DeepSeek's SuperAgent architecture questioned over risky multi-worker event design
teortaxesTex · x · 2026-11-02
After DeepSeek published its SuperAgent technical article, a developer questioned a key design choice: when a Thread is subscribed by multiple Workers simultaneously, subsequent events are broadcast to all Workers, letting them infer each other's actions from the shared event stream and 'resolve conflicts on their own' — with no point-to-point communication. Critics argue this invites collisions, since fully isolated Workers can only guess at each other's intent.
Research commentator teortaxesTex agreed, calling SuperAgent 'a hot mess' as written, and said it confirms his long-held view of DeepSeek: the team has evolved internal crutches that cap the costs of missing capabilities (chiefly hallucinations) in their own scenarios, rather than making the model generally robust — and the harness reflects the same philosophy.
More from coding & agent
- LlamaIndex launches Extract Max Effort, hitting 61% perfect extraction on ExtractBench Pro — llama_index · 2026-11-02
- LlamaIndex claims Extract Max Effort beats frontier agent harnesses on accuracy and cost — llama_index · 2026-11-02
- Founder shares the exact path to land a POC at a 60,000-employee multinational insurer — joecole · 2026-11-02
- Claude codes an animated pixel-art world in pure TypeScript — full prompt shared — AIandDesign · 2026-11-02
- Opus 5.5 builds a live painting studio from NVIDIA's free SimReady assets — petewoodbridge · 2026-11-02
- AI agents write 800% more code for 30% better delivery: a 2027 AI engineer roadmap — MaryamMiradi · 2026-11-02