The three workshops

Each session works on its own, in pairs with alternating roles of builder and reviewer. There is also a talk version, at 40+10 minutes.

1. From idea to a working agent

Bedrock, Strands and AgentCore in practice, with read-only tools and adversarial cases. Participants learn to separate observation from hypothesis and to reject a fluent answer that has no evidence.

You need: nothing for the simulation. An AWS account only for the optional mode.

2. Architecture under pressure

A fictional payments platform faces a spike, slowness and duplicates, with a scenario card every ten minutes and the capacity math in hand.

You need: pen and paper.

3. From briefing to an architecture you can defend

Requirements, primary sources, two alternatives and an ADR reviewed by someone else, with AI and MCP as support. The decision stays “proposed” until authorization, quality and cost are validated.

You need: a browser. Node.js for the architecture MCP, optional.

Success criteria

The participant presents one decision, the evidence, one risk and one next test. Likes don’t count: record completed exercises, questions and reproduction reports.

Full material

In Portuguese, in the repository:

• Participants’ workbook (in Portuguese): briefings, cards and answer key
• Facilitator’s guide (in Portuguese): minute-by-minute script
• Templates (in Portuguese): ADR, evidence log and rubric
• Proposals (in Portuguese, with English summaries): texts for organizers
• Presentation PDF

Note

Publishing these materials doesn’t mean acceptance on an agenda or a reserved slot. The audience and timing estimates are not measurements yet.