INTERACTIVE PRODUCT CONCEPTExamples are simulated. Explore the scope.

Acceptance assurance for agentic work

AI can do the work.Decide what
gets accepted.

Your agents produce the work. AAC’s acceptance model brings the evidence, policy and human authority together before that work moves forward.

AACWORKSPACE / SOFTWARE DELIVERY

REL-042 / CANDIDATE 01

Repair authentication regression

Proposed by a coding agent · example workflow

Evidence & requirements

HOLD

The agent finished.
Acceptance is still outstanding.

Explore this decision
POLICY / release-policy-v1INTERACTIVE EXAMPLE · NO RELEASE

A product concept for decisions that matter.
No live agent or production connection in this preview.

Explore the architecture
Evidence before acceptance
Explicit human authority
Policy-bound decisions
Inspectable records

01 / Watch the work, not a promise

A workflow.
A decision that matters.

Follow the agents, inspect the evidence, and see the exact moment a human decision is required. Then take control of the example.

AACWORKFLOW STORIES / GUIDED SIMULATION

Interactive workflow story

Follow the work.
Examine the decision.

Enable JavaScript to play or control this illustrative workflow. The worked-example pages below remain readable.

A proposed output starts the review. It does not finish it.

01 / 06
00:00 / 00:36

Fictional workflow. Agent activity, evidence, checks and approval are simulated. No downstream action is executed.

Read this use case ↗
Read the story without motion
Watch the story. Change the outcome. Inspect the record.Open the detailed evidence workbench ↗Watch the captioned film ▷

What changes for your team

Less reconstruction.
More informed decisions.

The value to test is not another approval screen. It is a clearer handoff between proposed work and an accountable decision.

Build your evaluation case ↗
01

Know what is still missing.

Make the outstanding condition specific: a source, a test result, a scope check or an approval.

02

Review the work with its context.

Bring the candidate, policy, evidence and named authority into one inspectable handoff.

03

Keep the decision with the version.

Preserve why work was accepted without letting changed work inherit that acceptance.

02People. Policy. Evidence.

More automation.
Not less accountability.

An agent can prepare the work. The acceptance conditions should still say what evidence is needed—and which decisions remain with a person.

01

Define the conditions before the decision.

Make the requirements, permitted scope and next action explicit.

02

Bring the evidence into view.

Distinguish a passing check from everything needed for acceptance.

03

Keep authority where it belongs.

Reserve human approval when the workflow calls for it.

04The integration boundary

Put the boundary
where the work
becomes consequential.

Producing work, accepting it and executing the next action are separate responsibilities. Make the hand-off between them explicit.

Select a component to inspect its responsibility.

Intended architecture

Make acceptance explicit.

AAC’s proposed role is to relate a candidate to its requirements, evidence, checks and reserved approvals. The connected downstream system must enforce the decision.

05Trust you can examine

A decision should
come with its reasons.

What was required? What was checked? Who approved it? Which candidate does the decision apply to?

AAC’s intended model makes those questions part of the record—not a reconstruction after the fact.

AAC / Assurance record

A reason to move forward.

Example record · software-release scenario

Work itemREL-042
Candidateauth-patch-v1
Acceptance policyrelease-policy-v1
Required authorityPlatform lead · simulated
DecisionACCEPTED · example
Downstream effectNot executed
Inspectable example. Unsigned. No attestation.
Inspect a working decision record →

A clearer line for agentic work

Bring one workflow
that matters.

Start where an agent’s output becomes your organisation’s responsibility. Define what has to be true before it moves forward.

A focused technical conversation. An explicit next step.

From completed to accepted · a 40-second walkthrough

00:00 / 00:40
Read the full walkthrough transcript
  1. The agent finishes. A coding agent prepares a repair for an authentication regression. A completion claim starts the review; it does not finish it.
  2. One requirement is missing. The current example has 17 of 18 evidence requirements. The missing requirement covers unauthenticated requests to the protected endpoint. The decision is HOLD.
  3. Evidence must be checked. The example submits the missing evidence, binds it to the current candidate and runs the fixture checks. Arrival is not acceptance.
  4. Authority remains with a person. The checks pass. The policy still reserves approval to the Platform lead. The example remains at a HUMAN GATE.
  5. The decision becomes inspectable. A simulated approval is recorded. This candidate is ACCEPTED against this policy. The record contains the evidence, scope and reason. No release executes.
  6. Changed work needs a new decision. Changing the candidate returns the example to HOLD. The historical acceptance remains recorded, but cannot be reused for the new candidate.

Animated explanation with fictional fixtures. No live agents, real evidence, authenticated approval or release integration.

Share the AAC site

A complete, clickable review

Send the site, not a screenshot.

Example detail