Magic employee reveals tiny-team culture: days-long RL runs with no compaction, cross-session agent memory
magicailabs · x · 2026-09-09
A Magic employee describes a very different lab culture: the tiny team sits in one room and everyone just does whatever is most important. He hints at technical breakthroughs — RL runs lasting days without compaction, and agents that learn from you across all sessions so you never repeat yourself. He says he's excited for the product experiences they're building.
More from coding & agent
- HeyReach hits $17M ARR running LinkedIn outreach agents via MCP for 7,000+ teams — thetripathi58 · 2026-09-09
- Final step of the AI agent playbook: run error analysis first, let eval metrics emerge naturally — PawelHuryn · 2026-09-09
- Agent-building steps 5 & 4: orchestrate with a manager, skip topology research; .md files often suffice for memory — PawelHuryn · 2026-09-09
- Agent orchestration tips: a manager or simple sequence beats complex topologies — PawelHuryn · 2026-09-09
- Built the same CRM four times in an hour: Lovable vs Google AI Studio vs Claude Design vs Claude Code — PawelHuryn · 2026-09-09
- Agent-building steps 3 & 4: turn fixed procedures into tools for determinism; .md files often enough for memory — PawelHuryn · 2026-09-09