Intelligence Track
2 June 2026
The Brief
Today's issue has a clear centre of gravity: AI in travel is moving from capability to economics, and the economics are increasingly adversarial.
The Skift pair on invisible AI and infinite search costs lands together — Booking.com is telling the industry that trust is the scarce variable, while the search cost piece explains why the agentic model is structurally destabilising the unit economics that OTAs depend on. These two signals do not cancel each other out; they point at the same problem from opposite ends. Building agentic travel planning is expensive and risky if the underlying cost model of unbounded search is not solved. The Scapia Series C adds a third pressure: a well-capitalised competitor is building financial ownership of the traveller before they enter the booking funnel. For Cleartrip, the threat is not primarily price competition — it is the loss of funnel entry. Meanwhile, MakeMyTrip's Creator Circle with Meta is a deliberate move to own the inspiration phase in the same way that Scapia is moving to own the payment phase. Cleartrip's current product surface addresses the consideration and booking phases well; the team should be asking what it owns at inspiration and what it owns post-booking in financial terms — because both ends of the funnel are being actively enclosed by better-resourced competitors right now.
Skift · 2 Jun 2026
AI agents eliminate the natural stopping behaviour of human search — people eventually stop comparing, but agents do not — which is exploding look-to-book (L2B) ratios for airlines and hotels. The cost of serving search queries that never convert is shifting from a nuisance to a structural cost problem for intermediaries and suppliers, with airlines particularly exposed given the direct infrastructure costs of GDS and NDC query processing.
The article applies Sturgeon's Law (90% of everything is mediocre) to AI-generated content, arguing that increased content volume from AI tools amplifies the mediocrity problem rather than solving it — more output means more noise, not more quality, unless curation and quality standards are built into the generation process.
Why it matters
Travel platforms using AI to generate itinerary copy, hotel descriptions, or destination content at scale will flood their product with generic, low-trust content unless they build editorial standards into the generation pipeline.
Gergely Orosz documents that developers using AI coding agents are generating twice as much code as six months ago — but code volume is outpacing review capacity, introducing quality, reliability, and tech debt problems that compound faster than teams can address. The rational fix proposed is deliberate upfront planning and specification before invoking AI generation, not more generation speed.
Why it matters
Product and engineering leads who are measuring AI-assisted development by velocity or output volume without a corresponding quality gate are accumulating invisible tech debt that will slow iteration speed in proportion to how much was rushed.
Sarvam AI is preparing to open its Samvaad voice agent platform to self-serve public access, moving beyond its current enterprise-only model. Samvaad supports 11 Indian languages, operates at sub-500ms latency, and connects to CRM, payment, and booking systems — the same integrations required for conversational travel booking across voice, WhatsApp, and web.
Why it matters
A self-serve Samvaad means any Indian startup or OTA can deploy multilingual voice AI for customer support, booking assistance, or post-booking servicing without needing an enterprise contract — lowering the barrier to voice-first travel UX significantly.
The piece argues that static screenshot-based inspiration archives teach designers to copy visual outcomes rather than understand motion, micro-interaction, and taste formation — the author identifies four motion-focused archives that capture the reasoning layer beneath visual polish. The core argument is that AI can replicate static aesthetics but cannot yet transfer the kinetic and temporal sensibility of strong interaction design.
Why it matters
For a travel product team shipping interactive search, date pickers, map surfaces, and booking flows, motion and micro-interaction quality is a trust and comprehension signal — not decoration — and teams without exposure to it will produce flat, unconvincing interactions.
The article argues that design taste trapped in individual contributors' heads creates a bottleneck — teams can't scale quality output unless the person with taste has built systems, principles, and critique frameworks that others can use autonomously. The cost of hoarded taste is slow review cycles, inconsistent execution, and designers who cannot grow.
Why it matters
For a product team scaling AI-generated UI or operating with a lean design org, externalising taste into documented principles and review rubrics is the only way to maintain quality as output volume increases.
The piece examines how default settings in digital products encode decisions made by designers or business stakeholders — not users — and how those defaults shape behaviour at scale without users consciously choosing them. It challenges teams to treat every default as a deliberate design decision with measurable downstream effects.
Why it matters
In a travel booking funnel, defaults around fare class, flexibility, ancillary pre-selection, and payment method selection carry real revenue and conversion implications — treating them as neutral is a design accountability gap.