Cloudflare's cf CLI exposes 3,000 API operations, built for agents
Wrangler's ~280 commands spoke to humans; cf's 3,000 speak JSON to agents. Plus eight LLMs in a browser tab and Google's AI-and-fuzzing Rust recipe.

Copy markdown
The whole Cloudflare API, in one agentic command
cf covers 3,000+ API operations — up from Wrangler's ~280 — and defaults to condensed JSON so agents skip the `--json | jq` dance, with `cf cli search` for natural-language lookup. Install with `npm i -g cf`; `cf migrate` ports existing projects, and Wrangler keeps 18 months of support. Your agent can now deploy Workers, set WAF rules, and manage DNS through a single tool.
Eight tiny LLMs, running in the tab you already have open
MicroLLM Lab is an Apache-2.0 WebGPU playground that runs and benchmarks eight small models (26M–362M params, Q4) fully client-side — no server, no API key, weights cached in IndexedDB. On an M4, SmolLM2-135M hits ~66 tok/s and PetitGPT ~115 tok/s. A weekend starting point for shipping in-browser inference on narrow tasks.
Google's recipe for killing memory bugs: Gemini plus a fuzzer
Google rewrote giflib (~3,000 lines of C) into Rust with Gemini, using differential fuzzing to catch behavioral drift across 30M real GIFs and 200M fuzz iterations over six days. The result hit runtime parity and proved structurally immune to an unpatched heap-write zero-day (CVE-2026-26740). A repeatable pattern for retiring your own risky C/C++ dependencies.