# Greenfield Development with Claude Code

> Bilingual (EN/ES) reveal.js slide deck for a ~30-minute live talk by **Martin Brian** (Senior AI Engineer) at Marvik's "¡IA en vivo!" meetup — Montevideo, 2026-05-28 — on building an app from zero with Claude Code, bookended by a live, audience-played showcase game ("Simon Sees") that is rebuilt on stage from a written spec.

The talk's thesis: we can't keep up with AI tooling, but we can teach the *thought process* for choosing it. It runs in four rounds — (1) the game, (2) why software is Complex (Cynefin) and AI only amplifies the system around it (DORA), (3) how to reason with rules of thumb and a four-gate tool audition, (4) a live rebuild that regenerates the opening game from a spec while the frameworks that justify that spec are presented. The site is a single-page reveal.js app; press `T` or the globe icon to toggle English/Spanish, or load with `?lang=es`.

## Talk

- [Slide deck (live)](https://meetup2805.martinbrian.com/presentation.html): The full reveal.js presentation, EN/ES, ~37 slides across four rounds plus the reveal.
- [Full talk content](https://meetup2805.martinbrian.com/llms-full.txt): Every slide rendered as plain markdown — the complete narrative, frameworks, and source attributions in one file.

## Key frameworks

- [The four gates](https://meetup2805.martinbrian.com/presentation.html#/gates): Every tool a project adds must pass — Observability & Ownership, Correctness of Output, Cost, Simplicity & Maintainability — each scored A+·A·B·C·D·F against a per-project budget; drop the tool if it fails any single gate.
- [Picking a Claude Code primitive](https://meetup2805.martinbrian.com/presentation.html#/decision-framework): A top-down decision tree mapping the need to the primitive — permissions, hook, MCP server, sub-agent, CLAUDE.md, or skill (plugins are packaging, not a primitive).
- [Coherence engines, not truth engines](https://meetup2805.martinbrian.com/presentation.html#/facts-not-specs): LLMs produce coherence, not truth — so the verifier and the fact must live *outside* the model; write the tool that retrieves the fact.
- [AI is an amplifier](https://meetup2805.martinbrian.com/presentation.html#/amplifier): Per Google DORA, AI magnifies the strengths of high-performing teams and the dysfunctions of struggling ones — returns come from the system, not the tool.

## Optional

- [Cynefin — A Leader's Framework for Decision Making (Snowden & Boone, HBR 2007)](https://hbr.org/2007/11/a-leaders-framework-for-decision-making): The complexity framework behind Round 2.
- [DORA — ROI of AI-Assisted Software Development (Google)](https://cloud.google.com/resources/content/dora-roi-of-ai-assisted-software-development): Source of the "AI is an amplifier" framing.
- [Performing a Project Premortem (Klein, HBR 2007)](https://hbr.org/2007/09/performing-a-project-premortem): The pre-mortem exercise used in Round 3.
- [Rewilding Software Engineering (Wardley & Girba)](https://medium.com/feenk/rewilding-software-engineering-900ca95ebc8c): The "build the tool that retrieves the fact" / facts-over-specs lineage.
- [Claude Code documentation](https://code.claude.com/docs): Canonical reference for the primitives (permissions, hooks, MCP, sub-agents, memory/CLAUDE.md, skills, plugins).