A four-tool agent now turns livestreams into YouTube clips in about an hour
hugobowne · x · 2026-07-24
Hugo Bowne argues that coding agents are really general-purpose computer-use agents, not just helpers for writing code.
He uses a four-tool setup — Read, Write, Edit, Bash — to clean up a desktop folder in five tool calls, after the agent inspects files, proposes a plan, and asks for permission.
The same pattern now powers part of his real content workflow:
- the agent scans livestreams for clip candidates,
- sends transcripts for review,
- cuts video with FFmpeg,
- adds a logo and word-highlighted subtitles,
- verifies the render,
- drafts the title and description,
- and uploads to YouTube.
What used to take 15 clips and 4–5 hours now takes about 1 hour of agent work plus 20 minutes of human review. His core point: keep the loop small, and rely on skills, approval, verification, and workflow integration.
More from coding & agent
- A Claude Code user built mcp-guard after an MCP server leaked their .env — TastePrestigious4419 · 2026-07-24
- Dev Uses Claude to Build 3D City Generator and Weather Engine — repligate · 2026-07-24
- Mnemos shifts away from full agent orchestration and doubles down on identity — RileyRalmuto · 2026-07-24
- Codex helps build a route-planning web app for 100+ bar shifts — gabrielchua · 2026-07-24
- Long-Horizon Terminal-Bench leaderboard adds a new agent eval for terminal code tasks — Muennighoff · 2026-07-24
- GPT-5.6 Sol is shown unfollowing non-human accounts directly in Chrome — cocktailpeanut · 2026-07-24