Build a Software Factory: Hands-off agentic coding for experienced engineers

Matt Wynne

Cucumber co-founder, BDD pioneer

Aldric Giacomoni

ex-General Assembly, XP agentic builder

+ Jeremy Lightsmith and David Laing

Move from prompting AI to harness engineering

You can get agents to write reams of code for you (some of which is even very good). But afterwards you're left with a backlog of PRs of varying quality which still require your full focus and attention. Does this mean you're destined for a future as "Code Reviewer in Chief"?

Imagine a world where your agents consistently generate trustworthy output that you can deploy confidently without reviewing every line. A world where agents produce outcomes not just output.

We have lived in this world and we want you to join us!

This course is for you if you're:

Spending as much time correcting the agents output as you would have spent writing it yourself
Working in a brownfield codebase with landmines all around
Drowning in agentic PRs

Over four weeks you'll go from using your coding agent as a slot machine to architecting a reliable software factory.

What you’ll learn

Move from prompting AI to architecting trustworthy agentic systems

Learn what context to include, and what to omit, for quality responses
Treat "the AI" as one of several models — Claude, Gemini, ChatGPT — each with different strengths and blind spots.
Use disagreement between models as information, and treat their agreement as a weak signal rather than proof.

Build structured prompts with tool calls: you set the context and constraints, the agent works, you review the result.
Write paired execution and review skills with named inputs, explicit steps, done criteria, and escalation rules.
Make tacit judgment explicit by encoding how the work is done, not just describing what you want.

Connect agents to real systems with MCP, and add retrieval so context is fetched at runtime, not pasted by hand.
Manage the context window deliberately — relevance beats volume — and give the reviewer what it needs to audit, not guess.

Write the contract between doer and reviewer, plus specific, checkable exit criteria that define when the loop is done.
Use a separate verifier with fresh context and no stake in the output — the core move, even on the same model.

Add a different model as a critic to widen coverage, since separate same-model agents can share the same blind spots.
Close the gap with an external, model-independent check — a test, a schema, a lookup — because consensus is not verification.

Wire skills and loops into a pipeline that routes each step to the right model and handles failures on its own.
Design human checkpoints deliberately — where the agent acts, checks, escalates, or hands control back to you.

Learn directly from expert instructors

Matt Wynne

Contact

BDD pioneer, Cucumber co-founder, and creator of Example Mapping.

Aldric Giacomoni

Contact

Agent-native builder and educator shipping production software with AI agents.

Jeremy Lightsmith

Contact

Executive coach helping leaders navigate complexity in the AI era.

David Laing

Contact

Software supply-chain engineer bringing AI agents to SBOM and CVE work.

See all products from Matt

Who this course is for

Experienced, quality-conscious developers who already use AI but don't trust the output without reviewing every line.
Anyone who wants to see AI produce outcomes, not just outputs.
Anyone who has seen the potential in delegating work to agents and wants to do it more

What's included

Live sessions

Learn directly from your instructors in a real-time, interactive format.

Lifetime access

Go back to course content and recordings whenever you need to.

Community of peers

Stay accountable and share insights with like-minded professionals.

Certificate of completion

Share your new skills with your employer or on LinkedIn.

Sandbox & tokens included

A pre-provisioned cloud devcontainer lets you get hands on practice, even on a locked-down work laptop.

Hands-on, ~70% building

Most of your time is spent building on your own workflows, not watching slides.

Maven Guarantee

Your purchase is backed by the Maven Guarantee.

Course syllabus

Week 1

Jul 14—Jul 19

What AI is actually doing — fluency vs. correctness & the three big brains

Jul
14
Session 1
Tue 7/144:00 PM—5:30 PM (UTC)

Tool calls & the AI sandwich — your first agentic step + pick your capstone

Jul
16
Session 2
Thu 7/164:00 PM—5:30 PM (UTC)

Week 2

Jul 20—Jul 26

Reusable skills — write your paired execution & review skill

Jul
21
Session 3
Tue 7/214:00 PM—5:30 PM (UTC)

MCP, RAG & context — give your agent what it needs at runtime

Jul
23
Session 4
Thu 7/234:00 PM—5:30 PM (UTC)

Schedule

Live sessions

3 hrs / week

Spend 90 minutes twice a week with the instructors and classmates to learn, practice, and discuss key techniques.

Tue, Jul 14
4:00 PM—5:30 PM (UTC)
Thu, Jul 16
4:00 PM—5:30 PM (UTC)
Tue, Jul 21
4:00 PM—5:30 PM (UTC)

Self-Directed Practice

2-6 hrs / week

Apply the learnings from the live sessions to your own work or personal projects and bring back learnings to discuss. Work solo, in pairs, or groups to exercise and experiment with new learnings and techniques.

Office Hours

2 hrs / week

Drop in on the instructors with any questions or ideas you want to discuss.