Empathy unlocks Claude Code

Watched the new YC interview with Anthropic’s @nottombrown and my big takeaway: Claude Code didn’t need secret sauce - any solid team could ship a CLI on the Claude API, but the leap was empathy for the model. Build a coding tool for Claude, and you get a better tool for us. 💻👇

Why this matters in 2025: Claude Code is a public, npm installable CLI and SDK on the same models we all can call, with first party integrations and guardrails. I’ve been using it for weeks/months. It feels like a reference app others can copy or rival.

What designing for the model looks like in practice: put the agent where it lives, the terminal and your repo. Write prompts for the agent to perform best, including internal evals not for benchmarks but results. Give it codebase search, permissioned edits, IDE hooks, and standard tools via MCP. Plans should become runnable workflows, not pasteable snippets.

Nuance: This is “not your mom’s” CLI - reliability, session state, permissions, and recovery are hard. Early reports show big wins and sharp edges, like context compression or risky cleanup commands. Platform politics about who gets model access also shape the field, see Windsurf for example.

Implication: Model empathy might be the new product sense. Design the environment the model can actually operate in and humans benefit.

Where else should we build for the model first, data analysis, ops, research tools.

And can the agent ecosystem stay open while labs ship first-party apps?

Watch the video with @garrytan, my YC partner @sdianahu and the rest of the team here:
https://www.youtube.com/watch?v=JdT78t1Offo