Demo
Try AgentProof on sample agents
Walk through a sample readiness review against two demo agents. Demo data lives in an isolated demo workspace and never appears in a real customer's report history.
- Step 1. Pick a sample agent below.
- Step 2. Answer the short readiness questions.
- Step 3. Generate the sample readiness report and review what AgentProof would surface for a real agent.
Microsoft-style sample
Demo Sales Support Copilot
A Microsoft Copilot Studio-style sample agent that helps a sales team draft replies and look up product information.
Demo / sample data — isolated from any real workspace.
Non-Microsoft sample
Demo Website Support Chatbot
A website-chatbot sample that answers product questions and escalates to a human when it cannot answer.
Demo / sample data — isolated from any real workspace.
Demo · value story (R35)
Demo is a complete value story — sample agent → readiness → score → improvement → radar → trial/workspace.
Chapter 1. A sample agent
Meet 'Sample Power Platform Sales Support Agent' — a non-confidential illustration of a real agent shape.
We show a realistic agent profile (capability zone, scope, data categories, write capabilities). No real customer agent is created during the demo.
- Agent name: Sample Power Platform Sales Support Agent
- Capability zone: Assisted Work
- Provider: Microsoft Copilot Studio (sample, not connected)
- Scope: Sales answers, not action-taking
Chapter 2. Why readiness matters
Even a small agent can mislead a sales conversation or echo stale policy. Readiness asks the right questions before deployment.
AgentProof's readiness check is a structured review covering scope, safety, evidence, oversight, recovery, and value. The output is a score + a list of specific improvements.
- Capability zone classification + risk profile.
- Six control families with expected evidence.
- Honest 'not enough evidence yet' signals when applicable.
Chapter 3. The assessment
A guided question set tailored to the sample agent — answers shape the readiness score.
Questions are written for an admin who understands the agent's deployment. Each answer carries an importance weight that feeds the scoring model.
- Sample questions covering scope, oversight, evidence.
- Three classification answers shape the question path.
Chapter 4. Score + report
The sample agent scores in the medium-readiness band, with a per-finding report that traces every score contributor.
The report carries four version stamps (methodology, scoring model, intelligence pack, report version) + capability zone + control family findings + evidence expectations.
- Sample score: 68 / medium.
- Per-finding cards with why-it-matters + what-good-looks-like.
- Visible export action.
Chapter 5. Improvement plan
Each finding has a clear improvement action with evidence-to-collect + reassess-after guidance.
Improvement cards aren't a checklist — they're an honest path from the current score band to the next.
- Sample improvement cards for top three findings.
- Per-card reassess-after guidance.
Chapter 6. AI Radar relevance
AgentProof tracks public guidance + market updates against an approved source registry; affected reports surface a newer-intelligence warning.
Approved-source-only. No broad crawling, no probing, no paywall bypass. Old reports remain immutable; new packs trigger reassessment warnings.
- Six monitored source categories.
- Sample candidate signals — clearly labelled.
- Sample adaptation proposals — never operational.
Chapter 7. Next step
Open Trial for a guided, persisted journey — or Workspace if you already have a signed-in pilot identity.
Trial creates a sample-only persisted workspace so you can come back. Workspace is for invited pilot customers signed in with a real identity.
Open Trial →
Demo is sample-only. No customer data is persisted, no agent is created, no Supabase write happens. To create a real agent and persist a real readiness journey, open Trial or Workspace.
See the methodology behind the demo
Want to read the framework first?
Public preview is free. The full library is available inside the trial workspace.