Matt Campbell

Software developer, formerly at Microsoft, now co-developer of the AccessKit open-source project (https://accesskit.dev/) and cofounder of Pneuma Solutions (https://pneumasolutions.com/). My current favorite programming language is Rust, but I don't want to make that part of my identity.

Music lover. Karaoke singer. Science fiction fan. Legally blind. Secular humanist

Pronouns: he/him
GitHub: https://github.com/mwcampbell

Public Key

npub19nhp7esyu59cw3v3e4jeac5nfy5zwj8cd2m2haa43h3apj58033qmzv7fc

NIP-05 Address

matt_at_toot.cafe@momostr.pink

Profile Code

nprofile1qqszemslvczw2zu8gkgu6ev7u2f5j2p8frux4d4t776cmc7se2rhccspz3mhxue69uhhyetvv9ujuerpd46hxtnfduqs6amnwvaz7tmwdaejumr0dsqk7hel

Publishing to

relay.damus.io

nos.lol

Author Public Key

npub19nhp7esyu59cw3v3e4jeac5nfy5zwj8cd2m2haa43h3apj58033qmzv7fc

Show more details

Last Notes

2026-07-30 13:22:06 UTC

- reply

I'll be curious to hear what my followers who are into programming language design and psychology (I do have at least one follower in each of those fields) think of this. To me though, it sounds like we still have room for improvement in programming language design. Not that there's much funding for that, with everyone chasing AI.

2026-07-30 12:48:20 UTC

Question for people who vibe-code full apps, especially blind people, as I believe that covers most of my followers who do this: What would it take for you to be comfortable coding directly in a programming language without an LLM? A high-level programming language with better tutorials and quickstart templates than are currently available, particularly for writing real apps? Or do you believe that having an LLM generate the code is the only way you could practically do what you do?

2026-06-21 12:15:55 UTC

Question: What's the best neural text-to-speech that I can run on an ARM64 Linux machine? Piper, Sonata, or something else?

2026-04-19 20:43:38 UTC

- reply

Is my Mastodon client broken, or am I missing an obvious reason why that image itself doesn't have alt text?

2026-01-28 19:39:07 UTC

- reply

As a former evangelical Christian, one of the last steps on my road to deconversion was finding that the Bible just didn't hold my interest. Mark is the shortest of the four gospels, but even so I'd say it's probably not worth your time.

2026-01-13 22:35:48 UTC

- reply

I wonder if you've tried Chrome OS, whether via a Chromebook or installing Chrome OS Flex on an old laptop, with people like him.

2025-12-18 12:54:59 UTC

- reply

That distinction reminds me of Ted Chiang's story "The Truth of Fact, the Truth of Feeling".

2025-11-05 15:17:05 UTC

Question for any blind desktop Linux users, particularly GNOME users, out there: What are the remaining practical issues with running Orca in a Wayland-based GNOME session? I'm guessing inability to synthesize mouse clicks is the main one.

2025-10-28 19:04:35 UTC

- reply

Is it a Linux-capable Raspberry Pi or a Raspberry Pi Pico? To me the word "full" implies the former, but the fact that you said the code is in MicroPython and there isn't much space for a keyboard driver implies the latter.

2025-10-27 16:51:42 UTC

- reply

How about Ghost? Can't it do email newsletters?

2025-10-05 10:30:44 UTC

I remain uncertain on the ethics of using generative AI, and I know I've posted some hard-line anti-genAI takes here and elsewhere. But if one is interested in locally runnable models, IBM Granite 4.0 is worth a look IMO. https://www.ibm.com/new/announcements/ibm-granite-4-0-hyper-efficient-high-performance-hybrid-models It uses something other than the typical transformer architecture, allowing for much longer contexts with much less RAM usage. Also interested in what they mean by "ethically acquired" data. cc @npub1gv2…tlwl

2025-10-02 12:24:47 UTC

- reply

Your conclusion, "The real reason not to use GenAI", seems at odds with your statement "I’m not a hater." Shouldn't your reasons for not using GenAI lead you to advocate that we don't even use it for coding, even in the areas where it'll probably work? I guess Those reasons for not using GenAI sound to me like reasons to fight the prediction and try to make it not come true.

