The Moving Frontier — On AI, business, and a changing world. By Dhruv Dhingra.

What AI-Native Companies Actually Compete On

By Dhruv Dhingra


THE MOVING FRONTIER · NOVEMBER 2026 RETROSPECTIVE

Model intelligence is becoming a starting point

The temptation in every technology wave is to mistake the new capability for a complete business model. In AI, that mistake is especially easy: a remarkable model can make a new company feel inevitable before the surrounding product has been proven.

Yet customers do not generally want a model. They want a job completed, a decision improved, or a capability they did not previously have. As more companies can access increasingly capable systems, the existence of intelligence alone becomes a less persuasive source of differentiation.

The enduring question is what a company can build around that intelligence that customers would struggle to replace.

Four sources of durable advantage

The first is workflow ownership. Does the product sit inside an important, recurring job? Does it understand the exceptions, approvals, and evidence that make the work real? A demo can be copied faster than a deeply integrated operating rhythm.

The second is trust infrastructure. Permissions, evaluation, reversibility, traceability, and dependable failure handling are not exciting slogans. They determine whether a customer will delegate higher-value work.

The third is a legitimate learning loop. Does the company receive useful feedback from real outcomes, with the right permissions, and can it turn those signals into a better product? A pile of data does not compound automatically.

The fourth is distribution and organizational design. Can the company reach the buyer, fit the user’s habits, and support deployment? Can its own teams change their workflows as quickly as the technology changes?

The false moats

There are several apparent advantages I would treat cautiously. A thin wrapper around a general model can be valuable as an initial product, but may be easy to reproduce. A claim of broad autonomy can sound powerful while obscuring the costs of supervision. Proprietary data may be less useful than expected if it lacks quality labels or legal reuse rights.

Even vertical specialization is not a moat by itself. Domain language on a landing page is different from knowing how real work is evaluated, handed off, and signed off.

A durable business must connect its specialization to outcomes that matter and operational knowledge others find difficult to acquire.

The tension between speed and reliability

AI-native companies have an opportunity to build and learn at a radically different pace. But there is a trap in equating speed with constant change. Customers often rely on enterprise software precisely because its behavior is predictable.

The better operating principle is to move quickly on hypotheses and deliberately on obligations. Prototypes can change daily. A production system entrusted with consequential work needs stable controls, regression tests, and clear communication about changes.

That is not a contradiction. It is a separation between the frontier of discovery and the contract of reliability.

The long view

My bet is that the strongest AI-native companies will look less like collections of spectacular features and more like systems of work that become incrementally more capable, trustworthy, and indispensable.

Technological change keeps moving the frontier. Competitive advantage comes from knowing which parts of that frontier matter to a customer—and building the organization that can turn them into enduring value.

Editorial note: This is a retrospective essay written in November 2026. November 2026 describes the period explored, not its original publication date.


← All essays