is how well a business can be found, read and safely acted on by AI agents. It means one current record of your facts, a person approving any step that could create liability, access limited to each customer's own data, a log of every action, and reviewers who really review.
Nothing you enter leaves your browser. No AI model and no network call: fixed rules score your answers on your device. Nothing is sent unless you choose to send your result.
This is a self-assessment scored by a fixed set of rules, not an audit and not legal advice. It reflects only the answers you give. A high score does not guarantee how any agent will behave or protect you from liability. Rules last reviewed 27 October 2026. Origin Pi accepts no liability for decisions made using this tool.
JavaScript is off, so the score cannot be worked out. The questions and every scoring rule are listed on this page.
Origin Pi. originpi.com/tools/agent-readiness-assessment/
Your agent readiness result
Send me this result only fills in our contact form with your scores. Nothing is sent until you submit it.
A rules engine, not a chatbot.
The assessment applies 20 checks in four dimensions, taken from Origin Pi's published agent-readiness audit and the definition of agent-ready. Rules last reviewed 27 October 2026.
Agent readiness framework: the four dimensions and the IMDA framework
Singapore's Model AI Governance Framework for Agentic AI was launched by IMDA on 22 January 2026. Version 1.5 was published on 20 May 2026 and updated on 5 June 2026. It is voluntary guidance, not law. Its four dimensions are: assess and bound the risks upfront; make humans meaningfully accountable; implement technical controls and processes; enable end-user responsibility. Our consent and scoped-access checks are also informed by the Model Context Protocol specification. Our reading: it lists user consent before tool invocation among its key principles, and its Tools page recommends (SHOULD) a human in the loop. Read the specification for the exact wording.
| Dimension | Weight | What it asks | IMDA framework |
|---|---|---|---|
| Data | 25 of 100 | One current, complete record of what you sell and promise. | Assess and bound the risks upfront |
| Governance | 30 of 100 | Limits on what an agent may do, and proof of what it did. | Make humans meaningfully accountable; implement technical controls and processes |
| Discoverability | 20 of 100 | Whether a machine can read clear, current answers about you. | Not an IMDA dimension: Origin Pi's addition for agents outside your business |
| Operational readiness | 25 of 100 | Whether the humans in the loop are real and accountable. | Enable end-user responsibility |
Scoring rules
- Answers score Yes 2, Partly 1, No 0, Don't know 0. Each check has a weight from 1 to 3.
- A dimension score is the weighted points as a percentage of the maximum, rounded to a whole number. The overall score is the weighted mean of the four dimension scores.
- Bands: Ready 80 or more, Nearly ready 60 to 79, Early stage 40 to 59, Not yet ready below 40.
- A must-pass check answered No or Don't know holds its dimension, and the overall result, at Early stage at best.
- A check marked No agent yet (only the override-rate check offers it) is left out of the score, not counted as a failure.
- A dimension is weak if it scores below 70 or has a must-pass gap. Each weak dimension gets its three highest-priority gaps as next actions, must-pass gaps first, then by weight.
Every check
| Check | Dimension | Weight | Must-pass | Why it matters |
|---|---|---|---|---|
| D1 Is each key fact (products or services, prices, availability, policies) held in one system of record, with identical values on every surface (website, quotes, chat, marketplaces)? | Data | 3 | Yes | An agent quotes whichever copy it reaches, so two copies mean two answers. |
| D2 When a price, stock level or policy changes in your system, does your public site show the change in under a minute? | Data | 3 | Yes | Stale data sells the wrong price to every machine reader at once. Under a minute is full marks; a day or more fails. This threshold is Origin Pi's own suggestion, not an IMDA requirement. |
| D3 Does your machine-readable record carry price, tax treatment, stock and the version of the policy in force, not only a name and address? | Data | 2 | No | Missing fields become guessed fields, and a guessing agent is a liability. |
| D4 Can you show which version of your terms, return policy or price list applied on a given past date? | Data | 2 | No | If you cannot reconstruct what was promised, you cannot explain or contest an agent-assisted decision. |
| D5 Have you decided which data an agent may read (for example the public catalogue) and which it must never reach (customer records, staff data, contracts)? | Data | 2 | No | Bounding what an agent can touch is the first step in the IMDA framework. |
| G1 Does every material action that could create liability (a refund above your set limit, an order, a commitment, a message that binds you) pause for a named person's approval before it goes out? | Governance | 3 | Yes | An agent committing your business at machine speed with no pause may create risk your insurance policy does not cover. The IMDA framework asks for approval checkpoints for significant actions. |
| G2 Is each agent or customer token limited to that customer's own data, and have you tested this with a second customer's token? | Governance | 3 | Yes | One broad key lets one agent read across accounts it should never see. |
| G3 Is every agent action logged with agent ID, request ID, timestamp and result, in a log staff cannot edit, kept for at least 90 days? | Governance | 3 | Yes | You cannot contest or explain a decision you cannot reconstruct. This threshold is Origin Pi's own suggestion, not an IMDA requirement. |
| G4 Does each agent have a written list of what it may do and which tools and data it may use, with everything else denied by default? | Governance | 2 | No | Least privilege limits the damage when an agent is wrong or manipulated. |
| G5 Can you switch an agent off, or roll back a change to its tools or instructions, within minutes, and is each such change reviewed before release? | Governance | 1 | No | IMDA Version 1.5 recommends change management so small changes do not have outsized effects. |
| V1 Are the answers to "what do you sell, for how much, where and when" present in the HTML your server sends, without JavaScript needing to run? | Discoverability | 3 | Yes | Machine readers need the answer present and current in the served response. |
