You just submitted another pull request that got away from you. The Claude session started well. Ten hours later, you’ve mostly rescued the main ideas — but the code is nowhere near what you know it could be. It’s too long. It doesn’t reflect the values or the prior art in the codebase. It gravitates toward the exact concepts your team works hard to avoid. You shipped it anyway because the deadline was real. Now it sits in your lead’s review queue alongside four others just like it, and you both know it.
What would have made that afternoon different?
This full-day, hands-on training teaches the prompt and project architecture that turns agentic coding from a coin flip into reliable, reviewable work. You’ll build the patterns on real Phoenix code, with feedback checkpoints at every layer, so you leave with muscle memory — not just notes.
DURATION
8 hours (including breaks and lunch)
WHAT YOU’LL BUILD:
- A layered prompt architecture you can apply to any Phoenix application.
- Specialized prompts for the targeted layers of your stack.
- Workflows with feedback checkpoints that catch drift early.
- Project anchors - markdown, comments, and memory files - that keep your prompts and your codebase aligned across sessions.
WHAT YOU’LL LEARN:
- Decomposing prompts into testable layers with fast feedback
- When to create skills vs. agents vs. tasks
- Using artifacts to anchor workflow state
- Debugging by isolating layers
AUDIENCE: Elixir developers seeking systematic AI-assisted development approaches. Applicable whether using Claude Code, Cursor, or building custom AI integrations.