Agentic Coding learnings and opinions

I've spent the last year using Claude Code, Codex, and spec-kit on real production work — not toy demos. This call is for engineers who want the honest version: what's working, what isn't, and what I'd set up if I were starting today.
What we can cover (you pick):
- Setting up Claude Code / Codex / Copilot CLI for your stack - hooks, skills, MCP servers, what's worth configuring vs. ignoring.
- Spec-driven and intent-driven workflows - when they help, when they get in the way.
- Compound engineering: getting agents to write the spec, the plan, and the code without drift
- Reviewing AI-written code without rubber-stamping it.
- Honest takes on where agents fail - refactors, debugging, anything stateful.
Picking models: GPT/Opus vs Sonnet vs Haiku, when each one earns its cost.
Who this is for:
Engineers (any level) who've tried Copilot / Codex / Claude and want to push further. Tech leads figuring out how their team should adopt this without shipping garbage. Founders evaluating whether the productivity claims are real.
Who this is not for:
"Will AI replace developers"
debates. I won't be useful there.
Come with a real problem from your codebase if you can - we'll work through it live.

Transcript