| V2 Do your key pages carry structured data (schema.org JSON-LD) with zero errors in a validator, matching what a visitor sees on the page? | Discoverability | 2 | No | Markup that describes content not visible on the page breaks trust. It is hygiene for correctness, not a ranking trick. |
| V3 Do you publish what automated agents may do on your site (allowed actions, rate limits, who to contact), kept separate from facts that change such as price and stock? | Discoverability | 1 | No | A hand-written file goes stale. Keep volatile facts out of it. |
| V4 Are your business name, address, hours and contact details identical on your website and every directory or profile that lists you? | Discoverability | 2 | No | Conflicting identity facts make an agent unsure which one to trust. |
| V5 For anything an agent should be able to do for a customer (book, quote, order), is there a documented, permissioned route such as an API or an MCP server, rather than only a form meant for humans? | Discoverability | 2 | No | Being agent-ready is not having an API; it is a permissioned route with a confirm step. The route still has to exist. |
| O1 Over the last 30 days, did reviewers change or reject at least one in ten of the agent's proposals? | Operational readiness | 3 | No | A change or reject rate of zero can mean approval has become a rubber stamp, but a low rate can also mean the agent is good. IMDA Version 1.5 adds monitoring human override rates against automation bias. One in ten is Origin Pi's own rough proxy, not an IMDA requirement; the stronger test is to plant a few known errors and see whether reviewers catch them. |
| O2 Can every change of state (an order, a refund, a record edit) be traced to an agent ID, a session and a named human reviewer? | Operational readiness | 3 | Yes | Accountability that survives after the fact needs a name on the record. |
| O3 For lower-risk actions that need no approval under the check above, are unusual ones (a price drop over a set percentage, a first-time customer) still routed to a person, rather than everything or nothing being reviewed? | Operational readiness | 2 | No | Targeted review of the remainder keeps attention where it matters and prevents alert fatigue. G1 covers the material actions; this check covers the routing of the rest. |
| O4 Does each agent have a named owner, and have its reviewers been told what it can do, what data it touches and what their own duties are? | Operational readiness | 2 | No | Users and reviewers who do not know an agent's limits cannot supervise it. |
| O5 Is there a written plan for when an agent gets something wrong, covering who tells the customer, who fixes the record and who handles any personal-data incident? | Operational readiness | 2 | No | Speed and clarity in the first hours decide how much damage an error does. |
The one-minute, 90-day and one-in-ten thresholds are Origin Pi's own suggestions, not requirements of the IMDA framework or of any law.
This is a self-assessment scored by a fixed set of rules, not an audit and not legal advice. It reflects only the answers you give. A high score does not guarantee how any agent will behave or protect you from liability. Rules last reviewed 27 October 2026. Origin Pi accepts no liability for decisions made using this tool.
Sources
Frequently asked questions
What is agent readiness?
Agent readiness is how well a business can be found, read and safely acted on by AI agents. It means one current record of your facts, a person approving any step that could create liability, access limited to each customer's own data, a log of every action, and reviewers who really review.
How is the agent readiness score calculated?
Each of the 20 questions is answered Yes (2 points), Partly (1) or No or Don't know (0) and has a weight from 1 to 3. A dimension score is the weighted points as a percentage of the maximum, rounded to a whole number. The overall score is a weighted mean of the four dimensions: Data 25, Governance 30, Discoverability 20 and Operational readiness 25. 80 or more is Ready, 60 to 79 Nearly ready, 40 to 59 Early stage and below 40 Not yet ready. Seven checks are must-pass: answering No or Don't know on one marks its dimension as not ready and holds the overall band at Early stage, whatever the total. The one-minute update time (D2), the 90-day log retention (G3) and the one-in-ten override rate (O1) are Origin Pi's own suggestions, not requirements of the IMDA framework or of any law; only D2 and G3 are must-pass.
Does anything I enter leave my browser?
No. The score is worked out by JavaScript on this page from rules that load with the page. There is no AI model and no network call. If you press Send me this result, the existing contact form opens with a short summary filled in; nothing is sent unless you submit that form.
Is this based on the IMDA framework, and is that framework law?
Partly. The Data, Governance and Operational readiness dimensions follow Singapore's Model AI Governance Framework for Agentic AI, which IMDA launched on 22 January 2026 (Version 1.5 was published on 20 May 2026 and updated on 5 June 2026): assess and bound the risks upfront, make humans meaningfully accountable, implement technical controls and processes, and enable end-user responsibility. Discoverability is Origin Pi's addition for agents outside your business, which is not one of IMDA's four dimensions. The framework is voluntary guidance, not law.
Should I answer for my own agents or my customers' agents?
Run it twice. First for the agents you deploy inside the business. Then again for the customer agents that will read your public record and act on it without asking you first. If you run no agent yet, answer for the process you would rely on today; a No simply becomes a to-do.
Does a high score protect me legally?
No, and no honest checklist should claim it does. A confirm step and an audit trail give you a defensible accountability record, but they do not shield you from every outcome when a customer's own agent acts on your data. Treat the score as a map of operational gaps, and take legal advice for decisions with legal effect.
What should I do after I get my result?
Fix the must-pass gaps first, then work through the three next actions listed for each weak dimension. Re-run the assessment after the changes. If you want help, use the Send me this result button and a person at Origin Pi will reply.
Want a person to read your result?
We turn a readiness score into a plan: the record, the confirm step, the access controls and the audit trail behind them. See how we approach agent readiness.
Talk to us