Directions Brain

2026-05-31-directions-connect-businesses-and-brain-scaling

8 min read

Directions sync — Connect-Businesses thesis + Brain-scaling concerns

Source: raw/meetings/Directions starting the week (2026-05-31 20.00).txt

Language note: original is a Hebrew ASR transcript, ~90 minutes. The opening ~5 minutes (football championship celebration), final ~10 minutes (Guy’s logistics: nanny costs, new house construction timeline), and assorted personal asides (Dor returning from Lebanon, Beer Sheva Hapoel anecdote) are excluded from this summary.

Summary

Sunday evening sync between saar-arbel, guy-barkat, and nizan-shifman. Two main threads. First, guy-barkat introduced a brand-new exploratory direction — a multi-vertical, “Plaid-but-everything” unified-API platform that monetizes distribution by letting any app plug in any service via standardized contracts + bundled compliance. He triggered on three signals: etoro’s in-flight partnerships with papaya-global and deutsche-bank, plus same-morning Israeli news that victory is selling apartments and rami-levy / osher-ad are selling cars. The thesis is captured as connect-businesses-multi-vertical-api — a non-Brain direction, sitting alongside vertical-use-case-led-brain and prediction-markets-regulatory-play as the team’s third open exploration. Second, the team revisited context-os-brain / Brain scalability. Guy argued the Brain is much harder to make work at organization-scale than the team has been treating it — partial data is worse than none, conflict resolution is non-trivial, and the system needs an “agent harness” (proactive validation, human-in-the-loop, conflict surfacing) before it can be trusted on a real org’s recordings. The pivot Guy reached for himself: this difficulty is the moat — large enterprises will not be able to build this in-house, which is exactly what makes it worth selling. Saar reinforced that the Brain shape is vertical-dependent (different verticals require different data models). Nizan defended the Holy-Grail framing (alignment + transparency + speed + source-of-truth) and reported he has personally “cracked AI” with the Brain in his daily flow. The meeting closed with concrete adoption logistics for Guy + Nizan and a Tuesday-evening follow-up.

Key takeaways

  • NEW DIRECTION — Connect Businesses (multi-vertical unified API). The thesis details, why-now, and open questions are written up at length on connect-businesses-multi-vertical-api. Headline: turn the N×N integration cost between “platforms with distribution” and “companies with capability” into N — by bundling unified API + standardized commercial/legal contracts + compliance + customer-support routing. Reference templates: plaid (banking, $15B+) and snaptrade (trading, smaller). Sweet-spot ICP per the discussion: low-tech / medium-tech distribution platforms that can’t run a 6-vendor compliance + integration process themselves.
  • Anchor evidence for Connect Businesses (Israeli low-tech retail trend). victory selling apartments through its retail channel; rami-levy (or osher-ad) selling Kia Sportage at ~140K NIS through a supermarket app; both stories broke the same morning. Plus international precedent: wolt expanded restaurants → groceries; grab (Thailand SEA) went broader to taxis + travel; Costco selling Porsches. Pattern: distribution-rich, low-tech platforms are reaching for category expansion now because cost of building / integrating dropped.
  • Brain scaling is harder than the team has been treating it (Guy). Partial data is worse than no data — the Brain confidently extrapolates from incomplete recordings and risks misaligning the whole org around its summary. Conflict resolution is unsolved (who’s right when saar-arbel said X in a call but nizan-shifman said Y in another? Who outranks whom? On paper vs in practice?). The Brain needs an “agent harness” similar to coding-agent harnesses: proactive validation, human-in-the-loop on uncertain inferences, explicit conflict surfacing. Naive ingest at scale will produce a worse-than-nothing source-of-truth.
  • Guy’s flip on Brain difficulty: the difficulty IS the moat. Mid-conversation Guy realized that because the Brain is so hard to make work at scale, even hi-tech enterprises won’t be able to build it in-house — which is exactly what makes it a sellable product. Reinforces vertical-use-case-led-brain direction (vertical scope shrinks the problem to something a startup can actually solve while incumbents can’t replicate it).
  • Vertical determines data model (Saar reinforces). A construction-vertical Brain wouldn’t have companies and entities — it’d have patients-equivalent things, treatments, sites. An eye-clinic Brain looks nothing like an enterprise-CRM Brain. The “vertical-use-case-led” frame is what makes the data-modeling tractable; a generic Brain has too many conflicting demands.
  • Source-of-truth debate. Nizan: the Brain is the org’s source of truth — alignment + transparency + speed all flow from it, which is the Holy Grail enterprises want from AI. Guy: it’s not source-of-truth because it’s not deterministic; at best it’s “high-quality aggregation that may be wrong.” Resolution they reached: it’s a useful source-of-truth when paired with conflict-surfacing UX (the agent says “Nizan changed the OKR — please confirm” rather than silently overwriting). Pattern matches a question Saar’s cousin at Amazon asked in his Brain-pitch dry-run: how do you handle conflicts?
  • Nizan’s “I’ve cracked AI” report. Reports he’s now in a sustained mode where the Brain holds his full context fresh, he wakes up and just runs Opus 4.8 on his open work, and the bottleneck isn’t AI quality anymore — it’s escaping the slop of “this looks like everyone else’s AI output.” Guy’s response: “I personally still need an assistant that knows everything I’m running across all my surfaces — Obsidian + Monday + ChatGPT — I can’t keep up.” Confirms the personal-productivity case at the user-level even while the company-level value is still being articulated.
  • Brain prototype usage update. Saar built a Claude project (in Monday’s Code-Workspaces) layered on top of the Brain content + alon-huri’s LinkedIn posts, with a 10-criteria scorecard ranking strategic directions; demoed live in this meeting. Nizan’s cousin at Amazon presented the Brain to colleagues, became “champion of the week,” and is giving a 40-person internal talk next week.
  • Adoption gap (action-bearing). guy-barkat still hasn’t installed or used the Brain; Nizan called him and Saar out for it (“how have you not had a setup session yet”). saar-arbel will pair with Guy this week. Guy can’t use monday NoteTaker at etoro (default opt-out at the org level) — substitute is Spinach (already installed for Guy) plus Monday’s NoteTaker MCP wired to the Brain via Claude Code. Saar will provision Guy + Nizan as collaborators on the Brain GitHub repo (private) once they send him their GitHub-connected emails.
  • Tomer Shalom / Toto chairman update. tomer-shalom’s assistant reached out to schedule the Toto-chairman conversation this week as high-priority. Guy is taking it. Triage gate for prediction-markets-regulatory-play.
  • Wonderful / FDE-model debate (sidebar). Guy: wonderful is real (jumped from 200M ARR in two quarters), well-oiled, surrounding Monday on enterprise. Nizan: real but Israbluff — the FDE-heavy model hasn’t been through a market cycle yet. Saar surfaced render-ai’s LinkedIn post arguing FDE doesn’t scale and they’re building platform-from-day-one. Resolution: model is unproven through a downturn, but in the meantime Wonderful wins.

