Theory sessions build the mental model. Lab sessions make it yours. Every lab runs in a pre-configured cloud environment — no setup, just building.
Tokens, context windows, temperature, hallucination — explained precisely, not simplified away. Why models say what they say, where they fail, and what that means for how you use them. The foundation every other session builds on.
Take 3 real AI failures from your own history. Diagnose each one with your new mental model. Then rewrite your 3 most-used prompts with precision — system prompts, structured outputs, chain-of-thought — and see the difference live.
☁ Cloud lab pre-configured. Accounts ready before you arrive.
Advanced prompting: few-shot examples, chain-of-thought, structured outputs. Then: retrieval-augmented generation (RAG) — how AI reasons over your internal docs, runbooks, specs, and requirements. Includes a concrete look at generating test cases directly from acceptance criteria and specs.
Upload your own internal documents — runbooks, requirements, test plans, retros. Query them in plain language. Then take a real requirements doc and use AI to generate a first-pass test suite from it. See where it's precise and where it needs your judgment — and why that distinction matters.
☁ Document intelligence pre-configured in your cloud lab.
Tool use, planning loops, memory, orchestration — demystified. Why agents sometimes fail catastrophically and the design principles that prevent it. Real use cases across operations, quality, and project work: incident triage agents, bug classification agents, coverage gap detectors, sprint health monitors. The session most people say they wish they'd had first.
Pick a real, repetitive task from your job and build an agent for it using a visual builder in your cloud lab — no code, just configuration. Common starting points: an agent that triages alerts, classifies and routes bug reports, generates test cases from a ticket, or produces a sprint health summary. Run it live, end-to-end.
☁ Visual agent builder pre-configured and ready.
How orchestrators direct specialised subagents — shared memory, parallel execution, failure modes at scale. Then: Model Context Protocol (MCP) — the standard that lets AI connect directly to your real tools, files, APIs, and data sources without a custom backend. Why these two concepts belong together: MCP is how agents connect to tools, multi-agent is how agents talk to each other. Understanding both is what separates someone who demos AI from someone who deploys it.
Use MCP to connect the AI in your cloud lab to a real tool from your workflow — a ticketing system, a log source, a document store, or an internal API. Then wire it into a multi-step automated pipeline: trigger on an event, process with AI, route the output. No code. By the end of this session, AI is no longer a chatbot — it's part of your actual workflow.
☁ MCP connectors and workflow tools pre-configured in your lab.
Model versioning, deployment patterns, cost monitoring, drift detection, observability. How to validate AI outputs systematically — including AI quality gates in CI/CD pipelines and using AI to detect regression in test coverage. The gap between "I built an agent" and "I run AI in production responsibly." Everything you need to know without being an ML engineer.
Deploy the agent from Session 6 to a live cloud service. Configure a scheduled trigger. Add cost and usage monitoring. Come back the next morning — it ran while you were offline. This is the session where it clicks that these aren't demos.
☁ Deployment, scheduling, and monitoring pre-configured.
How to evaluate AI outputs systematically instead of guessing. Confidence thresholds, human-in-the-loop design, how to catch failure modes before they cause real problems. Then: the frontier — what's coming and what it means for your role specifically.
Each person demos the agent they built for their real workflow. Show what it does, where it failed, and how you fixed it. Leave with a working agent in production, a complete mental model, and a written 90-day plan for AI in your role.
🎯 Outcome · Working agent in production + personal 90-day roadmap.
Cohorts are capped at 5. Labs customised around your actual role.