2025-09-14 15:15:40 UTC

There was a toot that went around a few years ago that I wish I had bookmarked. It said something like, and I'm badly paraphrasing from memory, "When you consider the usability of your software, think about what it will be like for an elderly person, like Margaret Hamilton, who worked on the software for the Apollo 11 guidance computer and would be dismayed at how modern applications require megabytes just to display a button." Does anyone know the one I'm talking about and have a link?

2025-08-31 11:33:53 UTC

- reply

Pretty much not at all. I've listened to audiobook versions of non-fiction books that probably do have references, but they're not mentioned by the narrator. That's for commercial audiobooks. Recordings made specifically for blind people, at least here in the U.S. by the National Library Service talking book program and Recording for the Blind (now Learning Ally), were better about this, at least when I was in high school and college.

2025-08-03 22:02:01 UTC

Do any #blind folks here use FreeBSD as a desktop OS? I've read that the audio stack is more sensible there than on desktop Linux.

2025-07-17 22:57:44 UTC

Question for any blind people who are on Bluesky: What's the most accessible way to access Bluesky on Windows? The official web app, or is there a good third-party client?
Just curious at this point. I don't want to argue Bluesky versus fediverse. I just think it's good to know what the options are for accessing Bluesky since it's increasingly popular.

2025-07-03 17:50:05 UTC

There was recently an article about using salvaged Sun Ray thin clients with a modern server. @npub1ks7…zt6a's response to that article on Hacker News mentioned his story about using DTrace to debug a performance problem on a Sun Ray server. He linked to a paper he wrote for Usenix, but I prefer his extended telling in this 2007 talk: https://www.youtube.com/watch?v=TgmA48fILq8&t=35m6s

2025-06-19 13:34:59 UTC

- reply

Take note, @npub19yg…w8um: Your claim that the developers of Wayland-based desktop environments, specifically GNOME and KDE, don't care about accessibility is plainly false, and I dare say irresponsible considering the platform you have as a journalist. To be clear, I'm not offended for my own sake. I'm angry for how that article has affected developers who have done far more for free desktop accessibility, e.g. https://tesk.page/2025/06/18/its-true-we-dont-care-about-accessibility-on-linux/

2025-06-19 13:24:26 UTC

- reply

Oh, and this addition shipped in KWin 6.4.0, which was tagged a week ago. Good work @npub1dpj…p0g8.

2025-06-19 13:08:09 UTC

Apparently this is slightly old news (the Git commit timestamp is just over three months ago), but KWin, the KDE window manager/compositor, now implements the solution pioneered by GNOME for letting screen readers monitor and intercept keystrokes on Wayland, specifically, the org.freedesktop.a11y.KeyboardMonitor D-Bus interface. I prototyped this solution in GNOME last summer. Thanks to @npub1hl6…hych for bringing it to production in GNOME, and now to Nicolas Fella for implementing it in KDE.

2025-06-09 17:25:29 UTC

- reply

FWIW, I'm re-reading your post, not listening to WWDC. And making sure that if I post another follow-up, I don't say something really dumb that misses the point of the post.

2025-06-09 17:04:13 UTC

- reply

Got it. You're right, you wrote about that at length.

2025-06-09 16:59:42 UTC

- reply

You argue that the only sensible action is to hold off, that is, not use genAI. Would that be your position on all new technology, all new technology in the current environment, or is there something special about genAI (e.g. being a psychological hazard)?
I ask because I don't like to contemplate how different, and possibly worse, my life would have been if my parents had taken such a cautious position about personal computers in the 80s.

2025-06-04 16:16:44 UTC

- reply

No need to be sorry. Thanks for taking time to reply at length. I'm sorry that, at least to some extent, I've made you rehash things you've already written about.

