Craig Abbott

Polypane logo

Something I did a while back but just noticed I never linked to it from a11y-tools home page: a GitHub repo for PolyPane workspaces geared to accessibility testing:

👉 github.com/lloydi/PolyP...

You might not be aware of these and find them useful 😊

#testing #tools #accessibility #a11y #polypane

Rearranging the layout of apps and widgets on iOS, continues to be one of the most horrific user experiences ever.

I’m talking about AI and accessibility at A11y North this month, in Leeds. If it’s something you’re interested in, it would be great to see you there!

www.linkedin.com/events/a11yn...

www.a11ynorth.com

#a11y #accessibility #ai

The Accessibility Crisis of 2026: What No One Wants to Admit Innovation is speeding up, but disabled people are the ones paying the price.

In my talks, I’ve been saying for a while now that as AI continues to dominate “innovation”, accessibility is going to pay the price!

In this awesome article by Tracy Stine, you can see first-hand how many people are affected!

vocal.media/01/the-acces...

#accessibility #ai #a11y

This looks interesting! Will check it out thanks!

I love this idea haha - like crowd sourcing posts… because ADHD 😆

Thank you! I need to write more. Or, should I say, publish more. I have about 100 posts in draft but I can never get any of them over the line!

Thanks for sharing @priyanca.bsky.social, it’s also helped me notice that the meta description for this deck is wildly incorrect! 😆

Overheard somebody in the gym this morning, saying they’re “full of miniature heroes and regret”. 😆

Happy new year!

TalkBack does not offer an option to use computer vision & LLMs (“AI” for the scope of this thread) to describe images lacking alt text.

If you do it in Chrome, it overwrites all the good alt text with, well, crap.

Try it on this page:
srt.csb-cde.ca.gov/jaws/jaws-l

[1/4]

Autistic burnout, at this time of year. Changes in routine. Increased social activity and masking. Expectations on how you’ll spend your time, your money and your energy. Sensory overload, different lights, textures, smells and sounds. It’s ok not to be ok, even when it’s “the season to be jolly”.

A spray painted shaggy looking raccoon holding a fork.

I really relate to this Holborn graffiti.

Die hard is a Christmas tradition in our house 😆

a man is holding a lighter with the words come out to the coast get together have a few laughs above him Alt: Bruce Willis as John McClane in Die Hard. He’s inside of air conditioning ducting holding a lighter with the caption: come out to the coast. Get together. Have a few laughs.

Movie you’ve watched more than six times with a gif. Hard mode: no Stars (Wars nor Trek), LOTR, or Marvel.

Yep! They’ll usually diagnose anything which falls under “common mental health conditions”. Anxiety and Depression are the main two, but there are others, like OCD and addiction. When the case is severe, complex or uncommon, then they’ll refer to a psychiatrist.

That’s not necessarily true. Sure, a psychiatrist will diagnose neurodivergent conditions like ADHD and Autism etc, but GP’s will readily diagnose and prescribe for mental health conditions like anxiety and depression, which the article seems to be more focused around.

If you flip the response around, if GP’s believe there’s a problem with over-diagnosis, do they also believe a lot of people are being mis-diagnosed? They’re the ones doing the diagnosing, so are they incompetent or do people fit the criteria?

Got you. Then yeah, that makes total sense!

I think this is what I was trying to understand when you said small websites. I was thinking about it from a traffic perspective, but if you mean moving parts and complexity, then GOV is obviously a very simple structure

I also appreciate that GOVUK websites are boring as f… They don’t need to handle a lot of complex interactions. It’s perhaps not a great comparison. But it does show HTML will scale far beyond what a lot of people think it’s capable of.

You’re right, it does, they use a CMS called Whitehall Publisher for parts of it. So not all of it is static. But most of their digital services which handle applications for everything in Gov, are mostly just static HTML pages, either written by hand or compiled from Nunjucks templates.

I think it depends what you mean by small. Websites like GOVUK and parts of the BBC handle millions of visits per day using good ol’ HTML. Inefficiencies, poor architectural decisions or committing to a framework which doesn’t scale is going to kill your website way faster than using stock HTML.

Doing the hard work to make things open Paul Smith, a frontend developer at DWP, writes about why it's so important to make our work open and how we can all help make this happen.

Not necessarily a list as such, but there are a few good things dotted about the GOVUK Service Manual and the GDS blog from back in the day designnotes.blog.gov.uk/2017/03/24/d...

Thanks dude!

For what it’s worth, I stand by the principle though. I’m tired of people discounting others on their own assumptions about what they can and cannot do, or should and shouldn’t do. I’ve met brilliant visually impaired designers. Tools and attitudes often hold them back far more than their eyesight.

Just had my first experience of being blocked on LinkedIn. Turns out people don’t like being called ableist when you call them out for suggesting tools like Figma are not for people who use screen readers. Being autistic though, I’m still left feeling like I was somehow in the wrong. 🙃

I think the expectation is that it should write code that works, but it often doesn’t. 😅

I’d expect a drop off once the context window is exceeded, but it doesn’t tell the user when that is. It just fails silently and sends them in circles suggesting fixes it previously tried that didn’t work.

Nah. My dreams also like to crash at runtime 😩

When agentic AI chains multiple steps to try and solve this issue, say it generates working code 95% of the time, by the time it chains together 20 steps, there’s only a 36% chance of it being correct. The probability drop off is not an engineering problem, it’s a mathematical inevitability.

I’ve found that it even when an AI writes “working code”, it’s cobbled together in such a way it only works for so long. It’s often not modular and often does not have clear separation of concerns. So once the codebase grows beyond the size of the context window, even the AI struggles to debug it.