Intelligence Track
22 June 2026
The Brief
The booking transaction is quietly becoming the least valuable link in the chain. At one end, a supplier can now assemble its own bundle using an intermediary's inventory as invisible back-end, the intermediary trading brand for raw volume; at the other, the moment of recommendation is drifting into conversational surfaces no aggregator owns, so the asset stops being the ranked list and becomes whether your inventory is machine-readable and present in the answer. The same migration runs underneath the design and PM pieces below: value is concentrating at the two ends — the owned customer relationship and the precisely-specified, retrievable substrate — and draining out of the generic middle that merely lists options. Defending that middle is no longer a position. The question worth sitting with is which end the team is actually building toward, because trying to hold the center is now the slowest way to lose it.
HeyDesigner · 22 Jun 2026
Distilling Soleio Cuervo's hiring lens, the argument is that the gap between competent and exceptional designers is posture rather than craft — communicating intent over showing work, confronting users while ideas are still cheap to change, asking whether a feature should exist at all, and refusing to delegate strategy and impact to PMs.
A playbook of 20-plus reusable agent loops for PMs, each defined less by its prompt than by an explicit stop condition and an independent grader — harden a PRD until two engineers reading separately would build the same thing (checked by a second model reading cold), cluster feedback until themes stop changing, plus competitor watch and ship checks.
The leverage is not the prompt but specifying "done" precisely enough that an agent and a grader can self-terminate, which reframes the PM skill to build as writing falsifiable acceptance criteria rather than collecting prompt templates.
Vendor-sponsored research (Aven Hospitality with Skift Studio) finds only 11% of hotel organizations run AI agents that can actually complete bookings or price inventory in real time, while roughly 80% confine AI to chatbots and marketing — the gap, it argues, decides whether a property surfaces in conversational recommendations at all.
The contested ground is the recommendation moment, not the checkout, so the defensible work is making rate-and-availability data machine-readable and retrievable by agents rather than optimizing rank position in a fading link-based results page.