Aaron Gustafson pokes the fire

WEBVTT
00:00:00.000 --> 00:00:04.500
Okay. Welcome, Aaron. How are you? Thank you
00:00:04.500 --> 00:00:07.120
so much. Good, good. It's been a while. I think
00:00:07.120 --> 00:00:08.960
the last time we hung out in person was like
00:00:08.960 --> 00:00:14.320
2013, maybe, in Sweden. Yes, yes. I remember
00:00:14.320 --> 00:00:18.960
that. That was, was that making of the web or
00:00:18.960 --> 00:00:22.059
something? It was the Funkus conference. Yeah.
00:00:22.839 --> 00:00:27.039
Oh, yes, yes, yes. That was a long time ago.
00:00:27.239 --> 00:00:30.460
Yeah. The thing I do remember about you, because
00:00:30.460 --> 00:00:34.460
we've talked on and off over the years, not often,
00:00:34.600 --> 00:00:41.780
but that you had, you've been the author of a
00:00:41.780 --> 00:00:45.579
number of books, right? Yep. A couple of few.
00:00:45.939 --> 00:00:49.390
A couple of few. How many? Too many. I don't
00:00:49.390 --> 00:00:51.850
know. I mean, I've written for a bunch of books,
00:00:51.890 --> 00:00:54.229
two books of my own, Adaptive Web Design, the
00:00:54.229 --> 00:00:56.409
first edition and the second edition, all about
00:00:56.409 --> 00:00:57.950
progressive enhancement that I've contributed
00:00:57.950 --> 00:01:01.210
to a bunch of books, usually chapters on progressive
00:01:01.210 --> 00:01:04.709
enhancement in JavaScript. And then the latest
00:01:04.709 --> 00:01:07.549
thing, I just contributed to the most recent
00:01:07.549 --> 00:01:10.409
version of Learning Web Design by Jennifer Dean
00:01:10.409 --> 00:01:13.510
Hurst Robbins. So I did a bunch of stuff about
00:01:13.510 --> 00:01:16.700
JavaScript in there for folks as well. I just,
00:01:16.760 --> 00:01:20.159
I remember your books because I did have them,
00:01:20.200 --> 00:01:23.980
but I'll put them somewhere else now. But you
00:01:23.980 --> 00:01:30.319
asked me to do some tech review of it. That's
00:01:30.319 --> 00:01:35.420
how. Yeah. So I was. Way, way back. Yeah, that
00:01:35.420 --> 00:01:40.900
was enjoyable. So now, I mean, you've been around
00:01:40.900 --> 00:01:44.829
for. I won't say too long, but you've been around
00:01:44.829 --> 00:01:47.730
for a long time. I've earned all the gray in
00:01:47.730 --> 00:01:51.510
my beard, is what you're saying. So, but when
00:01:51.510 --> 00:01:55.430
you started off, what were you involved in initially?
00:01:55.989 --> 00:02:00.109
Like, is your background web developer? So I
00:02:00.109 --> 00:02:02.670
kind of came in an interesting way. I think in
00:02:02.670 --> 00:02:05.090
a lot of ways, like people who came in in the
00:02:05.090 --> 00:02:08.370
late 90s, early 2000s did, which was backdoor
00:02:08.370 --> 00:02:12.370
or side door. I don't know. I was a music journalist
00:02:12.370 --> 00:02:18.849
and I was running a zine when I lived in Florida
00:02:18.849 --> 00:02:22.009
and wanted to take that from being just a print
00:02:22.009 --> 00:02:24.169
publication to online. And so I had a friend
00:02:24.169 --> 00:02:26.169
that helped build the first version of the website.
00:02:26.310 --> 00:02:28.080
And I was like, kind of want to learn how to
00:02:28.080 --> 00:02:30.400
do this. And so I taught myself how to do it.
00:02:30.719 --> 00:02:33.000
The first book I learned from was Creating and
00:02:33.000 --> 00:02:35.500
Enhancing Netscape Webpages, which was pretty
00:02:35.500 --> 00:02:43.680
awesome. So this was back in like 96 or so. And
00:02:43.680 --> 00:02:46.719
then I had no idea what CSS was the first time
00:02:46.719 --> 00:02:50.000
I wrote it, but very early on, wrote HTML, wrote
00:02:50.000 --> 00:02:52.699
a little bit of CSS, wrote some Perl to handle
00:02:52.699 --> 00:02:57.240
form processing, things of that nature. And then
00:02:57.240 --> 00:03:00.379
really just fell in love with the media and started
00:03:00.379 --> 00:03:02.500
doing more and more with it and then eventually
00:03:02.500 --> 00:03:05.099
decided that I wanted that to be what I did.
