Latest · AI oversight & governance
Greece told every public body to list its AI, but the list is the easy part.
A national register asks the questions human oversight exists to answer. Many bodies will answer them like an inventory.
Articles
What we learn on real projects, written up so your team can use it.
We write about keeping people in charge of artificial intelligence (AI), getting found by AI search (generative engine optimization, GEO), sharper design practice and accessibility.
Topic
Language
No articles match these filters yet.
Articles, newest first
AI oversight & governance
The human in the loop is a job
Role profiles, a checklist and a job description template for the AI Act’s rule on who oversees high-risk AI, which applies from December 2, 2027 in areas such as hiring, credit scoring and public services.
AI oversight & governance · Ελληνικά
Όροι σύμβασης για ανθρώπινη εποπτεία που αποδεικνύεται
Υπόδειγμα όρων για φορείς του δημοσίου που προμηθεύονται συστήματα τεχνητής νοημοσύνης: δοκιμή πριν από την παραλαβή, με όριο που ορίζει ο φορέας εκ των προτέρων.
AI oversight & governance · Ελληνικά
Δήλωση στο Μητρώο Συστημάτων Τεχνητής Νοημοσύνης: οδηγός για φορείς του δημοσίου
Τα έντεκα στοιχεία που ζητά ο νόμος για τη δήλωση στο Μητρώο, με υπόδειγμα, συμπληρωμένο παράδειγμα και μια δοκιμή ανθρώπινης εποπτείας μίας εβδομάδας.
AI oversight & governance
My fact-checker caught every error I planted. It missed the ones I made.
My AI fact-checker caught 108 of 108 planted errors and missed my real one. Seven steps for an audit that tests the errors you actually make.
Design practice
I made it impossible to publish a claim I hadn’t checked
Most of the debate about AI writing is about who wrote the draft. The check that matters comes after it — a gate no unverified claim gets through.
GEO
I built the GEO scoreboard. My score is zero.
Part three of the GEO series was meant to be about measurement. Here is the instrument for tracking AI citations, and the number it returned for my own site.
New here? Start with these
One piece for each topic we write about, chosen as the best way in.
AI oversight
Room to Disagree
The framework behind our AI oversight work: five capabilities and one test that shows where oversight fails.
GEO
The machines can’t cite what they can’t identify
Why AI search skips content it can't identify, and what to make explicit so it gets quoted.
Design practice
How to run a UX audit in under 2 hours
A repeatable way to evaluate a product and prioritize fixes in under two hours.
Accessibility
Why accessibility overlays are now a bigger legal risk than doing nothing
Why overlay widgets add legal risk instead of removing it, and where the real problems hide.
More articles
Accessibility
The accessibility annotations nobody reads
The annotations live in Figma. The handoff fails three sprints later, in a ticket where nobody has the Figma file open.
AI oversight & governance
Room to Disagree
A framework for designing and testing the interfaces through which people oversee AI systems. It sets out five capabilities drawn from the AI Act’s article on human oversight, one condition underneath all five, and a behavioral test that shows where the oversight you have documented fails.
AI oversight & governance
Telling people they are talking to a machine is now a design problem
The AI Act’s rule on telling people when they are dealing with AI has applied since August 2, 2026. Most of what it asks is decided on the screen: the wording, where it sits and whether everyone can perceive it.
Accessibility
Your design system cannot make a product accessible
Accessible components are a good start. Whether the product works for everyone is decided in how they are put together: the flows, the content and the choices no library makes for you.
Design practice
Nobody gets promoted for better screens.
Better screens rarely move a design career. I look at what does, and why the work many of us are proudest of is not it.
Accessibility
Why accessibility overlays are now a bigger legal risk than doing nothing
Overlay widgets promise compliance in one line of JavaScript. The evidence points the other way — they attract complaints, and they make the underlying problems harder to find.
Accessibility
Most accessibility fixes are cheap in Figma and expensive in code
A contrast or focus-order fix takes minutes in Figma. Found later in code, the same issue costs a ticket, a rebuild and a retest.
Accessibility · Ελληνικά
Πρόστιμα EAA στην Ελλάδα: τι λέει πραγματικά ο ν. 4994/2022
Ποιος ελέγχει τι, γιατί ο νόμος δεν ορίζει κανένα ποσό προστίμου και τι εκθέτει πραγματικά τα e-shops και τις επιχειρήσεις.
Accessibility
Nobody buys an accessibility audit. They buy the report.
The testing is the easy part. What clients actually pay for is a document their team can act on — and that is where most audits fall down.
GEO · Accessibility
GEO just became a budget line describing accessibility
Structured, machine-readable content is what AI engines cite. It is also, almost line for line, what accessibility has been asking teams to do for years.
GEO
The machines can’t cite what they can’t identify
AI engines cite what they can name with confidence. If your content doesn’t make its entities explicit, it doesn’t get quoted — however good it is.
Accessibility
The EAA, one year on: enforcement arrived — just not from who you expected
Twelve months after the deadline, most of the pressure is not coming from regulators. It is coming from procurement, customers and competitors.
Design practice
How to run a UX audit in under 2 hours
A practical, repeatable process for designers who want a fast, no-fluff way to evaluate a product and prioritize fixes.
Design practice
I built a Figma plugin to audit my own team's work
What 235 unnamed layers taught me about the gap between having design standards and actually keeping them.
Design practice
Why design systems are the only investment that gets cheaper as AI gets smarter
After 18 years and six enterprise rollouts, why design systems are a smarter bet than ever in the age of AI.
Accessibility
Accessibility debt and hidden costs
How delaying inclusive design quietly erodes product quality, user trust and revenue — and what it costs to catch up.
Design practice · Accessibility
The myth of the “average user”
We keep designing for someone who doesn't actually exist. A case for designing to the edges instead of the mythical average.
Accessibility
What most teams misunderstand about the European Accessibility Act
A few alt-text fixes don't make you compliant. What the EAA actually asks of digital teams — and how to get ready.
Get new articles by email
New articles as they go live, and the Room to Disagree newsletter on designing human oversight of AI.
Have a project in mind?
Tell us what you're working on. We'll recommend the right first step.
Start a project