Dev lists six fixes agent harnesses urgently need, from auditability to cloud-first sessions
threepointone · x · 2026-09-05
Developer darkmembo lays out what agent harnesses really need to fix: no manual chat/work switch, no model/effort picking, full auditability of executed commands, cloud-first sessions for seamless mobile-to-desktop handoffs, fast feedback loops with visible intermediate results, and richer outputs beyond plain text chat. threepointone amplified the thread.
More from coding & agent
- Agent Made 119 Tool Calls on One Task, But the Benchmark Only Scored the Final Answer — patience__01 · 2026-09-05
- lemire's layered model: a slow, hand-reviewed core, fast AI-driven outer layers — lemire · 2026-09-05
- Astra isn't live in Codex UI yet, but `codex -m gpt-6-astra` unlocks it today — lolapazoola · 2026-09-05
- qBotica: 335% growth going from 3-person RPA shop to ~200-person Agentic AI platform — rschmelzer · 2026-09-05
- Experimental vite-plugin-use-worker turns any async function into a Web Worker with one directive — cnakazawa · 2026-09-05
- anybridge open-sourced: let Codex and Claude Code drive websites via MCP — Just_Vugg_PolyMCP · 2026-09-05