AI Agents Series

Explore the five modes of collaborating with AI coding agents in Elixir — and why 'vibing' creates fragile systems.

Why AI agents lose context, how the conveyor belt model explains it, and a practical habit Elixir developers can use immediately to get better results.

Stop comparing features and benchmarks. Learn to match AI agents to your workflow with a practical decide-vs-execute rule you can use today.

The model didn't forget your architecture—it just never lived anywhere durable. Learn how to anchor decisions in structure instead of prompts.

Learn the Ask → Plan → Agent workflow, a two-level safety net, and context discipline for checkpoint-based AI coding in Elixir.

A 30‑minute Phoenix LiveView demo looks magical – until you see how the yes loop creates hidden scope and long‑term maintenance cost.

A fast Phoenix LiveView build can look impressive, but hidden structural debt appears when you inspect what AI generated under speed pressure.

If you keep retyping the same AI instructions, you do not have a workflow. Learn a practical three-layer structure to keep coding agents consistent.

Learn when to use rules, skills, sub-agents, and slash commands so your AI workflow scales with structured oversight instead of context-heavy repetition.

Learn when AI tooling should live globally versus in the project so shared workflows stay versioned, visible, and aligned with the codebase.

Learn how meta workflow generation turns prompts into reusable project commands, making AI development workflows durable, repeatable, and reviewable.

AI can produce a convincing architecture plan, but production systems need structured review. Bruce Tate shows how an Elixir architect catches security and design mistakes before implementation begins.

Bruce Tate shows why pure functions and the CRC pattern (construct, reduce, convert) don't just help humans reason about code — they help Claude generate, verify, and stay correct with fewer tokens.

Learn how Bruce Tate manages Claude context windows with compacting, clearing, and priming while preserving layered Elixir architecture and one source of truth per layer.

Why Elixir's functional core, process isolation, OTP recovery, and runtime metadata make it a strong language for AI-assisted development without losing architecture decisions.

Learn how Claude's context hierarchy, AutoCompact, and memory tradeoffs affect AI-assisted Elixir development, and why structured oversight starts with managing what enters context.

Install Tidewave in a Phoenix app and learn the two ways to wire Claude into your Elixir workflow through browser-based Tidewave or Claude Code MCP.

Learn how Tidewave's project_eval lets Claude Code inspect, test, and demonstrate behavior inside your running Elixir application through a shared IEx-style workflow.

Learn how to review AI-generated Elixir code by separating functional core from boundary code, reading public APIs first, and catching missed domain requirements.

Learn why AI-assisted projects need three sources of truth aligned: the codebase, the agent context, and the repository that preserves prompts, plans, and project guidance.

Learn how feature branches, binary verification, and pull requests create safe boundaries for AI coding agents while keeping humans in control of architecture decisions.