00:03:05.219 --> 00:03:11.780
And so I sort of took that leap in 2000 and started
00:03:11.780 --> 00:03:15.919
doing freelancing work and started working kind
00:03:15.919 --> 00:03:20.400
of through an agency called Aquent, which I think
00:03:20.400 --> 00:03:22.099
is global, but I know they have a big presence
00:03:22.099 --> 00:03:25.520
here in the States. I did work for a lot of large
00:03:25.520 --> 00:03:29.479
companies like Gartner and stuff like that and
00:03:29.479 --> 00:03:33.159
just kind of kept building up my chops. And then
00:03:33.159 --> 00:03:38.199
I became the web lead for an ad agency in Connecticut
00:03:38.199 --> 00:03:42.500
where I was living at the time. And sort of out
00:03:42.500 --> 00:03:45.139
of that and kind of leading that team, found
00:03:45.139 --> 00:03:47.199
my way. I think a couple of years earlier, I
00:03:47.199 --> 00:03:49.729
discovered List Apart. through an article Eric
00:03:49.729 --> 00:03:52.949
Meyer had written on print CSS. And so I'd kind
00:03:52.949 --> 00:03:55.610
of gotten a little bit of the web standards bug.
00:03:55.949 --> 00:03:58.210
And then by the time I had like a team that I
00:03:58.210 --> 00:04:00.830
could run my own way, I started to get more and
00:04:00.830 --> 00:04:04.870
more into accessibility and into CSS and like
00:04:04.870 --> 00:04:06.430
all the things that we could do. So this was
00:04:06.430 --> 00:04:10.629
around about 2003, 2004. Actually, no, sorry.
00:04:10.729 --> 00:04:13.789
It would have been like 2001, 2002. And then
00:04:13.789 --> 00:04:16.319
wrote my... or gave my first talk at a conference
00:04:16.319 --> 00:04:23.019
in 2003 on XHTML. And then, um, met Molly Holschlag,
00:04:23.120 --> 00:04:26.519
uh, at that conference and she and I became instant
00:04:26.519 --> 00:04:28.300
friends and she kind of took me under her wing,
00:04:28.439 --> 00:04:31.279
um, and really helped, helped to grow me into
00:04:31.279 --> 00:04:34.480
the person that I am today. And, um, yeah, I
00:04:34.480 --> 00:04:37.519
wrote my first article in 2004 and then, you
00:04:37.519 --> 00:04:39.240
know, kind of was off to the races from there
00:04:39.240 --> 00:04:43.180
and eventually found myself on the, uh, on the,
00:04:44.079 --> 00:04:46.100
I was going to say the concert scene, but I mean
00:04:46.100 --> 00:04:50.980
the speaking circuit. Yeah, I did a lot of the
00:04:50.980 --> 00:04:54.259
conferences, I would say from about 2000. It
00:04:54.259 --> 00:04:57.319
really kicked in like 2004, 2005. And I did a
00:04:57.319 --> 00:05:01.240
lot of that really up through the pandemic and
00:05:01.240 --> 00:05:04.259
then decided, well, I would say up through about
00:05:04.259 --> 00:05:07.740
2015 when my son was born and then kind of tapered
00:05:07.740 --> 00:05:09.939
off a little bit and then pandemic. And then,
00:05:09.939 --> 00:05:12.339
yeah, I got a little bit more choosy about what
00:05:12.339 --> 00:05:19.160
I was doing. So you worked for small shops, independent.
00:05:20.279 --> 00:05:24.860
When did you start working for larger companies
00:05:24.860 --> 00:05:32.740
like you do now? So I was a freelancer up until
00:05:32.740 --> 00:05:39.879
2005, 2006, somewhere in there. I forget exactly
00:05:39.879 --> 00:05:44.480
when. I think 2005. And then left the ad agency
00:05:44.480 --> 00:05:48.259
that I was at to start my own consultancy. And
00:05:48.259 --> 00:05:51.379
then I ran that for a decade and had a lot of
00:05:51.379 --> 00:05:56.100
large clients, like large investment companies.
00:05:56.160 --> 00:05:59.220
And Microsoft was a client for a project as well.
