"Stop naming your agents": dev argues multi-agent teams are just one LLM in costumes
ziv_ravid · x · 2026-10-26
Developer zivravid, quoting shgidip, criticizes the trendy "agent team" narrative: an agent is still the same LLM no matter what name or persona you give it. Role prompts, md files, context, and tools effectively turn it into a harness, not a separate agent. Some splits work — coder, code reviewer, bug hunter — but he mocks a marketing product with 10 named agents doing near-identical tasks (turn needs into briefs, copywriting, content). Takeaway: design honest harness-based workflows instead of dressing up one LLM as a team.
More from coding & agent
- Compound rewrites Excel, PowerPoint and Word engines in Rust for AI agents — peterjliu · 2026-10-26
- Dev vibecodes open-source suite: guided LoRA trainer, dataset captioner, image tournament tool — Speedyrulz · 2026-10-26
- Thorsten Ball publishes long-form guide on how he works with coding agents — vivekhaldar · 2026-10-26
- MCP explained: the open protocol standardizing how AI agents connect to tools — robertonappe · 2026-10-26
- Seeed Studio open-sources reBot-DevArm: a 6-DoF robotic arm with sub-0.2mm repeatability — lukas_m_ziegler · 2026-10-26
- Dev Predicts Personal Agents Will Replace Operating Systems, Generating UI on Demand — evielync · 2026-10-26