Agentic Systems · 14 modulesBuild an AI agent that's governed from the first line.
14 modules, from what an agent is to deploying one responsibly. Each module teaches one idea and has you build it into one working support agent, with tests that prove it works. Runs on your laptop, maps to any cloud. About 18 hours in all.
How this series is different
Any cloud, or noneEvery lab runs on your own laptop with Python and no cloud account. Each module then names the managed services that do the same job on AWS, Azure and Google Cloud.
Governed from module 1The policy file, scoped tools, tamper-evident audit log and cost ledger are in the first lab, not saved for the last module.
One project you keepFourteen labs build one support agent for a fictional store, piece by piece. Tests prove each piece works before you move on.
No forms, no drip, no upsellModules 1 and 2 are free with no signup. Pay once and all 14 open at once. No partner tools to buy along the way.
Every cost statedEach lab runs offline for $0. The optional live run uses your own model API key, and the kit prints what each run cost.
Proof you learned itA five-question check closes every module. Pass all 14 and print your certificate.
The 14 modules
Modules 1–2 are free. The rest open the moment you pay.
Foundation · modules 1–2
What an agent is, when not to build one, and the parts every agent is made of.
- Introduction to AI agents · 60 min · free · passed
What an agent is (model, tools, loop, stop conditions), when not to build one, and a first governed Kettle agent for order-status tickets. - Agent frameworks and building blocks · 75 min · free · passed
What frameworks add to the plain loop, the building blocks they share (model adapter, tool schemas, memory, MCP), and a refund tool under policy.
Building · modules 3–6
Evaluate it, split the work across agents, write its spec, and orchestrate the workflow.
- Agent evaluation and decision engines · 75 min · in the series · passed
Score the agent against a labelled answer key, block any change that makes it worse, and keep deterministic rules apart from model judgment. - Multi-agent architectures · 70 min · in the series · passed
When a router and specialists beat one agent, how each gets only its own tools, identity and policy, and how to pass context without passing power. - Domain-specific agent specifications · 70 min · in the series · passed
Write what the agent may and may not do as a machine-checkable spec, compile it into the policy file, and check every policy and every run against it. - Workflow orchestration · 75 min · in the series · passed
Wrap the agent in a deterministic workflow with retries, idempotent side effects, timeouts and a human approval step that can resume days later.
Operating · modules 7–12
Memory, identity, data lineage, integration, the dev lifecycle, security and observability.
- Agent memory systems · 75 min · in the series · passed
Short-term memory inside a token budget, long-term memory per customer, what must never be remembered, and expiry and deletion you can prove. - Agent identity and access management · 75 min · in the series · passed
The agent as its own identity, short-lived scoped tokens, acting for a customer or as the service, checks on the tool side, and a least-privilege review. - Data pipelines and lineage · 70 min · in the series · passed
Feed the agent through a pipeline that rejects broken files, quarantines bad rows, checks freshness and traces every answer and refund to its source row. - Agent exposure and integration · 75 min · in the series · passed
Put the agent behind a local HTTP API with key, size, input and rate checks, describe its tools for MCP clients, and verify signed helpdesk webhooks. - MLOps and the agent dev lifecycle · 90 min · in the series · passed
Version the prompt, policy and model as one release, ship it through an eval gate, checklist and tests, try it in shadow and canary, and roll back. - Security and observability · 90 min · in the series · passed
Hold the agent to its limits under prompt injection, filter its replies, keep secrets out of logs, trace every step, alert on spikes and verify the audit log.
Scaling · modules 13–14
Make it faster and cheaper without losing quality, and deploy it responsibly.
- Optimization and scaling · 90 min · in the series · passed
Measure cost and latency per ticket, route easy tickets to cheaper tiers with an eval proving quality holds, then cache, batch and budget at volume. - Responsible AI deployment · 90 min · in the series · passed
Kill switch, approvals and overrides; AI disclosure to customers; fairness checks across segments; an incident runbook; a readiness list that checks itself.
All 14 modules · $49, once
- Every module, lab, knowledge check and the certificate
- The code kit: starter code, solutions and tests for all 14 labs
- A license to use the kit's code in one business
Questions
Do I need an AWS, Azure or Google Cloud account?
No. Every lab runs locally with Python 3.10 or newer. The cloud sections show how each idea maps to managed services, so you can take the design to whichever cloud you use.
Do I need to pay for a model?
No. Each lab includes a scripted model so it runs, and its tests pass, without any API key. If you want to see a real model drive the agent, the kit supports your own Anthropic API key and records the cost of every run.
Who is it for?
Developers and technical leads who are building, or about to build, an agent that touches real customers, money or data, and need it to hold up in front of security, finance and compliance.
What do I get when I pay?
All 14 modules, the full code kit with starter code, solutions and tests for every lab, the knowledge checks and the certificate. A one-time payment, no subscription.