00:05:59.379 --> 00:06:01.759
And a few things here and there that were kind
00:06:01.759 --> 00:06:04.680
of larger companies. And then kind of got tired
00:06:04.680 --> 00:06:06.720
of running my own consultancy and was like, what
00:06:06.720 --> 00:06:09.060
do I want to do next? And a lot of the folks
00:06:09.060 --> 00:06:11.569
who I... felt were doing the work that I wanted
00:06:11.569 --> 00:06:16.269
to do. People who were evangelizing for web standards
00:06:16.269 --> 00:06:18.850
and for accessibility and stuff like that were
00:06:18.850 --> 00:06:21.529
working as evangelists for browsers. And I was
00:06:21.529 --> 00:06:23.230
like, okay, that's kind of an interesting idea.
00:06:24.089 --> 00:06:31.569
And so this was about 2014 -ish. And then, yeah,
00:06:31.670 --> 00:06:34.430
so I kind of reached out to a couple of browsers,
00:06:34.629 --> 00:06:38.829
talked to them, and really decided that I liked
00:06:38.829 --> 00:06:42.250
the idea of going to like a Microsoft where if
00:06:42.250 --> 00:06:44.350
you think about the Microsoft at the time was
00:06:44.350 --> 00:06:49.290
just the tail end of IE and it was just before
00:06:49.290 --> 00:06:53.449
the launch of Edge. And so, but it was still
00:06:53.449 --> 00:06:56.870
kind of on the back foot. It was very much like
00:06:56.870 --> 00:07:00.069
acting like a scrappy upstart, but with a very
00:07:00.069 --> 00:07:04.060
large. like backing of a huge huge company um
00:07:04.060 --> 00:07:07.100
so i saw it as an opportunity to really get a
00:07:07.100 --> 00:07:11.439
a much different um platform to to speak from
00:07:11.439 --> 00:07:13.779
and you know kind of made it clear to them like
00:07:14.079 --> 00:07:16.019
I'm not going to be here hawking Microsoft product,
00:07:16.180 --> 00:07:18.100
but I'm going to talk about web standards. I'm
00:07:18.100 --> 00:07:20.279
going to help you get better and so on and so
00:07:20.279 --> 00:07:22.600
forth. And I'd had some existing relationships
00:07:22.600 --> 00:07:24.639
with Microsoft from back in the web standards
00:07:24.639 --> 00:07:28.360
project days. And so I wasn't an unknown quantity.
00:07:28.480 --> 00:07:30.199
They kind of knew what they were getting with
00:07:30.199 --> 00:07:34.220
me. And yeah, so I joined in February of 2015
00:07:34.220 --> 00:07:37.199
and thought I'd stay for a year or two. You've
00:07:37.199 --> 00:07:40.560
been there for 11 years. Yeah, I thought I'd
00:07:40.560 --> 00:07:43.079
stay a year or two. And here I am 11 years later.
00:07:43.980 --> 00:07:50.519
So what I've noticed, not with you in particular,
00:07:50.600 --> 00:07:52.879
but just in general, when people join larger
00:07:52.879 --> 00:07:56.259
companies, they can often be swallowed up and
00:07:56.259 --> 00:08:02.420
not, you know, the public persona becomes less.
00:08:02.779 --> 00:08:07.709
Have you found that for yourself? No, I think
00:08:07.709 --> 00:08:10.709
I've had a good set of managers that have allowed
00:08:10.709 --> 00:08:13.949
me to continue to maintain my own visibility
00:08:13.949 --> 00:08:20.870
and my own voice. And each successive manager
00:08:20.870 --> 00:08:22.490
that I've had, I've made it pretty clear that,
00:08:22.569 --> 00:08:25.670
hey, I'm going to continue doing this. And I'm
00:08:25.670 --> 00:08:27.750
going to keep a little bit of a separation between
00:08:27.750 --> 00:08:31.189
events that I do on behalf of Microsoft, where
00:08:31.189 --> 00:08:32.710
Microsoft is paying the way, and events that
00:08:32.710 --> 00:08:37.450
I'm doing as Aaron. I've lucked out and that
00:08:37.450 --> 00:08:40.570
I've been able to maintain that separation. Well,
00:08:40.610 --> 00:08:43.049
that's good because, um, you know, I suppose
00:08:43.049 --> 00:08:46.830
not everybody has, you know, you, you had a sort
00:08:46.830 --> 00:08:50.389
of, uh, a reasonably well -known public persona
00:08:50.389 --> 00:08:53.750
and the, you know, they're already formed before
00:08:53.750 --> 00:09:00.090
you joined the. Yeah. I mean, yeah, I don't know.
