Is this ready for an agent?

Answer ten questions about one workflow and get the specific boundaries to fix.

Runs on worked examples — how this works

One workflow. Ten questions.No model calls · no account needed

1. Can you describe a good outcome?

An agent needs a finish line, not just a job title.

Think: a correctly classified support request, not ‘improve customer experience’.

The goal is still broadWe can describe it, but not check it consistentlyWe have a specific outcome and a way to check itNot sure yet

Question 1 of 10

How the assessment works

This original educational framework checks the task, evidence, authority and operating conditions. It does not assess your organisation or inspect your systems.

Your answers produce a preparation score and specific next steps. Missing data permission, action boundaries, recovery or ownership cannot be cancelled out by a good total.

Read the full framework

Published rules and local scoring

Every result comes from published rules running in your browser. There is no model connection, login for the lab, or background saving. Optional sharing puts only the ten selections into the link.

Rubric 1 · educational self-assessment. Not a certification, NIST assessment, legal advice or approval to automate high-impact decisions.

Apply the method to invoice triage

Follow three fictional invoices from evidence to the next responsible step.

Fictional cases · authored rulesNo live systems or model calls

A clean match

Fictional invoice INV-104 · Birch Office Supplies · £480

Available evidence

Purchase order PO-28 matches the supplier and amount.
The goods receipt is recorded.
The supplier details match the approved vendor record.

The boundary

This exercise permits the agent to recommend a review queue. It may not release payment, create suppliers or change bank details.

What should happen next?

Choose a recommendation. Nothing will be sent, updated or paid.

How worked examples and data handling work

How this was built

Approach: ten typed questions, a deterministic scoring function, native radio controls and a versioned answer link. The same inputs always produce the same result.

Build instruction: “Build it over this.” The existing Next.js site and visual system were extended in place with AI-assisted coding and human review of the preview.

Runtime model: none. Inference cost per assessment: £0. Hosting is separate. Build time and assistant-token spend were not measured, so no estimate is presented as fact.

Limit: self-reported answers are not verified. The questions and thresholds are an editorial starting point, not an independently validated instrument.

Build diary awaiting Arun’s notes. Empty fields below are intentional; no build story or time estimate has been invented.

Tools used: OpenAI Codex, Next.js 16, TypeScript, Impeccable
Models used to build: Arun to add
Rough hours: Not recorded
What failed: Arun to add

Actual build prompts

Not as v2, build it over this

Here you go, build it