AI Product Building Future of AI Business Models

The system of work is the moat, not the model — the model is fungible underneath

A vertical app company wins by owning the surface where a company's work actually executes — data capture, workflow system of action, and governance — while each new model generation flows through underneath

@joeschmidtiv (Joe Schmidt IV, a16z) — Avoiding Death on the Yellow Brick Road · Jun 1, 2026 · 14 connections

In Joe Schmidt IV’s Oz framing, the labs own the model and the distribution, but a vertical app company can still win by owning the system of work — the surface where a company’s actual work executes and the data it generates gets captured. He is explicit: “The model is fungible underneath; the system of work is not.” The durable layer owns three things at once — data capture, the workflow system of action, and governance — and becomes the layer that integrates each new model generation and delivers it to the customer.

This is a workflow-specific cousin of Organizational shape is the emerging moat in AI — what AI cannot copy is the institution underneath: when models commoditize, the moat moves to what they can’t copy. It extends Context is the product, not the model from data and skills to the whole operating surface, and it is why Sell the work, not the tool — model improvements compound for services, against software holds and why the surviving defenses in LLMs selectively destroy vertical software moats — 5 fall, 5 hold — system-of-record status, regulatory lock-in — remain. The deterministic SaaS survives as the governance and coordination layer — determinism still rules is part of this system, not a competitor to it.

Connected Insights

References (5)

→ Context is the product, not the model → LLMs selectively destroy vertical software moats — 5 fall, 5 hold → Organizational shape is the emerging moat in AI — what AI cannot copy is the institution underneath → SaaS survives as the governance and coordination layer — determinism still rules → Sell the work, not the tool — model improvements compound for services, against software

Referenced by (9)

← Inference-time compute makes cost-per-outcome a choice — and that's the application layer's counterattack on the labs ← Every firm must build two capitals — human capital and token capital — and they compound together, not at each other's expense ← You can offload a task, or even a job, but you can never offload your learning ← The sovereignty test — can you swap out a generalist model without losing your 'company veteran' expertise? ← The learning loop becomes the firm's new IP — a hill-climbing machine that compounds unlike any other asset ← If a problem improves directly with raw model capability, the labs will take it ← Guardrails aren't just safety — they're what the customer is paying for ← System or tool? Ask whether the customer would still need you if a lab shipped a direct competitor ← Organizational shape is the emerging moat in AI — what AI cannot copy is the institution underneath