00:09:00.149 --> 00:09:02.269
Cause I, you know, some people would work for,
00:09:02.350 --> 00:09:04.549
for large companies. I didn't even notice because
00:09:04.549 --> 00:09:12.899
they're so. They talk a lot, regardless of where
00:09:12.899 --> 00:09:16.080
they are. But then you've got people that work
00:09:16.080 --> 00:09:19.940
for interesting people to interact with, follow
00:09:19.940 --> 00:09:24.500
publicly, that go and work for someone like Apple.
00:09:24.559 --> 00:09:27.539
You never hear from them again. Yeah, I think
00:09:27.539 --> 00:09:29.340
it's the nature of the company that you end up
00:09:29.340 --> 00:09:31.299
going to work for or the team that you work for
00:09:31.299 --> 00:09:36.470
and stuff. So you still got a job? unfortunately
00:09:36.470 --> 00:09:41.509
fired a load of people. Yeah, yeah. It's definitely
00:09:41.509 --> 00:09:46.330
the restructuring is not awesome. It's not awesome.
00:09:48.210 --> 00:09:51.470
But at the same time, there's a ton of positions
00:09:51.470 --> 00:09:54.110
that are opening up. I mean, I've been lucky
00:09:54.110 --> 00:09:57.850
to kind of find my way to good media projects.
00:09:57.950 --> 00:10:02.539
And lately I've been focused a lot on... Shifting
00:10:02.539 --> 00:10:05.519
left tooling. So really focusing on accessibility
00:10:05.519 --> 00:10:08.059
tooling for the enterprise and getting what I
00:10:08.059 --> 00:10:12.399
can up into the... Yeah. So I guess probably
00:10:12.399 --> 00:10:16.659
about maybe a year and a half ago. So I'll take
00:10:16.659 --> 00:10:17.899
a step back actually a little further than that.
00:10:18.000 --> 00:10:21.360
So when coding models started to become a thing,
00:10:21.460 --> 00:10:25.179
I started looking at, okay, let's do some evaluations
00:10:25.179 --> 00:10:27.639
of how good is the quality of the code that these
00:10:27.639 --> 00:10:29.620
are producing. It's pretty garbage, which is
00:10:29.620 --> 00:10:31.039
not surprising because it was trained on the
00:10:31.039 --> 00:10:32.659
garbage that we've been producing for decades,
00:10:32.919 --> 00:10:37.240
right? So kind of started through that analysis,
00:10:37.419 --> 00:10:44.379
ended up really getting into looking at how Microsoft
00:10:44.379 --> 00:10:46.360
as an enterprise is doing the software development
00:10:46.360 --> 00:10:49.580
lifecycle, where we've got investments in tooling
00:10:49.580 --> 00:10:51.840
for accessibility, how can we accelerate things,
00:10:52.139 --> 00:10:57.100
shifting it further and further left. and identifying
00:10:57.100 --> 00:10:59.220
just kind of opportunities across that spectrum.
00:10:59.759 --> 00:11:03.259
And then one of the opportunities I identified
00:11:03.259 --> 00:11:05.559
is we had a plugin that we'd created for Figma
00:11:05.559 --> 00:11:07.539
that was used for annotating like focus order
00:11:07.539 --> 00:11:10.480
and stuff like that on designs. And it was pretty
00:11:10.480 --> 00:11:13.700
good, but it was written in 2017. And I think
00:11:13.700 --> 00:11:15.639
it had a little bit of updates over the years,
00:11:15.820 --> 00:11:18.220
but I think the last major update had been like
00:11:18.220 --> 00:11:21.720
2024 or something like that. And it really wasn't
00:11:21.720 --> 00:11:24.690
doing all that I thought that we could do. And
00:11:24.690 --> 00:11:30.009
so I put together a proposal for recasting that
00:11:30.009 --> 00:11:34.950
as a tool that could really help designers to
00:11:34.950 --> 00:11:37.129
create accessibility specs, but at the same time
00:11:37.129 --> 00:11:41.370
help to upskill them to help them better understand
00:11:41.370 --> 00:11:43.769
accessibility and how it applies to the work
00:11:43.769 --> 00:11:46.769
that they're doing. And so I've been leading
00:11:46.769 --> 00:11:50.879
that project for, I think it's been about nine
00:11:50.879 --> 00:11:53.659
months of of actual like engineering work and
00:11:53.659 --> 00:11:54.940
stuff like that and doing a bit of the engineering
00:11:54.940 --> 00:12:01.860
myself um and new release product you need to
00:12:01.860 --> 00:12:05.159
buy was it a product it's free free profit free
00:12:05.159 --> 00:12:10.039
product um and we've we've done one uh one public
00:12:10.039 --> 00:12:12.820
release uh so far of the updated version which
00:12:12.820 --> 00:12:17.440
um brought on a bunch of new uh w3c roles and
00:12:17.440 --> 00:12:19.259
stuff like that and kind of adjusted how it was
00:12:19.259 --> 00:12:23.179
handling the um the annotations doing actual
00:12:23.179 --> 00:12:26.179
like native uh figma annotations as opposed to
00:12:26.179 --> 00:12:30.429
doing annotations as like a layer So just from
00:12:30.429 --> 00:12:32.009
a management standpoint, a little bit easier
00:12:32.009 --> 00:12:34.309
to kind of turn on and off and adjust things
00:12:34.309 --> 00:12:37.269
that way. But the next release, which hopefully
00:12:37.269 --> 00:12:39.850
will be coming out in the, I would say next couple
00:12:39.850 --> 00:12:42.850
of weeks, is going to be a complete overhaul.
