Vertical AI startups can still build durable competitive advantages against frontier model giants, but only if they own the end-to-end workflow rather than competing on model quality alone. That is the core thesis of Tiffany Luck, a partner at New Enterprise Associates who backs AI application-layer and B2B SaaS companies.

Luck joined NEA roughly three years ago after stints at Anthropic, Morgan Stanley, and GGV Capital (now Notable). Her investment focus sits squarely on what she calls the "last mile" of automation — the gap between what a horizontal model like Claude can do and what an enterprise actually needs delivered.

Why the last mile matters more than the model

Most general-purpose AI tools take users from zero to 80% of a task, Luck told Crunchbase News. The remaining 20% — re-forecasting financials, flagging trade-offs between burn rate and growth, unifying data across disparate sources — is where vertical startups earn their moat.

"Startups that build these purpose-built product flywheels — and use forward-deployed engineers to sit alongside users and identify workflow holes — build a moat that general models can't easily replicate through scale alone," Luck said.

She pointed to two NEA portfolio companies as examples. August handles legal due diligence. Samaya AI produces equity research reports. In both cases, the output is a finished document that looks like what a team of analysts would create. The enterprise buyer does not care which foundation model sits underneath.

Luck drew a parallel to her early career at Anthropic, where she spent years persuading CPG manufacturers that e-commerce was inevitable. Fortune 500 companies face a similar adoption friction with AI today, she said — the potential is clear, but integration into daily workflows remains difficult.

Interoperability as the next frontier

Luck expects the way people work to shift toward a model functioning as a de facto operating system — a command center that calls specialized applications. She compared it to how startups once used Slack as their primary interface.

In that world, a tool like Samaya would plug directly into a horizontal model's UI. The startup retains its proprietary knowledge graph and data. Execution happens inside the user's primary environment. Interoperability, not model differentiation, becomes the competitive axis at the application layer.

For regulated industries, Luck flagged accuracy, auditability, and cybersecurity as the deciding factors for enterprise buyers. She highlighted AIUC — Artificial Intelligence Underwriting Co. — which is assembling more than 100 CISOs to build a for-profit certification standard for AI agents. Companies like ElevenLabs seeking enterprise contracts may eventually need that kind of third-party validation.

Luck's broader message to founders: stop worrying about whether Anthropic or OpenAI will swallow the application layer. Own the workflow, deliver the artifact, and let the platform wars play out underneath.