Architecture with Claude Code
Claude writes a convincing architecture document in under a minute — including the parts it made up. In one day you run a complete architecture engagement on a realistic case with Claude Code, and learn where AI genuinely accelerates your work and where your judgment has to stay in charge.
Any questions?
Melchior will gladly help you further with any personal or in-company needs you might have.
Melchior Schimmel
Chief Sales
What will I learn?
Claude will write you a polished architecture document in under a minute. It will also invent a constraint, over-trust a Slack thread, and present an inference as an established fact — all in the same confident tone. That is the real problem AI creates for architects: not bad output, but output that is too plausible to review casually.
This training teaches the working method that makes AI genuinely useful for architecture work anyway. Over one day you run a compressed but realistic engagement with Claude Code as your working partner: making sense of a vague request, following evidence as it shifts, judging when you know enough to move into design, reviewing a proposal against its alternatives, and recording a decision you can defend. Every architect does some version of this, whatever their organisation calls the stages.
You learn to bound Claude’s context, keep provenance intact, retrieve only what a decision actually needs, and prove that your documents and diagrams still agree instead of trusting fluent prose. Above all, you learn to draw a clear line between what AI prepares and what you, as the accountable architect, decide.
Key takeaways
- Run a complete architecture cycle with Claude Code — from a vague first request to a recorded, defensible decision — on a case that behaves like real work.
- Tell the difference between what AI can prepare and what only you can decide, and make that boundary visible in your artifacts.
- Leave with a reusable setup — project instructions, skills, and automated checks — that you can point at your own architecture repository. and advance on your journey to Professional Product Management.
Program
The day follows one continuous case. A B2B SaaS company wants selected customers to buy prepaid capacity and share it across several organisations — a request that quietly breaks its customer, commerce, subscription, and capacity models. You work it as the architect, from the first intake document to a signed-off decision record.
Every block follows the same rhythm: you hit the friction yourself, we name it, you get one technique to deal with it, and you apply it immediately to the case. There is no feature tour. Techniques are introduced when the work demands them, and you leave with the ones that survived contact with real architecture work.
Expect a mix of short theory, live demonstrations, and hands-on work in small groups, with discussion at every gate.
- Meet the case: you take on a real-looking architecture request at a fictional B2B SaaS company and put Claude to work on it straight away, with no method yet — so you feel the failure modes yourself before we name them.
- Making sense of the request: bound Claude’s workspace, separate source documents from generated output, and turn a vague, contradictory request into an analysis you can defend line by line.
- When the evidence changes: new Finance and Support findings arrive mid-analysis. Manage a long, drifting conversation, preserve provenance, and revise your conclusions explicitly instead of silently.
- Judging when you know enough: turn analysis into a recommendation, then decide for yourself whether it is ready to move forward. Fluent output has no decision authority.
- Moving into design and reviewing a proposal: retrieve only the evidence a decision actually needs from a large corpus, then work through a deliberately flawed design proposal (RFC) — comparing every option against the same rubric and delegating expensive claim checks to a subagent that returns evidence rather than a verdict.
- Keeping design artifacts aligned: a design change lands. Trace its impact across your decision records, model, and diagrams, and let automated checks prove alignment instead of trusting plausible output.
- Recording and defending the decision: break the outcome into decision records (ADRs), get challenged by an impacted team, disagree and commit, and map what you learned onto the way your own organisation works.
Who is it for?
This training is for people who own architecture decisions and have to defend them: software architects, solution architects, principal and lead engineers, tech leads, and engineering managers who write or review RFCs and ADRs.
It is aimed at practising architects who are new to working with Claude Code, or who have tried it and found the output impressive but hard to trust. You do not need prior experience with Claude. You do need real architecture work to apply it to.
Typical job titles: Software Architect, Solution Architect, Enterprise Architect, Principal Engineer, Lead Developer, Tech Lead, Engineering Manager.
Requirements
- Several years of hands-on software architecture experience: you have written or reviewed RFCs, ADRs, or design documents.
- Your own laptop, with rights to install software, and a working internet connection.
- Claude Code installed and authenticated before the day, plus an active Claude subscription or API access.
- Basic comfort with Markdown and with working in a terminal or IDE.
- No prior experience with Claude Code or any other AI assistant is required.
Why should I follow this training?
Judgment stays with you
Learn exactly where AI assistance stops and your accountability as an architect begins.
One realistic case, all day
No toy examples. You work one messy, evidence-driven case from first request to recorded decision.
A method you can reuse on Monday
Take home project instructions, skills, and checks that work on your own architecture repository.
What else
should I know?
After registration for this training, you will receive a confirmation email with practical information. A week before the training, we will ask you about any dietary requirements.
See you soon!
Course information
Lunch is included.
Bring your own laptop with admin rights and Claude Code already installed.
An active Claude subscription or API access is required and not included.
You receive a certificate of participation; there is no exam.
Meet the trainers
Rene Bulsing
Learn how René turns complex software and AI challenges into clear, practical solutions that work in the real world.
Also interesting for you
View all training coursesShip intelligent features to production—not prototypes. While others struggle with LangChain theory, you’ll master semantic search, RAG systems, and multi-agent workflows using battle-tested Spring-AI and LangChain4j. Deploy AI that actually works in enterprise JVM apps.
