Stage 1: An identity an agent can prove
Replace the shared secret with a registry of agent identities, each with its own key pair, fingerprint and revocation state.
£299 programme
Mai Studies Original · Computing
Build an AI Agent Identity and Permissions Platform
An extended build-along programme. Software agents now act on their own, yet most of them run on a shared secret key with no record of who asked for what, no limit on what they may spend and no way to prove afterwards what was decided. You build the missing control layer yourself, one stage at a time, with coding challenges marked instantly in the page and a stage deliverable marked against a published rubric. Every stage adds a working piece: a registry of provable agent identities, scoped capability grants, delegation chains that only ever narrow, spend and rate budgets that refuse politely, a tamper-evident log of every decision, and a final runtime gateway that puts all of it in front of a live request. This is a Mai Studies record of study, not a regulated qualification.
12 weeks 90 guided hours Advanced
No account needed to buy. Pay with your card and we create your student account from the email you enter, with this programme ready inside it for twelve months.
What you will achieve
- Give every agent an identity that can be proved and revoked
- Express permissions as narrow, time-limited capability grants
- Build delegation chains that can only narrow, never widen
- Hold agents to spend and rate budgets that refuse safely
- Keep a tamper-evident record of every decision taken
- Ship a runtime gateway that allows or refuses a live agent request
Entry guidance
- — Comfortable writing JavaScript functions
- — Willing to build and test your own code each stage
Original Mai Studies learning material. This is a professional completion programme, not a regulated qualification or university accreditation.
Programme syllabus
6 applied modules
15h
Stage 2: Scoped capability grants
Express permission as a narrow grant: one action, one resource, one expiry, instead of a blanket key that can do anything.
Stage 3: Delegation that only narrows
Let an agent act on behalf of a person or another agent, with a chain that can attenuate rights but never widen them.
Stage 4: Spend and rate budgets
Hold every agent to a budget of money and calls, refuse politely at the limit and make the refusal auditable.
Stage 5: A tamper-evident decision log
Chain every allow and refuse decision by hash so a later edit is detectable by anyone holding the log.
Stage 6 final build: the runtime policy gateway
Put the whole control layer in front of a live request and return a clear allow or refuse with the reason and the evidence.
What students say
No ratings yet.Sign in as a student on this programme to leave a rating.