AI-agent bans need a channel price

On 11 November 2026, United Airlines said it would restrict unauthorized personal AI agents across its own digital platforms, travel agencies and aggregators. United gave concrete reasons: privacy, security, operational reliability and regulatory obligations.

Tomasz Tunguz's economic reading goes one step further. Buyer agents can compare options without being persuaded by the screen that sells upgrades, warranties or sponsored placement. My claim: a merchant can block that machine channel, but the pricing conflict will return as soon as customers demand it.

United's filing shows why the channel matters

United's 2025 annual filing records $4.8 billion of ancillary fees within passenger revenue. It defines these as baggage, premium-seat, inflight-amenity and other ticket-related fees. The same filing reports $4.716 billion of operating income.

Those two numbers being close does not make ancillary fees pure profit, nor prove they caused the policy. The filing does show that extras are large enough to be part of any serious channel discussion. An agent trained to minimize the traveller's total cost may skip the prompts that make a direct interface commercially attractive.

An authorized agent channel needs commercial terms

United's policy release is about unauthorized agents. The unresolved question is what authorized access would require.

If I were reviewing such a channel, I would ask for five declared fields: allowed actions, agent identity, customer authority, liability for mistakes and the commercial terms for search, booking and servicing. This is editorial judgement, not a United design or a pattern I have deployed at Trigent.

The security layer can authenticate the machine and constrain its actions. It cannot decide whether the channel pays per transaction, shares servicing cost, preserves loyalty benefits or exposes optional products in a machine-readable way. Leaving that question implicit invites a cycle of scraping, blocking and exceptions.

What's in it for you

• Treat an agent API as a distribution channel with its own economics.
• Separate stated safety reasons from inferred margin pressure.
• Price the authorized path before users and agents invent an unauthorized one.

I did not test a United booking flow, and the inspected sources expose no approved-agent tariff or public agent API. Tunguz's piece is valuable because it names the commercial tension; United's own documents supply the boundary.

An agent channel is ready when its permissions and its price are both explicit.