Intelligence Track

Design

3 July 2026

The Brief

An essay begging the industry to stop performing AI capability it can't reliably deliver lands the same week a global leisure platform wraps its India corporate push in 'AI and seamless experiences' — and the juxtaposition is the story. Strip the marketing language away and the actual wedge that platform is driving is unglamorous: zero-fee onboarding and local tax compliance, not intelligence. Two practitioner pieces sharpen the same point from opposite ends — the distance between an AI demo and a load-bearing workflow is enormous, whether it's an agent booking flow that triggers half the time or a Figma-to-code pipeline where the human is still stuck in the seam because the semantic tokens were never authored. The value lives in the plumbing — evaluation, monitoring, machine-readable design contracts, funnel and compliance ownership — while the confident capability claim stays theater until that plumbing exists. Once every competitor's marketing reads 'agentic and seamless,' what buyer signal separates the platforms that built the infrastructure from the ones performing it?

Elena Verna · 3 Jul 2026

Please stop the AI Confidence Theater

Performative overstatement of AI capability — 'life-changing' agent workflows that in practice trigger half the time and need heavy hand-holding — is framed as doing measurable damage: it breaks hiring signals now that verbal fluency in MCP, RAG, and agents no longer proves competence, distorts genuine adoption, and manufactures a reverse-hustle culture where burning tokens replaces showing outcomes. The named drivers are attention economics, the difficulty of verifying anyone's claims, marketing that sells certainty AI can't deliver, and VC-to-exec-to-employee pressure to perform miracles.

Industry lens

As agent reliability improves through 2026, does the hiring signal recover — or do case-study and work-trial screens become the permanent default across product and engineering teams?

Reading as

AI & Design

You design it. Then what? A clear map of the Figma-to-code AI mess

A beginner's map through the Figma-to-code AI pipeline argues the determinant of usable AI-generated UI is upstream file structure — semantic tokens that carry intent, and Figma component-property names treated as a contract matching code props — not the generation tool itself; where the semantic layer is missing, the agent guesses, usually wrong in ways hard to catch before shipping.

The actionable read for a design lead is that AI code-gen quality is capped by design-system hygiene done before any agent runs, so the highest-leverage work is retrofitting semantic tokens and Figma-to-code prop-name parity now — and the designer's output shifts from pixels toward authoring machine-readable intent that both an agent and a developer can read unambiguously.

UX Collective·3 Jul 2026