Huddle — 2026-03-05

Decisions

Key Context

• Premera theory: They likely need internal committee approval / budget sign-off. Meetings booked weeks out. Not ghosting — just slow enterprise process. The "two-person startup credibility gap" is real but not fatal. Pinged Thursday/Wednesday last week, no response yet.
• McBee: Mikhael reached back out then ghosted again. Confusing.
• Rodrigo also never responded.
• Michael's 3-pillar vision:
  1. FDE/implementation work for health plans (Premera model) — generates services revenue, develops relationships, sets technical standards
  2. Acquire/techify a services group (like PrizeMD or McBee) and consolidate — the Ensemble model. Hire someone like Cassie to manage, go to insurers and offer to take over their UR groups.
  3. SaaS — sell the organizational OS to mid-size orgs like Dr. Bauman's group. Slow growth but sticky.
• Thomas's take: Each pillar is a standalone business; a roll-up is possible. But doesn't want to be the one finding/buying services companies. Would need someone like Dan for that.
• Fallback plan: If BD fails, Thomas would be open to being Premera's AI director (work with Colt directly). Dan joining calls could add credibility to get through contracting — "just sitting there on the call, not doing anything" but it legitimizes the team.
• Dan/Ford/Gary interest: Michael reads between the lines that Dan wants to work together, Ford too, Gary is impressed. Something to work with even if Premera stalls.
• Pipeline gap: No one else in the pipeline beyond Premera and McBee. Not pushing intros because there's nothing to push on yet. Need Premera to close to unlock everything (fundraise, more clients, credibility).
• GPT deep research: Thomas notes it's still unparalleled — 30 min runs producing massive plans. Claude/Google not close on that specific use case.
• Michael built: Ideas tab + Reddit-style forum for the OS dashboard (pushed during call)
• Thomas built: Portal agents deployed, graph visualization improvements ("poker chip" style)
• Daily agent didn't run this morning — needs debugging
• Google Meet recorder: Needs the shared API key Michael pushed. Ended call to test it live.
• OS philosophy: Michael's vision — each dashboard surface (hot board, graphs, forum) should eventually be owned by a specific person (e.g., chief of staff). Not Thomas/Michael's job to maintain everything long-term.
• Forgot to start Google transcription for this call. Will remember next time.

Action Items

---

Huddle #2 — 2026-03-05 (12:43-1:18 PM)

Decisions

Key Context

• Zen postmortem: Their entire value prop (scoping data transfer, interoperability) is now trivial with LLMs. Tech is ~20 years old. No sample ADT/ORU messages provided for sandbox testing. If they'd brought a client into this, it would've gone badly. Good signal on where the interop vendor market is — Zen isn't off-market, this is likely how they all work. Just not needed anymore.
• Thomas on interop vendors: "These vendors fucking suck. Anyone can build this technology. The hard part is business development."
• Accidental Stripe charge: Thomas forgot to cancel 1-month free trial on their own app — $250 charge. Money goes to Mercury but Stripe takes processing fee.
• Tom Gilligan / Westview call at 2pm: Thomas noted $20M minimum investment — not a fundraise call. Prep already done.
• AI / singularity discussion: Both think we're at or near AGI. Thomas cited Dario's chessboard analogy (square 44 of 64 — exponential progress ahead). AI R&D automation is real — even if human input flatlines, tech output is exponential. The constraint is the limited number of people who know how to use the systems.
• Indian dev agencies: Thomas researched publicly traded Indian outsourcing firms — thinks they're clear shorts. But also notes they'll learn AI tools too (looked at an AI playbook from one — "actually really good, we should run it"). Playing field leveled but US relationship advantage matters, especially with immigration headwinds.
• Claude Code stickiness: Thomas giving it max permissions (browser, email, all MCP tools). Integration + smoothness = doesn't want to switch. Michael still uses ChatGPT for research but builds nothing in it. Both agree GPT outputs have deteriorated — repetitive, not succinct.
• Epic won't figure out AI — Michael's explicit bet. Thomas mostly agrees.
• Anthropic releases: Thomas tracking extended Opus runs and new Claude Code capabilities. Speed of releases is unprecedented.

Action Items