2025-06-04 15:32:57 UTC

- reply

That project was featured on Hacker News the other day. To me, it's the most convincing of the recent burst of posts. But I still believe it's not _right_ to use these tools.

2025-06-04 15:31:13 UTC

- reply

Have you seen the case of the Cloudflare Workers OAuth provider library (https://github.com/cloudflare/workers-oauth-provider). Particularly see the README section titled "Written using Claude". Kenton says he was a skeptic, intended to prove his skepticism with this experiment, and was surprised. He documented the process, including prompts. He's certain he developed the library faster this way than he would have otherwise.

2025-06-04 15:25:09 UTC

- reply

At least in the posts I've seen, the authors seem to be seeing evidence that the tools can in fact deliver, at least in some contexts, particularly coding. So I've concluded that a decision not to use these tools has to be based purely on principle. I hope I'm somehow wrong about that. I hope there are practical reasons too.

2025-06-03 14:00:19 UTC

- reply

Also, I don't think all automation is equally bad. I've spent most of my career working on assistive technology for blind people. It's true that before personal computers, OCR, and ebooks, some people could get paid to read text aloud or transcribe it to Braille. But, on the whole, I think it's a good thing that those jobs are obsolete, because probably most blind people didn't have access to readers or transcribers. I guess it's possible my reasoning is self-serving though.

2025-06-03 13:34:50 UTC

- reply

> “many of you will lose your jobs but that’s a sacrifice I’m willing to make”
That wasn't what I took from that section. I think the point of that section, and the one after it about plagiarism, was that he thinks we programmers deserve what we've got coming, because of how our work has destroyed so many other kinds of jobs already, and how many of us have flagrantly violated copyright because we want our entertainment now and for free. And both of those sections did give me pause.

2025-05-14 05:18:48 UTC

My #RustWeek conference talk about #AccessKit is coming up in about 3 hours, at 10:25 CET. That's 8:25 UTC, or 3:25 AM US Central time (my usual time zone). It will be live-streamed here: https://rustweek.org/live/wednesday/ My talk is in the main track.

2025-05-01 13:20:43 UTC

- reply

Probably concern about support for proprietary stuff on either side of the OS, meaning hardware and applications. Winner-take-all feedback loops are hard to beat.

2025-04-25 12:36:41 UTC

- reply

FWIW, FastSpring's UI has been stable since at least July 2023, when my SaaS company started using them. If you looked at it more recently than that, then I'd say it hasn't changed in any significant way.

2025-04-25 12:20:34 UTC

- reply

Have you looked at FastSpring?

2025-03-26 15:52:34 UTC

- reply

I think I was starting to head that way with Claude before I quit a couple of weeks ago. Thankfully I didn't use it much for actual coding work; I gave it a few one-off tasks, but was mostly unsatisfied with the results. But I did use it for other things, and I admit, even started to talk to it like a friend sometimes. But a few weeks ago, I reevaluated whether I should use LLMs, and decided to discontinue all personal use. No withdrawal symptoms yet, so maybe I didn't go too far.

2025-03-14 15:27:48 UTC

- reply

I guess at this point I'm just having faith that @npub1hr7…jxyc was right to say, as the title of the second edition of his book puts it, generative models are *bad for business*, not just bad ethically, and that it won't be too long before that becomes impossible to ignore. If that's the case, then we can abstain from LLMs with full confidence that we're not missing out.

2025-03-14 15:08:48 UTC

- reply

I'd be lying, though, if I said I didn't feel some anxiety about the future of my own career. But you and I are in a better place than most, because we both have specialties; yours is web performance, and mine is accessibility.

2025-03-14 14:57:29 UTC

- reply

I think that by abstaining from LLMs and continuing to lean on your decades of expertise, you'll be ahead of the curve when it comes time to clean up after the messes that these reverse centaurs have made. I suppose that could just be wishful thinking though. But regardless, I think we who don't use LLMs should make a conscious effort to keep our skills sharp, in preparation for whatever's coming.

