A three-session, 90-minutes-each bootcamp that helps engineering teams move from AI-assisted development (one tool doing everything) to AI-native development (deliberately assigning the right tool to the right role).
If your AI plans the system, executes changes, edits files, reviews results, and declares success on its own, you have not built a workflow. You have abdicated responsibility.
This curriculum is free, open, and designed to be forked and adapted to your team. It is stack-agnostic — it works whether your team writes Rails, Go, Django, Next.js, or anything else.
Background reading: Upgrade path available — the essay behind this curriculum.
Most "AI for developers" training focuses on tools. This one focuses on roles.
1. Thinking / Planning — architectural reasoning, design decisions
2. Executing — writing code, running commands, making changes
3. Editing — modifying, refactoring, adjusting existing work
4. Reviewing — verifying correctness, quality, security
5. Iterating — cycling back through 1–4 until accepted
Most developers today collapse all five into one tool (usually ChatGPT or Cursor). That feels productive and is actually a liability: when a single tool thinks, acts, and grades itself, you lose every checkpoint that separates "works" from "correct." This bootcamp is the antidote.
See framework/five-role-model.md for the full argument.
- Engineering teams of 3–10 who are already using AI but sense they're using it badly.
- Tech leads who want to set workflow conventions their team can actually follow.
- Engineering managers shopping for structure before "let's use AI more" becomes a mandate.
- Solo developers — most of the team-coordination content doesn't apply to you. You can still skim the framework and gotchas.
- "How do I write prompts?" training — this is about workflow architecture, not prompt craft.
- Teams without any AI exposure — you'll get more from two weeks of experimenting first, then running this bootcamp.
- Read
SYLLABUS.mdtop to bottom. - Work through the Preparation Checklist in the syllabus — it has three phases (content prep, pre-work rollout, dry run).
- Send
exercises/participant-pre-work.mdto participants a week before Session 1. - Run Sessions 1–3 from
sessions/, one week apart. - Follow up with
reference/post-bootcamp.md.
- Fork this repo.
- Replace placeholders (
[your stack],[your codebase],[your VCS host]) with your specifics. - See
framework/adapting-to-your-stack.mdfor worked examples — Node/TypeScript, Python/Django, Go, and Rails. - Customize
templates/for your team's conventions. - Keep the five-role framework — that's the durable part. Everything else is scaffolding you should reshape.
The curriculum is intentionally short on specifics so you can bring yours.
ai-native-bootcamp/
├── README.md (this file)
├── SYLLABUS.md high-level syllabus + learning outcomes
├── CONTRIBUTING.md how to contribute back
├── LICENSE MIT
│
├── sessions/ detailed facilitator plans (90 min each)
│ ├── 01-foundations.md
│ ├── 02-agentic-workflows.md
│ └── 03-team-conventions.md
│
├── framework/ the durable concepts
│ ├── five-role-model.md why role separation matters
│ ├── orchestration-matrix.md which tool for which role
│ ├── capability-boundaries.md what each tool can/cannot do
│ └── adapting-to-your-stack.md worked examples per language/framework
│
├── exercises/ handouts for participants
│ ├── participant-pre-work.md send a week before Session 1
│ ├── session-1-role-aware-task.md
│ ├── session-2-self-calibration.md
│ └── session-3-team-orchestration.md
│
├── templates/ starting points for team artifacts
│ ├── CLAUDE.md.template project-context file for AI tools
│ ├── team-conventions.template.md what you decide in Session 3
│ └── conventions-decision-sheet.md the decision checklist for Session 3
│
└── reference/ supporting material
├── common-gotchas.md real friction you will hit
├── cost-reality.md honest view of what this costs
├── security-conventions.md what never goes in an AI tool
└── post-bootcamp.md follow-up + living documentation
| Durable (don't change) | Scaffolding (change freely) |
|---|---|
| The five-role framework | Specific tool recommendations |
| "Separate thinking from executing" | Session timings |
| "Trust but verify — reviewing is always yours" | Example tasks |
| "The human decides, the AI implements" | Which MCP servers are in scope |
| Security boundary around customer data / secrets | Specific stack references |
Tools change quarterly. The framework should still make sense a year from now even if every tool in the orchestration matrix has been replaced.
PRs welcome — see CONTRIBUTING.md. Good candidates:
- Field reports: "We ran this with X people on Y team, here's what worked / didn't"
- New gotchas that teams keep hitting
- Translations
- Clarifications to the framework
Not a good fit: prompt libraries, model benchmarks, tool rankings. This is a workflow curriculum, not a tool catalog.
MIT. Fork, remix, teach it at your company, turn it into a blog post, adapt it for your language. Attribution appreciated but not required.