Full syllabus

Agentic AI for Products & Systems

From use case to production decision.

A two-day hands-on workshop, run on a real project you bring: a product, an internal system, a workflow, or a new capability. This page is the detail behind the workshop overview: what is covered, how it is taught, and what you need before you arrive.

Format: Two days, sixteen hours, live online. In person or hybrid for private workshops.
Group size: 10–25
Prerequisites: Coding fluency, working knowledge of system design, experience on complex projects
Pre-work: The one-page Agent Opportunity Brief, submitted and reviewed before day one
Led by: Samir Joshi

Who it is for

Senior engineers, architects, and technical leaders responsible for turning agentic AI ideas into production decisions.

The room is deliberately mixed, because building and deciding are two halves of the same job. Everyone joins framing, the first build, evaluation principles, architecture, product strategy, and the capstone. Engineers go deeper into implementation, orchestration, and the security labs. Architects and leaders go deeper into operating model, build-buy-defer, and economics.

Designed for experienced professionals; we confirm fit before the workshop.

Before you arrive

The workshop runs on your own project, and the machinery for that starts before day one.

You submit the Agent Opportunity Brief: a one-page brief on your project, from “why agentic, not plain software” through build, buy, or defer.
We review it and reply with tailored suggestions before you arrive.
Every exercise is re-pointed at what you submitted.
The capstone is built against your project, and you leave with a reasoned decision about what to do with it.

The first working agent is low-code, so the first build excludes no one. Engineers rebuild it properly in code later.

The six parts

Part	Title	What it covers	Who
1	Foundations	Concepts to your first working agent; everyone builds it	Everyone
2	Core concepts & building blocks	Context engineering, RAG, and the protocols (MCP & A2A)	Everyone
3	Building agents	Single-tool to multi-agent orchestration, in code	Engineers & tech leads
4	Evaluation, AgentOps & production trust	Evals, observability, cost, security labs, deployment	Engineers & architects
5	Strategy	Where should agents belong, and under what operating model?	Architects (tech) & leaders (mgmt)
6	Productize & capstone	Package the capability; take the production decision on your own project	Everyone

The depth behind those titles

Evaluation as a first-class discipline, and AgentOps: DevOps for agents, covering tracing, evals, cost control, CI/CD, and guardrails.
One framework deep (LangGraph), with the rest mapped: LangChain, OpenAI Agents SDK, Google ADK, AWS Strands, CrewAI, AutoGen, LlamaIndex, and low-code tools. Everything is written against patterns, so your stack can swap in underneath.
The two protocols of 2026: MCP, how agents talk to tools, and A2A, how agents talk to each other.
Security, hands-on: live prompt-injection and data-exfiltration labs, over-permissioned tools, guardrail libraries, sandboxing.
Cost discipline: token budgets, tracking, alerts. From Part 2 on, every design exercise carries a latency, cost, retry, fallback, and escalation budget.
Where RL and RLHF fit, and where they don’t.

How it is taught

One case, end to end. A legal-document agent grows across the two days: clause summary, precedent retrieval, whole-contract review, hardening, business fit, packaging. Not disconnected toy projects.

A deliberately broken agent. You meet ClauseBot v0 and debug it. It fails a new way at every stage.

Applied assessment. Broken-agent debugging, an architecture design sprint, code review. No quizzes.

Satellite examples run alongside: customer support, software engineering, incident response, fintech, healthcare, education.

What you leave with

A working agent slice, built against your own real project rather than a sample app.
A production decision – ship, extend, buy, defer, or stop – with the evidence behind it.
A reusable working kit: the tests, checklists, and briefs used as gates through the workshop, a complete reference implementation (agents, tools, retrieval, evaluation, tests, CI), and a tracing, evaluation, and cost-control toolchain stood up on your own accounts. Free tiers get that started; beyond that you pay the providers, not us.
A year in the alumni community.

Two days will not make a team production-ready, and we don’t claim it will. What it gives you is a working slice, the failure modes that actually bite, and a reasoned view of what production would require. Production itself is the work that follows.

Confidentiality

In a public workshop, bring a project you’re free to discuss. Proprietary data never leaves your control (bring-your-own-key or local models throughout), and sensitive implementation detail should be sanitized before it comes into the room. The room runs on a Chatham House norm: what’s said in the room stays there, what’s learned travels. That norm is a working convention, not a legal instrument.

In a private workshop for your company, your real codebase and internal architecture are in scope, under a mutually agreed NDA. Nothing to sanitize, and nobody from another company in the room.

Attend

The workshop runs as a public cohort, live online, or privately for your team. Price is shared when you register interest, with no commitment; private workshops are quoted separately.