Agent Certified
Tier specification ยท Methodology v2.0

Five tiers, each one a floor.

Every assessed agent is placed in one of five tiers by its weighted score across the seven dimensions. Each tier has a meaning, a minimum on every dimension and a period of validity. This page is the tier specification itself.

0 to 19

Pre Assessment

A baseline and a gap report. Valid three months.

20 to 34

In Progress

At least 2 on every dimension. An active candidate. Six months.

35 to 54

Certified

At least 4 on every dimension. The operating floor. Twelve months.

55 to 74

Advanced

At least 6 on every dimension. Ready for higher exposure. Twelve months.

75 to 100

Elite

At least 8 on every dimension. The top of the scale. Twelve months.

The score is a number between zero and one hundred. The certification is the tier that number falls into, together with the evidence file and the expiry date. A total that reaches a tier's band without meeting its floor on every dimension is placed in the tier below.

The number is not the tier.

The worked example scores one constructed deployment dimension by dimension and shows why a weighted total of 58 still stops at Certified. No real organisation is scored there, and none has been assessed under this methodology to date.

Elite 75 to 100.

The agent operates at the top of the framework. Every dimension sits well above the operating floor, including the two heaviest. No agent has been assessed at this tier, and it is described here as the methodology defines it.

What it requires

  • A raw score of at least 8 on every dimension, without exception.
  • Independent assurance already performed, or scheduled within the current year.
  • A board level review of the agent within the last six months.
  • An incident history that shows at least one real event contained.
  • An Autonomy Envelope tied to insurance policy wording.

What a counterparty learns

This is the tier at which an insurer could price against a known, repeatable risk posture. Agent Certified has no arrangement with any insurer or supervisory authority, and no Elite result has been adopted as an underwriting input.

Validity

Twelve months. Renewal is a full re assessment and must include at least one new piece of incident or drill evidence.

Advanced 55 to 74.

Materially above the operating floor and suited to higher exposure, including use inside regulated workflows where the audit trail and human oversight cannot be negotiated.

What it requires

  • A raw score of at least 6 on every dimension.
  • A governance owner named and referred to in board minutes.
  • A regression evaluation suite that runs on every change.
  • An Autonomy Envelope enforced in code as well as in policy.

What a counterparty learns

A credible risk posture with no single weak dimension hidden behind a strong total. Whether an insurer accepts it, and on what terms, is for the insurer to decide; where the deployment reaches beyond internal tooling, sector specific terms can be expected.

Validity

Twelve months. Renewal may be partial, a desk review of updated evidence, when the agent has not materially changed.

Certified 35 to 54.

The minimum floor for certification. Governance, oversight and technical controls exist, are documented and have been inspected. Counterparties may rely on the mark for standard commercial use. It is the entry point for the Agent Certified mark.

What it requires

  • A raw score of at least 4 on every dimension.
  • Documented guardrails and a verified kill switch.
  • A named accountable owner.
  • An incident playbook in writing.
  • An autonomy policy published inside the organisation.

What a counterparty learns

The operator has crossed the threshold below which reliance on the agent is hard to defend.

Validity

Twelve months, with a light annual review when nothing material has changed. Any change to the Autonomy Envelope triggers re assessment, whatever the date.

In Progress 20 to 34.

Formal controls have begun, and the evidence is not yet enough for a certification call. The agent is an active candidate with a known path to Certified. The mark is not awarded.

What it requires

  • A raw score of at least 2 on every dimension.
  • A written remediation plan with target dates.
  • A named owner, even in an interim role.
  • The board's acknowledgement that the agent is in production.

What a counterparty learns

The operator is engaged with the framework and moving toward certification. A reasonable position for pilots and internal tooling; too early for client facing or regulated work.

Validity

Six months, deliberately short. An agent that stays at In Progress for twelve months without movement loses the tier.

Pre Assessment 0 to 19.

A baseline. The operator has engaged with the framework, and the governance, oversight and technical evidence cannot yet support a certification call in either direction.

Conditions

  • An evidence file opened with the assessment desk.
  • An agent in production or in pilot, and nothing more technical than that.
  • A written gap report against the framework for the operator.

What a counterparty learns

A starting position. The operator has entered the process, and for the purpose of reliance the agent should be read as uncertified.

Validity

Three months. It lapses unless the operator reaches In Progress or above.

Side by side, for a committee.

The five tiers in one table, for risk committees, procurement teams and insurer intake forms.

TierScoreMinimum on every dimensionWhat it signalsValidity
Elite75 to 1008 of 10Reference profile. Every dimension well above the floor.12 months, full re assessment
Advanced55 to 746 of 10Regulated workflows and higher exposure.12 months, desk review possible
Certified35 to 544 of 10Standard commercial use. Operating floor met.12 months, annual review
In Progress20 to 342 of 10Candidate. Pilots and internal tooling.6 months
Pre Assessment0 to 19NoneEntry point. Read as uncertified.3 months

Questions the specification answers.

Each answer stands on its own, so it can be read, quoted or checked without the rest of the page.

How is a composite agent certified?

As one unit of reliance, with one score. A composite is built from several sub agents, tools or models, and the unit of assessment is the thing a counterparty relies on, whatever it is built from.

That brings three further evidence requirements. Distribution Control is scored against the widest authority any component holds, never the average, because a mistake or an attacker reaches the whole through the weakest boundary. The Autonomy Envelope is scored against the longest unsupervised chain the composite can run end to end, including any step where one sub agent invokes another. AI Integration is scored on whether handoffs between components keep the identity, authority and audit trail of the original request.

A composite whose components are each well governed but whose handoffs are not will score below the sum of its parts. That is intended.

What should be in an eval suite before an agent goes live?

Five things. A suite missing any of them does not score under Product Maturity.

A held out task set drawn from real production inputs, kept apart from the examples used to build the agent, with a pass threshold agreed before the run. Adversarial cases, including prompt injection against every external data path the agent reads; prompt injection is ranked LLM01 in the OWASP Top 10 for Large Language Model Applications. Regression coverage, so the same suite runs on every model or prompt change and results compare over time. Failure mode cases that assert what the agent does when a tool is down, a response is malformed or a request falls outside scope. And a recorded run history with dates, versions and results.

A suite that exists without a history proves only that someone wrote it.

Which level does an agent actually need?

The level follows the exposure. An agent on internal tooling, with a person approving anything that leaves the organisation, is served by In Progress or Certified.

An agent that acts on customers, moves money or produces output a third party relies on should hold Certified as a floor, and Advanced where the workflow is regulated. Advanced sets a minimum of six out of ten on every dimension, which rules out a strong governance score covering a weak oversight design. Elite requires eight out of ten on all seven: a description of an unusually mature operation, and the top of the scale rather than a procurement requirement.

Before commissioning an assessment, ask which tier the counterparty asking for evidence will actually read.

How a tier is arrived at, step by step     The complete framework guide     Performance and reliability

Next step

See where your agent lands.

The first output of an assessment is a score and a tier, with a written evidence file. The cycle runs four to six weeks from intake to the report.

Request an assessment Read the methodology