Directions Brain

2026-05-06-brain-os-strategy-brainstorm

4 min read

Brain / Context-OS Strategy Brainstorm

Source: raw/meetings/Meeting (2026-05-06 16.30).txt

Language note: original is a Hebrew ASR transcript. Roughly the first ~10 minutes are off-topic chitchat (football, Zoom vs Teams) and have been excluded from this summary per repo convention.

Summary

A ~88-minute brainstorm between saar-arbel, guy-barkat, and nizan-shifman on whether the Brain — an LLM-maintained company context OS — is a standalone enterprise product or just plumbing for agents. Saar showed the working prototype, Nizan demonstrated end-to-end Harmony workflows powered by it, and Guy played designated red team. The meeting did not produce binding decisions; instead it sharpened the central tension (vitamin-vs-painkiller-framing), mapped the competitive landscape (glean, sierra, wonderful), and surfaced concrete research and prep for a pitch to yoni in 2–3 weeks.

Key takeaways

  • The product splits into two pillars — the Operating System / wiki layer, and the proactive agent layer that hands work from brain to agents. See context-os-brain.
  • The strongest unlock is cross-team context an individual cannot reach, not personal-context productivity. Most day-to-day tasks live inside one person’s existing context, so the brain only saves a few tokens of copy-paste — that is a vitamin, not a painkiller. The painkiller emerges when leaders need to align dozens of ICs or when an IC has to act on knowledge that lives in another team’s room. See top-down-bottom-up-alignment and vitamin-vs-painkiller-framing.
  • Agent-native day-1 is non-negotiable if the bet on widespread agent adoption holds. The framing is “companion” — make your agents better at doing your work — not “do your work for you.” See agent-native-go-to-market.
  • Competitive landscape is real and getting more crowded. glean is the closest analogue and claims to have solved permissions at query time (verify). sierra is the most credible enterprise player (Bret Taylor, Fortune-500 via FDEs). wonderful proves the segment can scale to ~$3B in a year. cursor/Claude/ChatGPT eat chunks of the use-case from below.
  • The Brain prototype is “bosr” (immature) but the underlying problem is real. Saar is confident the engineering scaling questions are solvable; the open question is the value-prop pitch.
  • First-mover matters but is not enough. A pitch needs a sharper “next 2 steps beyond Operating System” story — Guy explicitly says he does not yet have conviction on what those are.

Decisions

  • (no firm decisions — exploratory brainstorm. The closest thing to a commitment is the agreement to pitch yoni in 2–3 weeks; not promoted to a decision-slug yet.)

Action items

  • saar-arbel — provision guy-barkat a Monday Personal account and walk him through Brain + NoteTaker MCP setup. Due: next session.
  • guy-barkat — install the Monday NoteTaker MCP in his own Cursor and start his own Brain context. Due: before the next session.
  • Team — competitive deep-dive on glean (especially the permissions architecture) and sierra (try to get a demo via the eToro angle). Due: before the Yoni pitch.
  • Team — schedule the yoni pitch for ~2–3 weeks out (late May 2026); pitch must answer the “next two steps beyond OS” question before booking.

Open questions

  • Vitamin or painkiller? Guy rates 4–5/10, Nizan rates 9/10. Unresolved — see vitamin-vs-painkiller-framing.
  • What are the “next two steps beyond Operating System”? The team has no shared answer yet.
  • How does Glean’s permissions model actually work at query time? Direct product testing required.
  • How does the Brain handle context conflicts (e.g. KPI = 100 in one meeting, KPI = 80 in another)? Saar says it currently does either an append or an edit but they have not stress-tested this.
  • Privacy / surveillance dynamics — how to deploy in companies without people changing behavior to avoid being recorded.
  • Agent-to-agent workflows — if every employee has their own agent, how does the Brain mediate between them rather than just enhance one agent?

Graph View

Backlinks