2025-02-17 12:26:29 UTC

- reply

In case it helps, I know a young developer (first or second year in college) who is against the use of LLMs for code generation for at least some of the same reasons you are, and likes to do low-level coding in C or C++. For what it's worth, this developer is blind. This generation isn't entirely a lost cause.

2025-01-18 13:13:17 UTC

- reply

What UI framework (e.g. SwiftUI, Flutter, React Native) is this application using? I want to do what I can to point you in the right direction to resolve the serious accessibility problem pointed out in this thread.

2025-01-11 23:09:09 UTC

- reply

I deleted my Facebook account about 7 years ago, and I don't regret doing so. But, some of my friends at my favorite karaoke bar still use Facebook; they even have a Facebook group for coordinating events at the bar. Perhaps they give me a pass for not being part of that FB group because they imagine there's an accessibility-related reason (not entirely wrong). But I wonder what they/we should use instead. A private web forum? Maybe a Signal group? Something else?

2024-11-27 11:56:59 UTC

- reply

Any ideas about what it would take to make business outcomes depend on software quality? It seems to me that it doesn't naturally happen at small, bootstrapped software companies either, since building a great product isn't sufficient or arguably even necessary to earn sales.

2024-11-16 14:29:24 UTC

I'd like to see less criticism of how Bluesky isn't really decentralized, why it sucks that the Bluesky app uses React (I regret that I briefly piled onto that one), etc., and more discussion of why we're losing people to Bluesky and what we're going to do about it. For my part, I'm trying not to be a toxic reply guy.

2024-10-16 14:17:01 UTC

- reply

I'd like to read that new look at frameworks, and why you disagree with what most of them are trying to do.

2024-10-13 17:04:35 UTC

- reply

Have you decided what you're using instead?
My little company is also redesigning our website. We were already using WordPress, and we've started a new WordPress site. With everything Matt Mullenweg has been doing lately, I'm tempted to suggest using something else, but I don't know what. Having such a wide selection of themes and plugins is certainly convenient.

2024-10-13 16:16:35 UTC

- reply

I'm not sure I understand. Did you mean you categorically rejected WordPress for the new site, or something else?

2024-09-18 16:28:10 UTC

- reply

linenoise (BSD 2-clause): https://github.com/antirez/linenoise

2024-09-02 12:53:53 UTC

- reply

Thank you. The way you argued against site-provided estimated reading times is much better than what I wrote.

2024-09-01 16:57:20 UTC

- reply

The footer has everything I wanted to keep. I decided to drop the latest posts from the home page. People who want to see what's new can follow the link to the blog. But for first-time visitors in particular, the focus is on what the project is and how to get started, information that hopefully won't become out of date quickly.

2024-09-01 16:27:21 UTC

I need feedback from sighted people about the website for my open-source project.
Here's the current version, using a third-party theme; I know the typography on the home page has problems: https://accesskit.dev/
And here's my proposed new version, using a minimalist template and CSS derived from an existing site (I stripped it down quite a bit): https://preview.accesskit.dev/
My main collaborator says the new appearance is from a past era. What do you think? Is it off-putting?

2024-09-01 06:42:31 UTC

Hot take: Websites or publishing platforms including estimated reading time in blog posts is ableist. It says, "We estimate that a normal person reads at this speed, and if you don't, we're going to remind you at the top of every post that you're not normal."
For users who need or want an estimate of reading time, their user agent (e.g. the browser or a browser extension) should do that, in a way that's tailored to them.

2024-08-31 17:30:26 UTC

- reply

Good to know, thanks. Both of us who worked on the site have low vision.

2024-08-31 17:27:43 UTC

- reply

*checks to make sure the static site's theme doesn't do anything egregious in computational or bandwidth cost*

2024-08-31 17:25:54 UTC

- reply

And, come to think of it, why *should* 3 seconds be acceptable, with computers and (some) Internet connections being as fast as they are?

2024-08-31 17:23:44 UTC

- reply

I guess it doesn't really matter for my little open-source project site. But then again, by hosting only in North America, I'd be giving preferential treatment to those in or near North America, right? Shouldn't we care about doing what we can to treat everyone equally?

