What it actually looks like when a PM checks in every document, every metric, every customer call into a shared repo - and the whole team queries it themselves.
Hey Aakash! Sounds like a fascinating episode. I'm really curious—what's one key takeaway you both discussed about building a Team OS that might surprise most founders? Always looking for fresh insights!
Great episode, Aakash. Hannah's Team OS is the best documentation I've seen of a mature enablement layer and the 3%-context / 97%-reasoning breakdown is the number every PM team should be chasing. A note for the PMs reading this who are about to copy the pattern: the routing, plan mode, subagent temp-file rule, and 70% skill auto-invocation are all Claude Code runtime behaviors. If your company standardizes on a different runtime (Goose or Codex for example), you are not porting Hannah's Team OS, you are redesigning it from the primitives up. That makes the runtime choice an architectural decision with real lock-in, not a cosmetic one. I wrote the full comparison here: https://www.iqsource.ai/en/blog/harness-is-the-runtime/
Question about the written summary (reading it before I watch): when the summary says "3% of the context window", is that Opus' 1M token context window (also Sonnet's limit on Claude Platform or via APIs), or Sonnet's 200k token context window? I'm impressed either way, but one is even more impressive than the other! :)
Hey Aakash! Sounds like a fascinating episode. I'm really curious—what's one key takeaway you both discussed about building a Team OS that might surprise most founders? Always looking for fresh insights!
Reduce onboarding time like 10x!
Great episode, Aakash. Hannah's Team OS is the best documentation I've seen of a mature enablement layer and the 3%-context / 97%-reasoning breakdown is the number every PM team should be chasing. A note for the PMs reading this who are about to copy the pattern: the routing, plan mode, subagent temp-file rule, and 70% skill auto-invocation are all Claude Code runtime behaviors. If your company standardizes on a different runtime (Goose or Codex for example), you are not porting Hannah's Team OS, you are redesigning it from the primitives up. That makes the runtime choice an architectural decision with real lock-in, not a cosmetic one. I wrote the full comparison here: https://www.iqsource.ai/en/blog/harness-is-the-runtime/
Question about the written summary (reading it before I watch): when the summary says "3% of the context window", is that Opus' 1M token context window (also Sonnet's limit on Claude Platform or via APIs), or Sonnet's 200k token context window? I'm impressed either way, but one is even more impressive than the other! :)
https://knowledgenetworks.substack.com/p/david-cross-explains-ai-native-thinking?r=7ykchg&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true