ParallACT: when every dev has their own coding agent, coordination still breaks down
EliasEskin · x · 2026-11-04
Researcher rishabhmediratt announces ParallACT, a benchmark built with EliasEskin, xiaolingyun98, Peter Stone, and Joydeep Biswas.
The core setup: current multi-agent coding typically splits one request across parallel agents with centralized planning. But when many devs each give their own task to agents, nobody plans who edits what — and agents can't sort it out yet.
The quoted original thread details the multi-principal multi-agent setting where Opus 4.8 achieves only 16% success even with inter-agent communication.
Related event: UT Austin's ParallACT Benchmark: Parallel Agents' Success Crashes to 16%(9 posts)→
More from coding & agent
- Anthropic engineer warns of 'cognitive surrender': AI coding erodes skills needed to supervise it — gdechichi · 2026-11-04
- How LangChain optimized browser use for Deep Agents with Stagehand — LangChain · 2026-11-04
- Plane AI scores 165/165 on real workspace eval: 18s median latency, $0.10 per answer — JosephJacks_ · 2026-11-04
- SurrealDB 4 launches with SurrealDB Earth: serverless agent database on Cloudflare — ritakozlov · 2026-11-04
- Karpathy's "vibe coding" post hailed as a shoo-in for the test-of-time award — MillionInt · 2026-11-04
- OpenAI SDK ships official WebSocket helpers for Responses, GPT-Live and Realtime APIs — craigsdennis · 2026-11-04