AI Agent Path

Decide when an agent is justified, then implement bounded tool use, operate it, contain failures, repair it, and prove ownership.

Core purchase history is required. Agent access has its own independent 12-month term.

12 calendar months · English

Outcome before content

Finish with a least-authority operational agent, a learner-owned deployment, trajectory evaluation, incident containment, verified repair, and portable evidence.

Module 1

Agent action boundaries

Choose deterministic code or agent control · Bound model-directed actions · Reject unjustified autonomy

Action-boundary decision map + Deterministic fallback trace

Module 2

Tool contracts and structured execution

Define strict tool schemas · Validate tool outputs and side effects · Reject ambiguous execution states

Versioned tool contract + Malformed-output recovery trace

Module 3

Durable state and memory freshness

Separate working and durable state · Expire and revoke stale memory · Resume from a verified checkpoint

Durable state model + Freshness and revocation evidence

Module 4

Least authority and delegated identity

Scope delegated authority · Enforce tenant and actor boundaries · Record accountable approvals

Authority matrix + Cross-tenant denial evidence

Module 5

MCP interoperability and trust

Trace MCP protocol exchanges · Authenticate servers and capabilities · Constrain tool discovery and consent

MCP trust decision + Protocol failure diagnosis

Module 6

Retries and uncertain side effects

Classify retry safety · Use idempotency and reconciliation · Resolve unknown outcomes before replay

Idempotency design + Uncertain-side-effect recovery trace

Module 7

Human approval and escalation

Route consequential actions for approval · Preserve informed human control · Fail closed when approval is unavailable

Approval policy matrix + Escalation and timeout evidence

Module 8

Agent evaluation and evidence integrity

Evaluate complete action trajectories · Validate automated judges · Separate scoring defects from learner failure

Trajectory evaluation portfolio + Evaluator calibration report

Module 9

Agent operations and containment

Observe tool and policy decisions · Contain runaway or compromised execution · Restart from a safe durable state

Agent operations runbook + Containment and safe-restart receipt

Module 10

Agent production mission

Deploy a least-authority agent · Prove end-to-end operational controls · Repair an injected uncertain-side-effect incident

Learner-owned agent deployment receipt + Independent containment and repair evidence

Production validation is in progress

Public sales are disabled. No payment can be created until infrastructure, tax, legal, accessibility, and operational launch gates are independently approved.