2024-08-31 17:12:50 UTC

- reply

Isn't it more like hundreds of milliseconds depending on how far the visitor is from the single server?

2024-08-31 16:33:48 UTC

- reply

Context: The AccessKit website (https://accesskit.dev/) is on GitHub Pages, but I'm wondering now if there's a better option.

2024-08-31 16:33:11 UTC

What is the least objectionable content delivery network (CDN) for a fully static website? That is, the CDN from the company that does the fewest bad things. Yes, I could run the site on a single server (dedicated, VPS, or even a machine in my home if I get a static IP address), but that would privilege visitors in a single geographic location, and static sites are of course the ideal use case for a CDN.

2024-08-31 15:00:44 UTC

- reply

Yes, and there's a subtle difference between "The solution is [describe solution]" and "The solution is called [proper noun]". It may be the case that the proper noun is the only thing that (currently) implements the described solution, but it still matches a pattern that has negative associations for some. Perhaps I was overly sensitive, but I'm sure I'm not the only one, so I thought it was something worth keeping in mind.

2024-08-31 12:13:32 UTC

- reply

> The solution is called Rust.
I use and like Rust, but such a definitive statement ("the solution") does sound uncomfortably like religious propaganda (e.g. "Jesus is the answer"). Just saying we have to be careful about how we advocate for Rust, which is indeed a useful tool.

2024-08-31 12:11:02 UTC

Ugh, the maintainer of the Debian bcachefs-tools package just orphaned it because its Rust dependencies are moving too fast. https://jonathancarter.org/2024/08/29/orphaning-bcachefs-tools-in-debian/
So, I need to be careful not to cause that same problem in AccessKit, especially if my current (draft) merge request to use AccessKit in GTK 4 (via the C API) is finished and accepted.

2024-08-31 10:06:31 UTC

About my last boost from @npub1d88…57ex, this makes me ashamed to consider myself part of the Rust community, especially since some of the comments he quoted in the thread were posted in places where the Rust code of conduct are allegedly in force (that is, it's not just Reddit and Hacker News). We've got a problem with tribalism in our community. I know it's not up to me to police the community, but still, what can we do about this?

2024-08-23 17:09:07 UTC

- reply

He also wrote some of the songs for the 2007 romantic comedy Music and Lyrics, including the song "Way Back Into Love" that the two leads cowrite in the movie.

2024-08-19 15:20:16 UTC

- reply

Definitely agree with not adding an audio engine and screen reader into m1n1 or u-boot. I much prefer the Apple approach of having no UI at all in the happy path (other than the logo and sound that happen in all cases), and falling back to a full OS for anything other than the happy path.

2024-08-19 10:13:34 UTC

- reply

What I really meant is that there's no inaccessible firmware/BIOS UI to deal with.

2024-08-19 10:13:04 UTC

- reply

Pretty sure GRUB has no screen reader. And I forgot, Fedora Asahi Remix does include a GRUB menu, so that's a possible challenge.

2024-08-19 09:10:39 UTC

- reply

The Asahi installation process is even accessible with VoiceOver, including the part that runs in the recovery OS. Even the boot volume picker is accessible, since that's running a mini macOS.

2024-08-19 08:48:10 UTC

- reply

If I get serious about developing for PMOS, do you think it would be better for me to use my old Pixel 3 or get a PinePhone or other better-supported hardware? I guess the most important things for screen reader development and testing would be reliable touch input and sound.

2024-08-18 21:23:48 UTC

- reply

Now I want to try PostmarketOS myself sometime. I have a Pixel 3 that's sitting mostly unused, still running the stock Android image. I'll get a newer device for Android testing, then I can put PMOS on the Pixel 3. Would be interesting to hack together a prototype of a screen reader running on GNOME Mobile with touchscreen support.

2024-08-18 20:22:52 UTC

- reply

Is that running Mutter or some other Wayland compositor?

