What's your real entry point for AI coding?

Setting IDEs aside for this one - Cursor, Windsurf, VS Code + Copilot live in a different conversation. I'm specifically curious about CLI-and-agent territory.

A year ago the answer for most of us was straightforward: open Claude Code or Codex, talk to it, ship code. That CLI was the entry point.

Now the stack is wider. Hermes, OpenClaw, Klesh, custom MCP-based orchestrators, personal AI runtimes like PAI (Daniel Miessler's Personal AI Infrastructure), homegrown agent setups - some of you have built a layer on top of the CLI, and that's where your day actually starts.

So genuinely curious - when you sit down to do real work, what do you OPEN FIRST?

• Claude Code / Codex / Gemini CLI directly - still the cleanest

• A specific agent on top (Hermes, OpenClaw, Klesh, etc.) - name yours

• A personal infrastructure layer (PAI, custom runtime, your own framework)

• A custom orchestrator you built yourself

• A mix - depends on the task

For me: I'm a PAI user (https://github.com/danielmiessler/Personal_AI_Infrastructure). Claude Code lives underneath, but my actual entry point is PAI's runtime - skills, subagents, memory, and a 7-phase algorithm that forces structured execution for non-trivial tasks. I don't open claude and freestyle anymore - every session starts already shaped by my own infrastructure layer. Same Claude Code under the hood, completely different daily experience.

What's your setup? And if you've moved past direct CLI - what triggered the change?