Decisions

Action items

  • saar-arbel — provision guy-barkat and nizan-shifman as collaborators on the Brain GitHub repo (the private personal repo, not Monday’s). Awaiting their GitHub emails. Source: this meeting.
  • saar-arbel + guy-barkat — schedule a 1-hour pair session this week to install Spinach + Monday NoteTaker MCP + connect Guy’s recordings into the Brain. Source: this meeting.
  • guy-barkat — start recording his eToro meetings (Spinach is already installed) so the Brain has data to ingest; without recordings the Brain has nothing to compound on. Source: this meeting.
  • guy-barkat — take the Toto-chairman call this week (assistant reached out, high-priority). Source: this meeting.
  • Team — for the next sync, each member arrives with insights pulled from the Brain (use the Claude-project ideation surface Saar demoed). Source: this meeting.
  • Team — discuss whether to move to a half-day in-person working session on a Friday or upgrade to 3×/week cadence; deferred to next sync. Source: this meeting.
  • Next sync: Tuesday evening, 2026-06-02. Source: this meeting.

Open questions

  • Defensibility for the Connect Businesses thesis. Saar pressed: how is this not a services consultancy? Guy’s cut: speed-to-market for low-tech / medium-tech ICPs is the moat. Unverified.
  • Which of the three open directions does the team commit to (vertical-use-case-led-brain, prediction-markets-regulatory-play, connect-businesses-multi-vertical-api) — and on what evidence? No process for converging yet.
  • What’s the agent harness shape for the Brain? Conflict resolution, proactive validation, human-in-the-loop on uncertain inferences. Not yet designed.
  • What’s the org-level Brain pricing model? Guy keeps returning to “we don’t know what value to attribute, so we don’t know what to charge.” Holy-Grail framing (Nizan) doesn’t yield a number.
  • Onboarding behavior with non-tech-native orgs. Will an enterprise be willing to feed the Brain everything (recordings, Slack, email, code) on day 1, or will they ask for sandboxed silos? eToro-shape orgs may be different from Monday-shape orgs.
  • Sheba / ari-leshno partnership — still overdue per 2026-05-19-saar-leshno-glaucoma-data-tracking action item; not addressed in this sync.

Graph View

Backlinks