Every cloned its Slack with 51 AI coworkers trained on chat history via DSPy GEPA
lateinteraction · x · 2026-11-07
Every's @hammermt built a simulated Slack with 51 AI coworkers, with prompts trained from historical Slack messages using DSPy GEPA — Every's editor calls it "we've all been cloned."
The screenshot shows the optimized 3,000+ char system prompt for a digital "Dan" (Dan Shipper): default replies are 1–6 word reactions, idea approvals get brief praise, tasks are delegated by @-tagging teammates, and longer replies only when genuinely explaining a decision. The author notes reading the prompts feels "like reading someone's mind."
The project doubles as a testing ground: simulated coworkers can reply to each other, and even to the real person they're simulating — a novel setup for simulating organizational behavior.
Related event: Every Clones Entire Team Into 51 AI Coworkers Trained on Slack History(3 posts)→
More from AGI Musings
- Devin Guillory essay: the public powers AI with data but shares none of the rewards — databoydg · 2026-11-07
- LLM-generated reviews hit ICLR/NeurIPS papers; CMU essay warns of review monoculture — brwilder · 2026-11-07
- Steve Jobs on obsolescence: no Principia in this field, just sedimentary layers — phroo · 2026-11-07
- Human-level AI may arrive without automating most jobs, argues viral thread — gleech · 2026-11-07
- Can AI ever produce a band like Tool? Recombination vs lived experience — krishnabrodhi · 2026-11-07
- Hardware x AI x Bio intersection remains a startup white space — jungofthewon · 2026-11-07