00:12:42.950 --> 00:12:44.529
And so what it does is it actually takes you
00:12:44.529 --> 00:12:46.490
through the process of like you say, okay, here's
00:12:46.490 --> 00:12:48.950
this project I'm going to work on. I'm going
00:12:48.950 --> 00:12:52.110
to name it and I can maybe connect it to GitHub
00:12:52.110 --> 00:12:54.830
or to an ADO work item or something like that.
00:12:56.289 --> 00:12:59.110
And here's the platform that it's for. Here's
00:12:59.110 --> 00:13:02.409
the design system that I'm using. And then it
00:13:02.409 --> 00:13:04.809
walks you through the process of annotating that
00:13:04.809 --> 00:13:07.950
design and then walks you through the process
00:13:07.950 --> 00:13:10.669
of assessing that design for overall accessibility
00:13:10.669 --> 00:13:14.529
and usability. And through that process, instead
00:13:14.529 --> 00:13:16.370
of it being just kind of a manual checklist of
00:13:16.370 --> 00:13:19.129
like, yeah, that seems like it worked. Some of
00:13:19.129 --> 00:13:21.250
those checks we've moved into being automated
00:13:21.250 --> 00:13:24.090
checks that can run right in the plugin. So for
00:13:24.090 --> 00:13:26.389
things like color contrast, even like gradient
00:13:26.389 --> 00:13:28.149
against gradient and stuff like that, we can
00:13:28.149 --> 00:13:30.950
check contrast. We can check for things like
00:13:30.950 --> 00:13:33.570
images of text, all of those sorts of things
00:13:33.570 --> 00:13:37.289
just directly within the Figma plugin. And then
00:13:37.289 --> 00:13:39.529
you can, if there are issues that are found,
00:13:39.649 --> 00:13:41.669
you can track those issues all within the context
00:13:41.669 --> 00:13:43.330
of the plugin so you're not having to go to some
00:13:43.330 --> 00:13:48.580
separate system to manage that. Once you've fixed
00:13:48.580 --> 00:13:50.279
those problems and done all of the assessments,
00:13:50.460 --> 00:13:52.480
your project is considered ready and you can
00:13:52.480 --> 00:13:54.980
actually export that as an accessibility spec
00:13:54.980 --> 00:13:57.980
as an HTML document or copy it and drop it into
00:13:57.980 --> 00:14:02.340
a work item if you wanted to have an engineer
00:14:02.340 --> 00:14:05.879
pick it up. Or if you're working with an AI coding
00:14:05.879 --> 00:14:07.820
agent, you can actually give it explicit instructions
00:14:07.820 --> 00:14:09.639
as to like, here's how to build this and here's
00:14:09.639 --> 00:14:11.360
what the expectations are for accessibility.
