01-projects/personal-brand

Edge to opportunity: which ownership bet, and when

2026-10-12·decision-memo·status: pending-founder·source: founder iMessage 2026-10-11 22:46 ET (the question), then "Yes please" to this memo; Karpathy X post 2109361546505966046
decisionambitionownershipvibe-manufacturingfamily-venturecredibilitykarpathy

Decision needed: which ownership bet gets your side hours between now and Nov 30: vibe manufacturing, the family revenue cycle management (RCM) venture, or neither yet (credibility first). Ray recommends neither yet, with a dated pick on Nov 30.

The question

You asked (iMessage 2026-10-11 22:46 ET) how to turn the edge into an economic opportunity "outside of moving to Silicon Valley." Two paths already run: phData pays for the edge (Ray's read: as the person who explains frontier agents to client executives), and the Nov-30 credibility targets build demand. This memo covers the third path, ownership.

The reframe: the problem is the asset

Karpathy's napkin funnel: about 75% of people barely use artificial intelligence (AI) at all, 20% use free tiers, 0.2% use frontier tools professionally in math and code, and about 5,000 have internal frontier access. His first driver of impact: "you need a reservoir of big, difficult problems that you really care about."

Ray's read: you sit near the top of the 0.2%, and the edge shrinks as tools reach the 20% tier. An edge with no problem attached pays whoever owns the problem, and today that owner is phData. So the asset to acquire is the problem. The edge is the tool you use to find it.

The test for any candidate problem comes from Karpathy's jaggedness point (models are strongest where work is digital and checkable): can you build an automatic verifier for this work?

Options

Quotes and 2026 dates mark your statements. Every other cell is Ray's scoring or research.

A. Vibe manufacturing B. Family RCM venture C. Neither yet: credibility first
FOUR-WALLS (mission sentence, whole lane, build-then-teach, upside stake; you agreed 07-24) Real pull: "much more up my alley" (10-05). No mission sentence yet. Lane, teach, stake: yes Stake, shared lane: yes. Pull: weak (Ray's read). On 10-05 you said the healthcare/life sciences niche "just doesn't inspire me as much", about your moonshot area, not this venture. On RCM itself you said "not sure that's the issue we want to tackle together" (09-16) Teach: yes. Stake: no, phData keeps the upside
Automatic verifier Partial. Designs are code, but your validate step is "assemble and use in projected application" (10-05): physical and slow Strong for payment matching (a payment matches the claim to the cent or it does not). Weaker for denial handling, which needs judgment Not applicable. This builds buyers, not a product
Buyer gap (20%-tier owners) Weak. You framed it as building things for your family first (10-05). Outside buyers would be storefront shoppers, who buy objects, not your edge Strong, but held by incumbents. Practice owners are reachable through a provider-side family network (09-16). One venture-backed company already names applied behavior analysis and substance-use billing, two of the family's specialties, and only 1 of 24 behavioral-health record vendors publishes an offer to run billing on a rival's system (Ray research, 10-01 and 10-03) Strong. Vice President (VP) and Director data buyers, through phData (09-27)
phData conflict Low. A product, not a service. Ray flags a written carve-out before the first paid sale; you said that would be "a letter from the CEO" (10-05) Medium (Ray's read). A billing product can drift into data and AI implementation work for practices, which is phData's service: "cannot sell the same services" (10-02) None. "Thought leadership is fine" (10-02)
Horizon and Nov-30 fit Months to a first product. Weak fit: the build-in-public pieces are scoped to Snowflake About 3 years of residency left for one brother-in-law (09-16); Ray reads a 3-6 year runway. Medium fit: healthcare billing is one example domain under a locked topic in your Ship 30 writing program (10-07) 7 weeks. Full fit: it is the target sheet

Recommendation: C, neither yet, with a dated pick on Nov 30

  1. It is your own sequence. On 2026-09-29 you set "credibility → capital → talent → moonshot," with the moonshot still to be chosen. A stake now skips two steps.
  2. It works the real constraint. Ray's 2026-07-31 read (you have not confirmed it): the constraint is demand generation, not capital. Every ownership bet needs buyers; C builds them now.
  3. The hours are taken. The target sheet budgets about 5 hours a week, and its tallies read 0 as of its last update (2026-10-04). The storefront and Scribble Works already share your other side time (10-07).
  4. Neither A nor B passes yet. A lacks an outside buyer; B lacks your pull. Ray's read: a pick that fails a wall now will not hold.

C does not mean waiting: each buyer-facing conversation doubles as a problem interview. Vibe manufacturing stays a print-on-demand layer on the storefront plus the home lab (your 10-05 and 10-07 statements). Skip any conversation where the person has asked you for contract work; that edges toward services. The family venture stays parked behind the trigger below.

Cheapest next step this week

Book the Snowflake sales reps (#4 on [[2026-09-29-conversations-to-have]]). Ray's read: they talk to VP-level buyers every week. Keep their listed ask (what content they would forward to a VP of Data) and add one question: "Which recurring work do your data buyers say they would hand to an agent if they could check the output automatically?" Ray logs each answer as a pattern only (no names, no client details) and scores it on the verifier test. These answers come through a phData partner relationship, so use them for writing and phData pipeline first. If one later seeds an outside product, log where it came from and include it in the CEO-letter carve-out. Booking takes about 10 minutes and the call about an hour; holding it moves the networking row off 0/8.

What would change this