AI Evidence Plane
Discuss one workflow

Enterprise control for agentic AI

When AI acts,can you provewhy?

AI Evidence Plane puts controlled execution around probabilistic agents—connecting every decision, approval and action to evidence.

Explore how it works Discuss one workflow
Agent workflow graphic showing a request, proposal, policy gate, evaluation, human approval, commit proof and blocked remediation branch.
    1. Request
    2. Agent proposes
    3. Policy + evaluation
    4. Human approval
    5. Commit
    6. Proof
    BlockedRemediate

    Programme selection

    Selected for the National Digital Track

    AI Evidence Plane has been selected for the National Digital Track of the Buy Australian AI Partnership, delivered by Stone & Chalk, with the National AI Centre as Principal Sponsor.

    Meet the team

    Why this matters now

    AI adoption is accelerating. Operational confidence is not.

    1. Adoption≈90%

      use AI

      McKinsey State of AI 2026
    2. Value gap37%

      report positive EBIT impact

      McKinsey State of AI 2026
    3. Project risk40%+

      of agentic projects may be cancelled

      Gartner June 2025 forecast

    Adoption is scaling faster than enterprise control, proof and measurable value.

    Core insight

    Keep reasoning probabilistic. Make execution controlled.

    The agent remains probabilistic. Its permissions, transitions and commit conditions are controlled.

    Probabilistic reasoningAgent proposes
    Versioned execution contract
    ReadPrepareEvaluateApproveCommit

    Policy · Identity · Limits

    FailBlock · Remediate · Re-evaluate
    Authorised outcomeCommit with proof

    Read → Prepare → Evaluate → Approve → Commit

    The product wedge

    Design. Enforce. Prove.

    01

    Design

    Graphical workbench for build, test, approve and promote.

    Workflow workbench showing connected configuration objects and execution states.
    AgentModelToolPolicyEvaluationApprovalRemediationEvidence requirement

    Output: an approved, versioned execution contract

    02

    Enforce

    The approved execution contract controls every consequential action before commit.

    Governed agent execution path with policy, evaluation, approval and proof nodes.
    Identity resolvedTool permittedPolicy passedEvaluation passedHuman approvalCommit authorisedCommit deniedGoverned remediation

    Output: a governed action path

    03

    Prove

    Evidence Graph connects approved design, observed execution and proven outcome.

    Connected evidence graph with approved and blocked execution paths.
    Approved designObserved executionContractModelToolsPolicyApprovalsEvidence record

    Output: a connected, verifiable execution record

    Pilot success measures

    What changes for the enterprise.

    Controlled executionFaster approval to productionFewer unauthorised actionsShorter investigation and recoveryAudit-ready evidenceCost per successful outcome

    AI Evidence Plane overview

    An independent assurance plane between approved AI design and real-world execution.

    This answers where it runs: inside the enterprise boundary, with optional cloud control by metadata reference.

    Business workflowsConsequential decisionsEnterprise approvals
    AI Evidence Plane
    Execution contractIdentity + authorityPolicyEvaluationApprovalRemediationEvidence
    ModelsAgentsAI gatewaysMCP toolsSystems
    Customer boundaryLocal evidence storeOptional cloud control: metadata reference only

    Why it is different

    Not another gateway. Not another observability dashboard.

    GatewayRoute
    ObservabilityInspect
    AI Evidence PlaneEnforce · Prove

    Complements gateways and observability; governs the workflow and its commit authority.

    Design-partner pilot

    Start with one consequential workflow.

    Map its execution contract. Govern one live path. Produce the evidence required to scale.

    • Approved execution contract
    • Governed pilot run
    • Evidence and scale decision
    Discuss a design-partner pilot