AI PM Hub Weekly #2: Your Agent Isn't Invited on the Flight
Article5 min read
Published · By Ofer Regev
Issue #2 · November 15, 2026
Everyone talked about agents this week, and not always kindly. United banned them from booking flights, an Anthropic study found that when everyone has their own agent the whole group does worse, and one startup let agents run three real businesses. Along the way we learned that a cheaper model per token can cost more per task, and that the chatbot recommending where to shop doesn't always know who's really selling.
1. United blocked your agent from booking flights
On November 11, United updated its contract of carriage to ban using AI agents to book, change or service a ticket, or to buy any service, including agents working through a travel agency. According to View from the Wing, the ban applies even when the customer approves every step, and United reserves the right to cancel reservations made that way. Agents can still show flights and prices, but you do the buying yourself. CEO Scott Kirby explained that he doesn't want to create another middleman.
The bottom line: Your agent can be excellent, but it operates in someone else's world. Whoever owns the customer relationship won't give it up easily, and that's true for the product you're building too.
Further reading: Skift · View from the Wing
2. Fable 5.2 is cheaper per token, and pricier per task
Anthropic launched Fable 5.2 on November 9 and cut the price by 20%: $8 per million input tokens and $40 per million output, down from $10 and $50. But on SemiAnalysis's GDPx leaderboard, where it ranks first at 16.4%, an average task costs $88.15, compared with $67.16 for Fable 5.1. That's a 31% increase. The explanation going around: the model writes far more tests and spins up subagents that check the work again.
The bottom line: Price per token isn't the price. What matters is what it costs to finish the job, and sometimes a smarter model simply does more work.
Further reading: Latent Space · GDPx leaderboard
