Workshop · Tokyo
Tokyo · 25 – 29 May 2026 · On-site · Hands-on
A five-day workshop for an engineering team in Tokyo. Every engineer leaves Friday morning having presented a working MVP they built across the week — twelve engineers, twelve different MVPs, all on top of the Claude Agent SDK and the orchestration substrate walked across Days 1 – 4.
The shape of each day is short framing (~15 – 30 minutes) followed by long hands-on (~1 – 2 hours), repeated through the day. Lectures stay concise; the majority of time is building.
The five days
Mon · The Why + SDK basics
Morning · platform vision, live multi-agent demo, Q&A. Afternoon · build a single agent in the Claude Agent SDK, then convert to multi-agent — the MVP starting point.
Tue · Orchestration + the nine agents
Morning · build together — engineers run the nine-agent pipeline in Claude Code by pasting prompts that point at instruction files. Afternoon · browser logins, modify agents, test the generated guardrails actually work.
Wed · Source + office library + memory
Morning · open the source, office library walkthrough, memory + context-window strategies. Afternoon · wire generated hooks into the SDK runner and test, or build a tenth agent or custom front-end.
Thu · IoT demo + MVP prototyping
Morning · production IoT agent-team demo, then free prototyping time for engineers to refine their MVPs. Afternoon · AI-ready data preparation, joint design session.
Fri · MVP presentations
Morning · engineers present their MVPs to the senior audience — twelve people, twelve working systems. Afternoon · AI guardrails opinion exchange, reflections, close.
Source guides — download
The three step-by-step guides each engineer follows during the week. Same files attached to the workshop kickoff email — keep them offline, paste into Claude before the workshop, or share with a teammate.
What every engineer leaves with
- A working MVP they built themselves, presented on Friday morning
- A working coding agent on their own laptop (Claude Code + Claude Agent SDK)
- Browser access to the hosted multi-agent platform with twelve logins for the team
- Cloned source for the orchestration runtime, with branch access
- A pattern vocabulary — heartbeat-watching, multi-perspective fan-out, approval gate, composed — for agent-native applications
Session types
Two types, formalised so participants know what to expect:
- Discussion sessions — open for both in-person and remote attendance. Vision presentation, architecture walkthroughs, opinion exchanges.
- Hands-on workshop sessions — in-person engineers only. Building, modifying, testing, prototyping.
Format
Sessions follow the host’s standard timings. Morning blocks are typically two hours. Afternoon sessions are two to three hours, hands-on. The week is delivered bilingually — English and Japanese — with diagram-heavy slides and a printed glossary of key terms.