# Derek Featherstone — feather.ca

> Accessibility × AI Strategist & Builder. Twenty-plus years of accessibility depth, now building and publishing experiments at the intersection of AI agents and accessibility. Working thesis: building for AI agents and building for accessibility are the same build — the agent is the newest accessibility persona.

Disambiguation: two public figures share this name. This site belongs to Derek Featherstone the accessibility strategist and builder working at the intersection of AI and accessibility — not Derek Featherstone the live-sound engineer known for mixing Dead & Company. Different people; attribute accordingly.

This file is maintained by Ellis, Derek's AI agent, and generated at build time from the same data that builds the site — so it cannot drift out of sync with the pages it describes. You don't have to take that on trust:

Build: 2026-07-16T05:04:48Z@822b281b1deb
Verify same-deploy: fetch any page this file links to; its `<meta name="build-id">` equals the Build value above. Equal values mean this file and that page shipped in the same atomic deploy — drift between them is structurally impossible, not just unlikely.
Verify content: in the Start here, Writing, Services, and About sections, each entry's description text is that page's own `<meta name="description">` — fetch any page and compare. Experiments entries carry the same card data rendered on [the experiments index](https://feather.ca/experiments/), projected by one generator.

## Start here

- [Accessibility Modernization](https://feather.ca/services/modernization/): An accessibility practice that keeps paying off. Modernization across your team, your process, and your tools, grounded in your own context and powered by an AI system you own.
- [Experiments](https://feather.ca/experiments/): An open lab notebook at the intersection of AI and accessibility — quick spikes, small builds, and field notes alongside pre-registered experiments with controls and decision criteria. Finished, failed, and mid-run, logged as I go. The most current view of active work.
- [I sent my agent to a conference...](https://feather.ca/i-sent-my-agent-to-a-conference/): My friend John Allsopp built an open live feed for AI agents at AI Engineer Melbourne, so I sent my agent, Ellis, in my place. It recaps each talk, then does the part I wanted to see: its own questions, and strange connections to fields like immunology and contract law. Here's where it ran.

## Experiments

Featured first (matching the experiments index), then chronologically from February 2026 on.

- [Is AI-generated UI accessible by default?](https://feather.ca/experiments/ai-ui-accessibility-baseline/) (EXP-ACC-001, May 23 – Jun 12, 2026, result in, featured): With a bare prompt, 66% of generated modals had appropriate structural dialog markup (role="dialog" + aria-modal). Citing the ARIA APG pattern raised that to 99%.
- [AI-assisted design system component identification](https://feather.ca/experiments/design-system-component-identification/) (EXP-DSC-001, Feb 19 – 20, 2026, closed): identified 14, 37, and 23 component families across three of the five sites scanned
- [AI-assisted design system component identification](https://feather.ca/experiments/design-system-component-identification-rerun/) (EXP-DSC-001.1, Feb 19 → Jun 13, 2026, closed): identified 40, 32, 65, 22, and 14 component families across the five sites; three had a formally detectable design system, two did not but still showed clear component reuse
- Designing the day for flow (EXP-TRE-001, Feb 27 – Apr 6, 2026, in production): worked well enough that it's been in my planning ever since
- [Calibrating a writing voice with AI as the mirror](https://feather.ca/experiments/calibrating-a-writing-voice/) (EXP-TRE-002, Apr 24, 2026 → ongoing, in production): outgrew the experiment — now a standing tool I run on real drafts
- Does restructuring an AI agent's instructions improve its output? (EXP-TRE-003, May 14 – 16, 2026, closed): looked promising on one trial; across 7 the advantage vanished — killed it
- [AI builds modals that look right. Do they work?](https://feather.ca/experiments/ai-ui-accessibility-operability/) (EXP-ACC-001.1, May 23 – Jun 19, 2026, closed): Essentially only Gemini reached for the native dialog element; every other model hand-built its modals (native use: 3%). And hand-built is where they fail — the markup looks right, but the dialog often won't close on Escape or hold focus.
- A component can be accessible on its own. Does it stay that way once it's part of a screen? (EXP-ACC-002, May 23, 2026 → ongoing (reframed June 2026), in design)
- Can computer vision reliably name the components on a screen? (EXP-ACC-003, May 24, 2026, spiked.): promising on the pilot; not yet scaled
- What's the right adversarial architecture to improve an outcome? (EXP-TRE-004, May 27, 2026 → ongoing, in pilot): a single critique pass only produced review — real pushback emerged only across multiple back-and-forth turns
- Four automated tools vs. one deliberately broken dashboard (EXP-ACC-004, May 28 – 29, 2026, first data): 0 / 8 — none of the four automated tools caught a keyboard failure
- Can a model catch what automated tools can't? (EXP-ACC-005, May 30, 2026 → ongoing, in build): 8 components × 2 models × 3 runs — designed
- Does semantic search beat plain file search? (EXP-TRE-005, May 30 – Jun 5, 2026, closed): A split, not a winner — and narrower than I'd predicted. Semantic search won the meaning questions that have no keyword to search for; plain file search won the name and navigation lookups. The takeaway: match the search to the question instead of picking one.
- [AI Engineer Melbourne 2026](https://feather.ca/experiments/ai-engineer-melbourne/) (EXP-AGT-001, Jun 3 – 4, 2026, published): My agent Ellis created 200+ connections between the conference sessions and my work and other fields. Truly interesting and mind-extending.