00:14:13.120 --> 00:14:16.220
And we've done some preliminary testing of handing
00:14:16.220 --> 00:14:19.559
off those accessibility specs to an agent to
00:14:19.559 --> 00:14:23.320
do remediation work. And it's able to take just
00:14:23.320 --> 00:14:25.600
a single example, let's say in a complex pattern
00:14:25.600 --> 00:14:29.139
like a grid, and it's actually able to see that
00:14:29.139 --> 00:14:31.580
example and then extrapolate that to all the
00:14:31.580 --> 00:14:33.960
rows in the grid, assigning appropriate names,
00:14:34.120 --> 00:14:35.759
descriptions, all that sort of stuff, which is
00:14:35.759 --> 00:14:42.879
pretty impressive. It sounds great. I mean, you
00:14:42.879 --> 00:14:47.379
may or may not be aware, but we do a fair bit
00:14:47.379 --> 00:14:51.559
of consulting at Tetralogical, and we have people
00:14:51.559 --> 00:14:55.940
that work with the designers who are accessibility
00:14:55.940 --> 00:14:59.600
designers, consultants or whatever. So that sounds
00:14:59.600 --> 00:15:01.440
like something we work with Figma. What's the
00:15:01.440 --> 00:15:03.539
other? There's Figma and there's something else.
00:15:04.580 --> 00:15:07.700
There's another design thing that is used, but
00:15:07.700 --> 00:15:11.559
I think Figma is the main one. But, yeah, it
00:15:11.559 --> 00:15:13.919
definitely sounds like something that we should
00:15:13.919 --> 00:15:19.860
look into. One of the things that I do is I'll
00:15:19.860 --> 00:15:24.980
gather some links for when I publish this. So
00:15:24.980 --> 00:15:27.419
if you can send me a link to that info, it would
00:15:27.419 --> 00:15:31.399
be great, or anything else that you want. So
00:15:31.399 --> 00:15:34.700
you'll be doing that. And what other – so it
00:15:34.700 --> 00:15:39.299
does have – Does it use AI to actually produce
00:15:39.299 --> 00:15:45.299
anything or do? So the public release right now
00:15:45.299 --> 00:15:49.000
won't have any AI related features. The main
00:15:49.000 --> 00:15:51.659
reason for that is this is a free plugin and
00:15:51.659 --> 00:15:55.860
we don't really have. like honestly pay for that
00:15:55.860 --> 00:15:57.659
you know i haven't been able to find a team that's
00:15:57.659 --> 00:15:59.779
like okay you know we'll we'll pay for this to
00:15:59.779 --> 00:16:01.740
have ai features and we've been offering it for
00:16:01.740 --> 00:16:04.980
free so um we've talked about so we have some
00:16:04.980 --> 00:16:06.659
ai features that will be part of the internal
00:16:06.659 --> 00:16:09.399
release of it right um and we've talked about
00:16:09.399 --> 00:16:12.860
potentially um setting it up so that folks could
00:16:12.860 --> 00:16:16.279
basically bring their own ai like you know oh
00:16:16.279 --> 00:16:20.620
yes yeah something like that i'm not sure how
00:16:20.620 --> 00:16:23.259
that would work yet you know i think tbd on on
00:16:23.259 --> 00:16:28.639
on that i mean it would be really cool yeah yeah
00:16:28.639 --> 00:16:32.679
i mean i don't uh what what are the things that
00:16:32.679 --> 00:16:36.899
ai brings the table that couldn't be done otherwise
00:16:36.899 --> 00:16:41.590
yeah So one of the things that we've been doing
00:16:41.590 --> 00:16:45.789
is like assisted heuristic analysis based on
00:16:45.789 --> 00:16:50.179
disability personas. Yeah, that's something that
00:16:50.179 --> 00:16:53.320
we're working on. We'll see where it goes. But
00:16:53.320 --> 00:16:56.159
I'm able to do, for one of the internal things,
00:16:56.500 --> 00:16:59.419
I can have it do a review of the design. And
00:16:59.419 --> 00:17:01.860
in the plugin, it'll actually attach not only
00:17:01.860 --> 00:17:05.099
the Figma node tree, but a screenshot of it and
00:17:05.099 --> 00:17:08.119
pass that over and get feedback on it. So there's
00:17:08.119 --> 00:17:10.200
some of the stuff that we're able to start getting.