2024-07-26 13:58:50 UTC

- reply

Because employers want people who can manage themselves. Less overhead. We may disagree about whether that's reasonable (I personally think it is), but surely it's important to set up students for success in the world as it is.

2024-07-26 13:29:58 UTC

- reply

Why should it be up to the teacher to break large projects into chunks? Isn't that a skill that the student will need to learn to do for themselves when they want to take on a large project in the workplace?

2024-06-18 11:14:22 UTC

- reply

I'm curious if you've tested any of these apps with Newton yet.

2024-06-17 11:59:30 UTC

- reply

I did see that post. In the accessible name calculation used by the GTK AT-SPI backend, the decision to walk the subtree for some roles (like buttons) but not others (list items) seems to be deliberate. I don't know why. The GTK AccessKit backend is using AccessKit's own name calculation, which currently doesn't walk the subtree for list items either, but I could easily change that.

2024-06-17 11:36:30 UTC

- reply

Thanks. So then, if I remember correctly, the only issue you're having with Fractal now is the lack of expanded state on the tree items, right?

2024-06-16 21:24:46 UTC

- reply

The next time you update, it's crucial that you update newton_atspi_compat (again from the newton_consumer repo), the AccessKit C bindings, and GTK all at the same time, and restart any running GTK 4 apps using the AccessKit backend before you restart Orca. I made a protocol-breaking change. Also, labels now have the correct role.
I'll look at your other reports later. Thanks again.

2024-06-16 16:32:21 UTC

- reply

I think I got flat review working correctly. Please pull the latest update to the newton_consumer repo and rebuild and reinstall the newton_atspi_compat Python package from there.

2024-06-16 15:17:28 UTC

- reply

Those differences between the Arch Linux GTK package and my version when using atspi are weird, because, as of a few days ago, my branch is based on GTK 4.14.4, with few modifications to GTK outside of the AccessKit backend itself.

2024-06-16 11:50:11 UTC

- reply

Please try my latest Orca update and let me know if it improves flat review for you.

2024-06-16 10:31:24 UTC

- reply

Please rebuild the newton_atspi_compat Python package from the newton_consumer repository here: https://gitlab.gnome.org/mwcampbell/newton_consumer
I moved the package so the low-level and high-level Newton consumer crates are consolidated in one repo. And since doing that move, I've pushed a couple of important fixes. Yes, I've updated my Orca README.

2024-06-15 23:58:09 UTC

- reply

Would you please test it with your Newton installation and let me know how it compares?
By the way, with the GTK AccessKit branch, you can easily switch back to the AT-SPI backend by setting the GTK_A11Y environment variable to atspi

2024-06-15 23:46:15 UTC

- reply

I think I heard the name Fractal before, but didn't know what it was. How accessible is Fractal with AT-SPI?

2024-06-15 23:37:39 UTC

- reply

I haven't used gvsbuild yet, but I have built GTK with the Visual Studio tools, just running Meson from a command prompt and letting it pull in dependencies via subprojects. I have also built it using msys2, which is where I've done most of my Windows testing lately, because that was the only way I knew that I could get it working with gnome-text-editor for some real-world GtkTextView testing.

2024-06-15 23:07:37 UTC

- reply

Hi, thanks a lot for building and testing.
Yes, the GTK AccessKit branch is already working on Windows. And it shouldn't be hard to make it work on macOS; I just have to write a bit of platform-specific code and figure out how to build on that platform.
I need to look into the first two bug reports. As for the third one, about stopping and restarting Orca, I think that's actually the same issue I'm working on right now.

2024-05-17 21:34:21 UTC

I recorded a quick demo of the current state of the Wayland-native accessibility stack I'm developing for @npub1dug…kxeq (and eventually other free desktops), code-named Newton. In the demo, I run the GNOME Podcasts app on the Newton accessibility stack, using my GTK branch that integrates AccessKit. So in principle, the same app should also work on Windows, and soon, macOS. The demo was featured in the latest edition of This Week in GNOME: https://thisweek.gnome.org/posts/2024/05/twig-148/ This work is funded by @npub1yng…nhrk.