AgentProof Learn & Authority Centre
Agentic AI Library
A living, source-backed authority on what good agentic AI workloads look like — and how to prove yours is ready before real users see it.
What you can learn here
Six practical tracks. Each track explains why it matters, what good looks like, what can go wrong, and what AgentProof checks for you in the readiness assessment.
Capability Zones
Place your agent so you choose the right controls.
- Why it matters
- Controls and evidence expectations differ by capability zone — applying chat-style guardrails to an action-taking agent is how serious incidents start.
- What good looks like
- Every agent is consciously placed in Informational, Assisted Work, or Action-taking — with the placement written down.
- What can go wrong
- An informational chat quietly becomes an action-taker over time, but still operates under chat-style oversight.
- What AgentProof checks
- The agent's classification, the actions it can technically reach, and whether the controls match the zone.
Good Agent Design
10 design principles that separate a credible agent from a brittle demo.
- Why it matters
- Most agent incidents trace back to a small set of design choices made (or skipped) before go-live.
- What good looks like
- Scope is written, sources are cited, refusal paths are tested, evidence is captured by default.
- What can go wrong
- A demo gets promoted, the design assumptions never get written down, and 'why did it do that?' becomes a regular Monday.
- What AgentProof checks
- Whether each principle is honoured for the specific agent being assessed.
Controls & Oversight
Six control families with maturity ladders from Basic to Evidence-ready.
- Why it matters
- Mature controls earn the right to operate higher-zone agents without operational drama.
- What good looks like
- Each control family has an owner, a written expectation, and a way to show it was honoured.
- What can go wrong
- Controls exist as PowerPoint, not as runtime behaviour — auditable only when challenged.
- What AgentProof checks
- Coverage of the six families, the maturity level reached, and whether evidence is captured.
AI Landscape Radar
Watch the market without being whiplashed by every press release.
- Why it matters
- Agentic AI guidance is changing every quarter — your readiness has to be updatable, not frozen.
- What good looks like
- A documented source watchlist with a human-approved release cadence and report-impact flagging.
- What can go wrong
- Guidance changes silently, recommendations drift, and old reports become quietly wrong.
- What AgentProof checks
- Source coverage, pack versioning, human-review status, and which reports are affected by each pack change.
Reference Library
Curated, source-backed guidance by topic and capability zone.
- Why it matters
- Buyers want to read the framework before they trust the report — give them a credible library to read.
- What good looks like
- Each entry shows the source family, the last-reviewed date, and human review status.
- What can go wrong
- A 'library' that's actually a marketing brochure with no source attribution.
- What AgentProof checks
- Source coverage and freshness — and binds report recommendations back to the entry they came from.
Report Guidance
How AgentProof turns your assessment into a report you can stand behind.
- Why it matters
- A report is only useful if you can explain what's in it, what's not, and where it came from.
- What good looks like
- Every finding is tied to a control family, an evidence expectation, and a next action.
- What can go wrong
- A score that nobody can defend in a meeting because it lacks provenance.
- What AgentProof checks
- Source coverage at report time, pack version binding, and credibility band derivation.
A training ground, not a brochure
Each track in the Library reads like a coaching session — what good looks like, the failure pattern it prevents, a practical example, and the question AgentProof will ask your team during the assessment.
What good looks like
Concrete patterns you can mirror, not vague principles.
Failure prevented
Each principle is paired with the failure pattern it prevents.
Evidence to keep
What to capture so the next review is faster and cheaper.
The Agentic AI readiness journey
Seven ordered stages. AgentProof is built around this journey — Discover, Classify, Assess, Control, Evidence, Improve, Reassess.
- 1
Discover
Surface the agents that already exist in your environment.
- 2
Classify
Place each agent in a capability zone — informational, assisted, action-taking.
- 3
Assess
Walk a source-backed readiness review for the agent in front of you.
- 4
Control
Pick the right controls for that zone — not generic AI policy.
- 5
Evidence
Capture audit-safe proof of decisions and actions.
- 6
Improve
Close the highest-impact gap first, then the next.
- 7
Reassess
Re-score when the agent, the data, or the AI landscape changes.
Public preview vs full workspace
Public preview — free
Useful knowledge before you sign in
- Capability Zones with examples and risk profile
- Good agent design — 10 principles with failure patterns
- Control families with maturity ladder
- Living AI landscape radar
- Reference library index with last-reviewed dates
Inside the free trial workspace
Deeper, report-linked guidance
- Full per-zone control checklists
- Report-linked deep dives on each finding
- Improvement-cycle suggestions wired in
- Stale-source warnings on affected reports
- Reassessment trigger when pack version changes
No payment to start the trial. No card required. AgentProof is a readiness assessment, not an official audit.
Why you can trust this Library
Source-backed
Each recommendation cites the intelligence pack version it came from.
Human-reviewed
No guidance is published to the product without a named human reviewer signing it off.
Honest scope
AgentProof is a readiness assessment. It is not legal advice and it is not an official audit. It does not speak on behalf of any vendor.
Ready to assess your own agent?
Start the free assessment to apply this guidance to a real agent. No payment. No public registration required.