Skip to content

Day 4 · IoT demo + MVP prototyping + Data prep

Thursday 28 May 2026 · Tokyo · Engineers, in person

Day 4 is where the week’s MVP work consolidates. The morning pairs an end-to-end IoT agent-team demo with open prototyping time so each engineer can shape what they will present on Friday. A senior external-meeting track runs in parallel.

Morning · parallel tracks

Main venue — IoT demo + free MVP prototyping (~3 hours)

  • End-to-end IoT agent-team deployment shown as a working production reference (~1 hour)
  • Free prototyping time so engineers can refine their own MVPs for the Friday presentation (~2 hours)
  • The engineering team is on hand throughout — questions, pairing, design discussion

External meeting (parallel) — a senior meeting at a partner organisation, running in parallel for part of the team.

IoT demo + prototyping detail · Partner meeting detail

Afternoon · ~2-hour block + open

Joint design session on AI-ready data preparation.

  • The team’s data-prep work — metadata catalogue, search, lineage
  • The five-stage office-library pipeline walked end-to-end
  • Joint design exercise — one in-house dataset mapped through every stage
  • Open time after for any topic the team wants to surface

Data-prep detail

How the IoT demo connects to the workshop

The IoT agent-team is the same Claude Agent SDK system the workshop teaches — built with the patterns from Day 1 (single → multi-agent SDK), driven by instruction files like the ones in Day 2, with guardrails of the kind tested in Day 3 — but deployed at production scale.

In an IoT monitoring context: alerts arrive in email or a dashboard. An agent team picks them up, researches across telemetry, manuals, and regulatory circulars, drafts an email reply, and a human reviews and approves. The system saves the resolution pattern and learns over time. Engineers monitor; agents do most of the work.

This is a concrete reference for how the workshop’s learnings deploy in real operations — a useful starting point as engineers shape their own MVPs.

Four agent-pattern reframes — reference

The week surfaces four reusable patterns. Each engineer’s MVP usually combines two or three.

  1. Heartbeat-watching — agent watches a stream and acts when conditions trigger, replacing “user opens app and clicks button”
  2. Multi-perspective fan-out — many lenses on the same source, recallable separately
  3. Approval gate — agent prepares, human approves via channel, agent executes
  4. Composed pattern — heartbeat + multi-perspective + approval gate for complex workflows

Materials

Next: Day 5 — engineers present their MVPs →