The topology of an LLM agent: where the model, loop, and tools run shapes cost and security
davidcrawshaw · x · 2026-10-21
exe.dev's blog proposes analyzing agents by topology: a model (defines smarts), the agentic loop (defines UX), and tools (define capability and blast radius)—the latter two forming the harness. Where these run affects cost, capability, and security. Key points: choosing hosted vs. own GPUs carries training-data leakage risk (local models suit privacy-first, latency-insensitive tasks like OCR-ing scanned bills); remote MCP servers can expose tools running elsewhere; and Ken Thompson's 'Reflections on Trusting Trust' reads differently in the agent era.
More from coding & agent
- Beat AI's samey frontends: import a design system before generating UI — dotey · 2026-10-21
- Trust slider over Clairic: 102 manual reviews catch 74 of 76 extraction errors — michellechen · 2026-10-21
- FreePi Bets on Ad-Supported Coding Help to Undercut $20/Month AI Coding Subscriptions — heyneighbor · 2026-10-21
- RedMonk: MCP and A2A Verify Publishers, Not Safety — Registries Become the Trust Layer — rseroter · 2026-10-21
- Agentic refactor: 600 PRs moved OpenClaw's web UI from Lit to Solid 2.0 in 2 weeks — steipete · 2026-10-21
- 19 Recorded Runs of 18 Agents Building a Lisp Interpreter, Released on Hugging Face — strickvl · 2026-10-21