00:17:11.019 --> 00:17:14.059
a bit more diagnosis of like, okay, this makes
00:17:14.059 --> 00:17:16.299
sense, or maybe this word could have been a better
00:17:16.299 --> 00:17:19.000
choice or something like that. But honestly,
00:17:19.119 --> 00:17:20.640
a lot of the tools that we've built into it,
00:17:20.680 --> 00:17:24.740
things like being able to within, so there are
00:17:24.740 --> 00:17:28.519
these concepts of annotation sets. So if you
00:17:28.519 --> 00:17:31.400
imagine that a project is encompassing several
00:17:31.400 --> 00:17:33.799
screens, you could do an annotation set for each
00:17:33.799 --> 00:17:37.160
screen. And then when you're annotating that,
00:17:37.220 --> 00:17:38.700
you've got certain items that are focusable,
00:17:38.740 --> 00:17:42.099
certain items that are maybe not. And we have
00:17:42.099 --> 00:17:43.880
a visualizer now that you can click and it'll
00:17:43.880 --> 00:17:45.680
actually do kind of like accessibility insights
00:17:45.680 --> 00:17:48.440
does do the line that takes you from, you know,
00:17:48.440 --> 00:17:49.839
one to two to three, four. So you can actually
00:17:49.839 --> 00:17:53.019
visualize what the, what the, the focus order
00:17:53.019 --> 00:17:55.279
is and you don't have to do a darn thing except
00:17:55.279 --> 00:17:57.099
click a button and it'll just visualize it for
00:17:57.099 --> 00:18:00.539
you. So just little, little things like that
00:18:00.539 --> 00:18:04.099
that can help designers to, to really tune, tune
00:18:04.099 --> 00:18:06.680
their thing as they're working and have less
00:18:06.680 --> 00:18:09.930
rework. And, and ultimately. The way that I kind
00:18:09.930 --> 00:18:14.289
of think about it is the costs overall of having
00:18:14.289 --> 00:18:17.309
to fix bugs, any bugs, but accessibility bugs
00:18:17.309 --> 00:18:20.480
certainly. It's zero if you do it in planning.
00:18:20.759 --> 00:18:23.700
It's maybe like $100 if you're doing it in the
00:18:23.700 --> 00:18:25.640
design phase, about $1 ,000 if you're doing it
00:18:25.640 --> 00:18:27.359
during engineering, and then $10 ,000 if it makes
00:18:27.359 --> 00:18:30.680
it into production as a bug. So that scale just
00:18:30.680 --> 00:18:35.240
goes up pretty incredibly. So if we can catch
00:18:35.240 --> 00:18:37.720
stuff at the beginning, it's much cheaper and
00:18:37.720 --> 00:18:41.440
easier to change those things early on. and it
00:18:41.440 --> 00:18:43.299
ends up saving, you know, obviously a ton of
00:18:43.299 --> 00:18:45.019
money. But the other thing that it saves is for
00:18:45.019 --> 00:18:47.000
the accessibility subject matter experts that
00:18:47.000 --> 00:18:48.759
you have on your team that are having to do the
00:18:48.759 --> 00:18:50.819
accessibility reviews before a product goes out,
00:18:51.359 --> 00:18:55.019
they can focus on the actual, like the gnarly,
00:18:55.019 --> 00:18:57.960
complicated interactions and not be dealing with
00:18:57.960 --> 00:19:01.160
like the very baseline stuff. Yeah, exactly.
00:19:01.500 --> 00:19:03.220
Just the noise that should have been caught,
00:19:03.400 --> 00:19:10.000
right? Yeah. So you're saying that there's, what,
00:19:11.150 --> 00:19:13.569
In the public one, there's not going to be any
00:19:13.569 --> 00:19:17.529
AI. What are the particular things that AI, that
00:19:17.529 --> 00:19:22.869
you see as potential for accessory testing? Because,
00:19:22.869 --> 00:19:25.769
I mean, from my point of view, I've looked at
00:19:25.769 --> 00:19:31.190
it, I haven't really seen much, if anything,
00:19:31.410 --> 00:19:33.549
that at this point in time, that is actually,
00:19:33.650 --> 00:19:38.710
will actually make my job easier. Yeah. So I
00:19:38.710 --> 00:19:40.859
think there's... The way that I kind of look
00:19:40.859 --> 00:19:43.140
at it is that I think there's an opportunity,
00:19:43.359 --> 00:19:47.140
if you think of it like bowling, there's an opportunity
00:19:47.140 --> 00:19:51.220
to put the little bumpers in the gutters so that
00:19:51.220 --> 00:19:54.240
you end up with more of the balls going down
00:19:54.240 --> 00:19:55.759
towards the pins as opposed to ending up in the
00:19:55.759 --> 00:19:57.460
gutter, right? So it's about putting guardrails
00:19:57.460 --> 00:20:02.079
on the system. At the same time there, sorry,
00:20:02.220 --> 00:20:04.220
but at the same time, when you put guardrails
00:20:04.220 --> 00:20:07.680
on the system, that means that people don't learn.
