Essay Hook — "Whose Keys Are in the Lock?"

Use this outline for a Substack or Medium post that drives readers to the repo.

Working title

Whose Keys Are in the Lock? Why Assistant vs Claw Is the Agent Decision That Actually Matters

One-line thesis

Building agents is easy now; authorization is the bottleneck — and LangSmith Fleet's Assistant vs Claw split is the first design choice that determines whether you leak data or ship something useful.

Suggested structure

1. The bottleneck moved — from "can the LLM call a tool?" to "whose identity calls it?"
2. Two stories — Alice/Bob onboarding (Assistant) vs @product-bot in #general (Claw)
3. The common mistake — Claw wired with personal OAuth (include avc-slack output)
4. Beyond credentials — memory, inbox, audit (link to Fleet Engineering patterns)
5. CTA — clone assistant-vs-claw, run avc-compare, read Fleet Engineering

Pull quotes

Same word — agent. Two completely different trust boundaries.

Not autonomy. Not model choice. Authorization.

Links to include

• Repo: https://github.com/cobusgreyling/assistant-vs-claw
• Interactive docs: https://cobusgreyling.github.io/assistant-vs-claw/
• Colab: https://colab.research.google.com/github/cobusgreyling/assistant-vs-claw/blob/main/notebooks/assistant_vs_claw.ipynb
• Substack (Fleet Engineering): https://cobusgreyling.substack.com/