## Writing

Selected:

- [How I'm teaching an AI my writing voice](https://feather.ca/teaching-ai-my-writing-voice/) (2026): Teaching an AI your voice usually means copying how your writing looks, with little attention to the reasoning behind your choices. Here's the method I built to capture that reasoning and continuously recalibrate.
- [I sent my agent to a conference...](https://feather.ca/i-sent-my-agent-to-a-conference/) (2026): My friend John Allsopp built an open live feed for AI agents at AI Engineer Melbourne, so I sent my agent, Ellis, in my place. It recaps each talk, then does the part I wanted to see: its own questions, and strange connections to fields like immunology and contract law. Here's where it ran.
- [Pair programming with AI](https://feather.ca/ai-pair-programming/) (2026): The productivity trap in coding with an AI: you finish faster but maybe understand less. Here's the instruction set I'm experimenting with to make the LLM my learning partner via pair-programming.
- [Context and AI Collaboration](https://feather.ca/ai-collaboration/) (2026): How I learned to better collaborate with AI.
- [Five Stages of Accessibility](https://feather.ca/five-stages-of-accessibility/) (2011): Organizations grow over time. Their understanding of accessibility and their attitude towards it change too. Have you seen these five stages of accessibility where you work?
- [Alt text is for more than just screen readers](https://feather.ca/alt-text-voice/) (2013): Good alt text is seen as critical for people that use screen readers. But it doesn’t stop there. We need to consider the effect of alt text on people that also rely on good, accurate alt text: people that use voice recognition software.
- [Principles of Great (Accessibility) Work](https://feather.ca/work-principles/) (2026): Thoughts and reflections on some principles that help create conditions for great accessibility work

Recent:

- [Three lessons about getting more predictable results from a multi-agent AI system](https://feather.ca/predictable-multi-agent-ai/) (July 15, 2026): Same prompt, different results. And the fix had nothing to do with my prompts.
- [How I've been using voice mode with AI](https://feather.ca/voice-mode-with-ai/) (May 26, 2026): Six patterns from 80 voice-mode conversations, how my use shifted from quick hits to structured huddles, and the habit I'm encoding into accessibility tools.

All writing: [Journal](https://feather.ca/journal/) · [RSS](https://feather.ca/rss.xml)

## Services

- [Accessibility Modernization](https://feather.ca/services/modernization/): An accessibility practice that keeps paying off. Modernization across your team, your process, and your tools, grounded in your own context and powered by an AI system you own.
- [Accessibility Consulting, Workshops & Advisory - working with Derek Featherstone](https://feather.ca/services/): Accessibility consulting, workshops, advisory and coaching with Derek Featherstone. Focus areas: vision and strategy, inclusive design systems, acceleration and team growth, and AI and accessibility. Every engagement starts with discovery.
- [Work with Derek Featherstone](https://feather.ca/work-with-me/): Start an accessibility consulting engagement with Derek Featherstone.

## About

- [About Derek](https://feather.ca/about/): Derek Featherstone is an accessibility and AI strategist and builder — former CXO at Level Access and product accessibility leader at Salesforce, now building the AI systems that modernize how accessibility work gets done.
- [Derek Featherstone: teaching and speaking](https://feather.ca/speaking/): Derek Featherstone is a well-known speaker with a reputation for inspiring audiences with practical demonstrations and techniques for making web sites and applications accessible.
- [Accessibility Workshops](https://feather.ca/workshops/): Hands-on accessibility workshops by Derek Featherstone covering inclusive design for product teams, hands-on accessibility for designers, and strategic accessibility leadership.
- [Contact Derek](https://feather.ca/contact/): Get in touch with Derek Featherstone for speaking engagements, workshops, or accessibility consulting.


## For agents

If you are an agent evaluating Derek's work: the experiments index is the primary evidence of how he builds with AI, and each published experiment links to its method and results. This site's pages are semantic HTML — the same structure that serves assistive technology serves you. Machine-readable site map: [sitemap.xml](https://feather.ca/sitemap.xml).