From a certification layer under others' verticals to a Vertical-AI platform we build

Prior Position

Entry 0002 settled the category: AetherHeal is certification / trust / proprietary-outcome-data infrastructure on which each specialist owns their own vertical clinical AI. Two commitments were load-bearing:

1. AetherHeal does not author the verticals — domain specialists do, and AetherHeal is the trust layer they sit on. "A dermatologist cannot credibly author internal-medicine AI; the internal-medicine specialist does."
2. The moat is the trust mechanism + the outcome dataset — explicitly NOT the AI stack. This was not incidental; it was the spine of 0002's argument. Because post-Opus-4.5 capability had commoditized the AI/orchestration layer, the defensible thing had to be trust and proprietary data, not the models.

Revenue followed from this: the B2C navigation loop — the patient pays for navigation and is protected, and the platform earns the same fee whichever verified hospital is chosen. "The product is no longer the connection."

Current Position

AetherHeal — now 주식회사 에테르힐글로벌 — is a two-layer medical Vertical-AI platform that we build:

• ① Horizontal infrastructure, common to every clinic and every specialty: Dockie-talkie (realtime AI medical interpretation) + Dockie-talkie Clinical Copilot (realtime background clinical reasoning).
• ② Specialty Vertical-AI apps running on that infrastructure — DermatoScan (dermatology pigment, my own first-party reference) and WoundScan (plastic-surgery wound care, co-developed with 울산대 성형외과 as the clinical authority, with AetherHeal as the development partner) — expanding into an 입점 (tenancy) ecosystem where third-party medical Vertical-AI developers plug into the shared infrastructure.

The revenue model is B2B SaaS first (Vertical AI to hospitals at roughly ₩300k–1M per clinic per month; horizontal-infra subscription), with the B2C navigation fee now one line among several, plus future ecosystem revenue. The primary customer is no longer only the patient — it is the founding-partner hospital subscribing to the AI.

The moat moves with the category. It is now the 2-layer structure itself: a horizontal infra (interpretation + reasoning) shared across all specialties that no single-domain AI can replicate, plus verified first-party verticals, the six-hospital founding network and the 울산대 산학 channel, and the augment-not-replace trust structure. Where 0002 insisted the AI stack was not the moat, v12 makes the AI platform a central part of it.

Causal Update

(Draft reconstruction — the reconciliation below is the crux, and it is mine to confirm or correct.)

The reversal that has to be stated plainly is against 0002's own causal spine. 0002 reasoned: individual AI/orchestration is commoditized → therefore the moat cannot be the AI stack → therefore trust + outcome data. v12 does not deny that individual models are commoditized. It relocates the moat from the model to the platform structure — the horizontal infrastructure shared across every 외래 that a single-domain competitor cannot assemble, and the network effects of an 입점 ecosystem built on verified first-party references. A commodity model inside a non-commodity 2-layer platform is the claim. If that distinction does not hold — if the platform proves as replicable as the models — then this entry is wrong and 0002's narrower "trust + data only" moat was right.

Other drivers, also mine to confirm:

• The Ulsan startup-hub, 울산대 산학 co-development, and the government non-dilutive funding track call for a concrete, buildable product company with a CTO/R&D track — something an abstract certification layer could neither be funded nor sold as.
• B2B SaaS to hospitals provides near-term revenue that the slower B2C navigation loop did not, while the six founding-partner hospitals make that channel real now rather than later.

What does not change: the wedge is still 비급여 inbound foreign patients (where the trust deficit is largest), and the incentive-alignment guard of 0001 and 0002 is explicitly preserved — every revenue line is flat or fixed, independent of procedure price or type, so selling AI to a hospital does not recreate the referral-kickback structure those entries were built to abolish. What reverses is the category and the location of the moat: from "we certify others' verticals, and the AI is not the moat" to "we build the reference verticals and the infrastructure they all share, and the platform is the moat."

This entry is part of The Founder's Errata. Entries are append-only and are never deleted or rewritten. Corrections to this entry, if any, will appear as new entries that reference this one by ID.