Governing AI Agents: Gateways, Registries and Controls
Put every AI agent in your organisation behind controls that code enforces. Apps ask for a capability instead of a model, every agent is registered with its own identity, and one model gateway and one tool gateway check, limit and record every call.
Read every lesson as a guest, or sign in to enroll and track your progress.
Enroll to save the lessons that you complete. It is free.
You are enrolled. From now on, the lessons that you complete are saved.
To get your certificate, mark every knowledge check as matching. These lessons have checks to finish:
Is this the right course for you?What this course covers and does not cover, and 5 courses to take for other needs.
This course teaches the technical controls that apply to every agent in an organisation: a capability catalog, registration and identity, a model gateway and a tool gateway, budgets, policy as code, kill switches and usage evidence. It does not teach how to build one agent, the MCP protocol itself or the laws and policies of AI governance.
Explain which governance rules can be enforced in code at the platform, and where each control point sits (identity, registry, model gateway, tool gateway, audit).
Design a capability catalog, map capabilities to models and providers with data rules and fallbacks, and change a mapping safely.
Register every agent with an owner, purpose, risk tier, permissions, budget and review date, and give it its own identity and credentials.
Run a model gateway with authorisation by capability, quotas, budgets and guardrails, and decide how it behaves when it fails.
Control tools and actions per agent with a tool gateway, policy as code and kill switches.
Produce usage and cost evidence per agent and team, find unregistered AI use, and choose between building and buying a gateway.
About this course
Larkfield now runs six AI systems. The help desk triages tickets with a small, fast model. The support workflow resolves refunds with a strong model. A release judge checks new prompts before they ship. The data team’s document assistant answers staff questions. A tool server gives agents the tickets and the policies. And one more script sends a weekly digest of customer messages to a provider that Larkfield has no contract with. Nobody registered it, and the bill arrived before anyone knew it existed.
In this course, you follow Kwame, Larkfield’s solution architect, as he becomes the owner of the AI platform. Each app asks for a capability, such as “fast-small” or “strong-reasoning”, and the platform decides which model serves it. Every agent is registered before it can call anything, with an owner, a risk tier, a budget and its own credentials. One model gateway and one tool gateway check every call and record who made it.
You also learn the limits of the idea. A gateway sees model calls, not the agent’s work. Models behind one capability name are not interchangeable. A shared gateway fails only as badly as it is designed to fail. And registration works only when there is no easy way around it.
Everything runs on your computer with no account and no key. The models’ answers are real recordings from earlier courses, replayed by a provider simulator. Outages, slow answers and the unregistered script are practice cases, and each lesson says which evidence is replayed and which is simulated.
Before you start
You completed AI Agents and Workflow Orchestration, or you have the same skills. You can read and change Python, edit YAML, use a terminal and read JSON and HTTP requests. The authentication, MCP and evaluation courses help but are not required; lessons that use their ideas explain them briefly and link to them. You need no account, no key and no money. The estate runs on your computer, and every model answer is a replayed recording of a real model.
Platform, backend and AI engineers who run more than one AI application or agent in one organisation, and the security and architecture leads who set the rules for them.
Lead or use AI, but do not build it?
You can study the ideas of this course without writing code. Read modules 1, 2, 6 and 7 (about 8–9 hours), answer their knowledge checks, and get a free concepts certificate. You can come back for the practice later.
The paths for people who lead or use AI take this course this way. See the paths
How we made this course
Experienced AI practitioners designed this course. AI helped us draft the content; people chose the teaching approach and the examples, and reviewed every lesson. The outputs that you see are the actual results of running the code. How we make our courses
Inside the course
Module 1ConceptsWhy agents need platform controls
3 lessons
Why agents need platform controls. Meet a script that nobody registered, sort governance rules into those a document asks for and those code enforces, and start Larkfield's AI estate on your computer.
Ask for capabilities, not models. Write a capability catalog with contracts that include features, map each capability to a model with fallbacks and data rules, and treat a remap as a release with an evaluation gate per consumer.
Register every agent. Fill in a registration with an owner, a risk tier, data classes, tools, a budget and a review date, give each agent its own short-lived credentials, and run the agent lifecycle from registration to retirement.
The model gateway. Send every model call through one door that checks identity, registration, capability, data class, budget and guardrails, records usage, and keeps working when a replica, the configuration store or a provider fails.
Govern tools and actions. Put MCP servers behind a tool gateway, write approval rules as policy files with tests, and run a kill-switch drill on the support workflow.
Usage, evidence and shadow AI. Report usage and cost per agent and team, answer the governance board's questions from evidence, and find unregistered AI in bills, egress logs and code.
Build, buy and roll out. Compare the course's gateway with open-source and managed products, and plan a rollout that starts with registration and new agents.
Three end-to-end case studies on open data. Build a capability catalog from public evaluation results, contain a runaway agent replayed from open trajectories, and review public MCP servers for registration. Complete at least one for the certificate; each one is a portfolio piece.
Complete at least one case study for the certificate. The others are optional. What each one covers
Realistic projects from start to finish, on real data, with every output explained. Each one makes a portfolio piece. Complete at least one for the certificate; the others are optional.
Build catalog contracts and a mapping from openly licensed benchmark results and published prices, and show how the mapping changes when one model is retired.
Replay open agent trajectories through the gateway with budgets, step limits and tool policies, and measure what each control would have stopped and what it would have blocked wrongly.
Assess servers from an open registry of MCP servers: fill in the registration for each, set risk tiers and tool permissions, and list what the published information does not tell you.
Reviews
Loading the reviews…
Took this course? Sign in to review it. Learners can write a review after they complete 25% of the lessons.