Intelligence Track
1 June 2026
The Brief
Today's issue is dominated by a single structural theme: the gap between AI as a product announcement and AI as a working operational system is closing — and the teams that close it first are changing what competitive means in travel.
The Expedia and Indian OTA AI divide stories read together, and they should: one shows the global ceiling rising, the other shows that most Indian players are still finding the floor. Cleartrip's absence from the Skift earnings analysis is the sharpest signal in today's batch — not because it confirms a product gap, but because it confirms a perception gap, and in a market where investor and supplier confidence follows narrative, that distinction matters less than it should. The three product and design pieces — on agentic PM workflows, probabilistic done criteria, and motion judgement — point toward the same operational conclusion: shipping AI features with pre-AI processes will produce AI products that feel like pre-AI products. The team should be asking not just what AI features are on the roadmap, but whether the rituals, roles, and review practices around those features have been updated to match. The window to get ahead of this is narrower than it looks.
Skift · 1 Jun 2026
Expedia's newly appointed Chief AI and Data Officer Xavier Amatriain — formerly of Netflix and Curai — outlines the company's AI strategy around trust, agentic booking flows, and proprietary travel data as a moat. The interview signals Expedia is moving from AI as a feature layer to AI as the core product architecture.
Gothelf argues that AI features break the standard sprint-completion model because behaviour is probabilistic, not deterministic — 'all tests passed' no longer means the feature works as intended in production. He proposes replacing binary done criteria with ongoing outcome monitoring, explicitly separating technical completeness from behavioural correctness.
Why it matters
Product teams shipping AI-assisted search, pricing recommendations, or chat support cannot rely on QA-gate release cycles — without updated definitions of done, they will consistently ship features that pass review and fail users.
Paweł Huryn documents a working model for PMs in agentic engineering: instead of reading code, PMs review the outputs of AI-assisted development — specs, test results, generated documentation, deployment artefacts — and use a prompt library to direct agent workflows. He frames this as the new core PM competency, not a shortcut.
Why it matters
If artifact review replaces code review as the PM's technical interface, the skill profile for product roles shifts — travel PMs who can direct agentic workflows will operate with a larger effective surface area than those who cannot, compressing delivery timelines on feature work.
Also in AI & Design
A new recurring digest curating open-source agentic tooling from GitHub, covering a YC-founder personal knowledge agent, a unified filesystem abstraction for AI agents, and high-star repositories reaching mainstream adoption. Framed for practitioners building with or evaluating agent infrastructure.
A non-technical talent leader documents shipping a fitness app to the App Store using AI coding tools and AI-generated video assets with no engineering support, from idea to live product in weeks. The piece functions as a case study in what AI-assisted development makes possible for non-engineers today.
Also in Design Systems
A tool that generates mathematically optimal font pairings by analysing typographic compatibility metrics rather than relying on curatorial taste or convention — outputs ranked pairing recommendations with rationale.
A practical training exercise in motion design judgement: side-by-side animation comparisons force the reader to identify what separates technically correct motion from motion that actually feels right — naming the specific properties (easing, timing, spatial arc) that produce the difference. Written as a pre-emptive argument against shipping AI-generated motion code uncritically.
Why it matters
As AI code generation reaches UI animation, design teams that cannot articulate quality criteria will lose the ability to QA their own output — which degrades perceived quality at precisely the moments (transitions, feedback states) that build or break booking trust.
Expedia has hired a former Pinterest advertising executive to run its ads business with a repositioning pitch: selling suppliers on conversion-linked inventory (confirmed bookings) rather than intent-signal impressions. The move targets hotel and airline advertisers frustrated with upper-funnel spend that doesn't close.
Why it matters
A booking-outcome ad model changes the supplier relationship — hotels and airlines will shift co-op budgets toward OTAs that can prove closed revenue, which reshapes who gets premium inventory placement and at what margin.