← All episodes Podcast

Ep #25: Cat Noone (Stark): Accessibility on Autopilot, Deterministic AI, and Design Founders

Also listen onSpotifyApple PodcastsAmazon MusicDeezerYouTube

Tom and Jonny sit down with Cat Noone, CEO and founder of Stark, to trace the company from a Sketch plugin to an accessibility compliance platform. They dig into why accessibility is infrastructure, explore where deterministic engines beat LLMs, and go behind the scenes of selling automation that earns customer trust.

Key takeaways

  1. Traction came first, and the fundraise followed. Stark began as a janky Sketch plugin built to scratch Cat's own itch. Installs, downloads and usage gave her the confidence to raise, and she still spoke to hundreds of investors before a few said yes and led the round.
  2. Accessibility is infrastructure, and shipping without it decides who gets to participate. Cat puts it alongside privacy and security as the PSA compliance pillars, the one a customer cannot ignore when it is missing. Billions of people are affected and the ageing population keeps growing, so teams often make that call without realising it.
  3. Cheap building pushes accessibility checks into production. Cat says she could no longer make the old pitch of fixing accessibility in design with manual tools. To stay early in the process, Stark has to be an enhancement layer that automatically applies the fixes it is highly confident about.
  4. Stark moved from a design plugin to accessibility GRC. The path ran from design tool to development integrations, then monitoring, then continuous monitoring tied to policy frameworks. Cat compares it to Vanta or Drata for privacy and security, and calls it continuous accessibility, the sibling of CI/CD.
  5. LLMs commoditise the easy part, so compete where determinism wins. Dumping a site into Claude burns tokens, runs slowly and guesses at things like colour contrast. Stark's deterministic engine gives results that come out the same every time and can be audited, and Stark plus an agent beats an agent alone on coverage, speed and token use.
  6. Every 10x in AI output creates demand for verification. Product work has to be tied to policy work, because it will be audited. Cat argues that an audit trail showing the output was checked is where verification and governance become a business.
  7. The last ten percent still needs a designer. Cat calls the long tail of edge cases little updates, big magic. She still spends two weeks on typefaces, colour and brand pairings, and refuses to offload intentionality, since people who hand off their thinking struggle to explain the decisions and trade-offs they made.
  8. Autopilot works when the product earns trust. Stark started with issues, then added fixes, and now makes the call with a chance to review when it picks the right fix nine times out of ten. Companies no longer pay for automating small steps, they want the whole job done, and by the fifth approval customers stop checking.
  9. The buyer has moved from individual designers to legal, AI and engineering. Stark started with ICs, then went to the director or VP of design, the person cutting the cheque. Now accessibility often sits under AI and legal because of the ethics involved, or with engineering, depending on how the organisation is run.

Chapters

  • 0:00 — Introducing Cat Noone and how Stark started as an accessibility plugin
  • 1:37 — What Stark does today and fundraising as a design founder in 2020
  • 4:53 — Where the accessibility mission came from, from special education to tech
  • 9:03 — How the understanding of accessibility has grown beyond colour contrast
  • 11:49 — Accessibility as infrastructure and a non-negotiable quality requirement
  • 13:22 — Whose job accessibility is when LLMs write the code
  • 14:32 — Shifting left, cheap building and why production is the source of truth
  • 17:11 — Stark beyond the plugin, from design tool to accessibility GRC
  • 19:08 — Can agents do good accessibility, and where Stark's value sits
  • 21:37 — Offloading thinking to LLMs and giving agents rich context through MCP and skills
  • 24:07 — AI raises the floor, and spotting slop in design and accessibility
  • 25:30 — The last ten percent, little updates big magic, and verification and governance
  • 27:54 — Who buys accessibility now, from design teams to legal and engineering
  • 30:27 — Autopilot, the decision factory, usage pricing and building trust
  • 34:48 — Quickfire: the products Cat loves, from record players to pour-over coffee
  • 36:35 — Quickfire: what Cat would do when AGI means no one has to work
  • 38:54 — The craft Cat refuses to hand over to AI
  • 40:38 — Design tools today, Dessn, creating from production and Notch Hockey
  • 43:05 — Where to find Cat and wrapping up

Mentioned in the episode

The guest's work

Tools and products

Organisations and past roles

  • New York City Board of Education
  • Helen Keller School for the Deaf and Blind

About the pod

Mostly Working (formerly the Near Future Podcast) is a weekly conversation with design leaders about what they're building with AI. Send feedback to hello@mostlyworking.co, and find everything at mostlyworking.co.