00:20:08.340 --> 00:20:11.259
as well do they like no it's absolutely true
00:20:11.259 --> 00:20:14.059
but simultaneously you could probably argue that
00:20:14.059 --> 00:20:16.579
a lot of people haven't learned it anyway that's
00:20:16.579 --> 00:20:20.180
true and so so you know if we can you know i've
00:20:20.180 --> 00:20:22.880
always looked at these things as you know how
00:20:22.880 --> 00:20:25.359
can we do kind of a wraparound how can we provide
00:20:25.359 --> 00:20:27.119
better support for the people who are doing the
00:20:27.119 --> 00:20:29.160
right thing in order to be able to not be doing
00:20:29.160 --> 00:20:31.519
as much busy work and to get to the the right
00:20:31.519 --> 00:20:34.599
end point but also how can we for all the people
00:20:34.599 --> 00:20:36.730
who aren't going to do that um that are just
00:20:36.730 --> 00:20:39.349
like clock in clock out i'm just you know churning
00:20:39.349 --> 00:20:41.930
code or whatever how can we do an end run around
00:20:41.930 --> 00:20:46.369
them so that they're they're not producing bad
00:20:46.369 --> 00:20:50.240
code How can we kind of force them to produce
00:20:50.240 --> 00:20:51.779
stuff that's going to be good? Because there's
00:20:51.779 --> 00:20:53.240
always going to be people who are interested
00:20:53.240 --> 00:20:54.460
in learning. There's going to be people who aren't
00:20:54.460 --> 00:20:56.880
interested in learning. And this was the same
00:20:56.880 --> 00:20:59.240
thing in the web standards space. It was, how
00:20:59.240 --> 00:21:01.720
do we get the concept of web standards? How do
00:21:01.720 --> 00:21:03.460
we get CSS? How do we get people to adopt these?
00:21:03.819 --> 00:21:06.299
And really it was sort of like, we got to go
00:21:06.299 --> 00:21:08.759
from both sides and we got to find a way to do.
00:21:09.549 --> 00:21:12.029
an end run around the people who aren't going
00:21:12.029 --> 00:21:14.230
to do the right thing. And how do we provide
00:21:14.230 --> 00:21:17.269
support for the people who are? So, you know,
00:21:17.269 --> 00:21:18.809
some of the things that I think are really interesting
00:21:18.809 --> 00:21:22.890
is potentially being able to write better testing
00:21:22.890 --> 00:21:26.430
software to be able to, and to be able to do
00:21:26.430 --> 00:21:29.710
so really at a scale that is really challenging
00:21:29.710 --> 00:21:31.950
with engineering and especially with budgets
00:21:31.950 --> 00:21:34.220
getting tighter and tighter. So I think that's
00:21:34.220 --> 00:21:36.539
an interesting space. But one of the ones that
00:21:36.539 --> 00:21:38.720
I'm really excited about, and it's maybe a bit
00:21:38.720 --> 00:21:42.700
off still, but if you think about, you know,
00:21:42.700 --> 00:21:45.039
Jeremy always talks about top tasks and like,
00:21:45.160 --> 00:21:47.079
you know, for a given interface, these are the
00:21:47.079 --> 00:21:48.759
key things that somebody needs to be able to
00:21:48.759 --> 00:21:54.579
do. If I can point a computer use agent at that
00:21:54.579 --> 00:21:57.799
particular URL and say, okay, your task here
00:21:57.799 --> 00:22:00.420
is to, you know, book a plane ticket from here
00:22:00.420 --> 00:22:04.549
to here. and have it be able to go through and
00:22:04.549 --> 00:22:07.369
try and do that process and highlight for me,
00:22:07.470 --> 00:22:09.849
where did it run into problems? Where did it
00:22:09.849 --> 00:22:11.589
hit a keyboard trap? Where was it not able to
00:22:11.589 --> 00:22:15.990
reach something? And to be able to have some
00:22:15.990 --> 00:22:19.029
help to recover if it gets stuck somewhere. But
00:22:19.029 --> 00:22:22.170
to give me that information as feedback, that's
00:22:22.170 --> 00:22:25.230
pretty valuable. But beyond that, being able
00:22:25.230 --> 00:22:28.549
to take that that run through for that particular
00:22:28.549 --> 00:22:32.089
top task and then translate it from that experience
00:22:32.089 --> 00:22:34.750
that the agent had completing that task to, let's
00:22:34.750 --> 00:22:37.309
say, a playwright end -to -end test that you
00:22:37.309 --> 00:22:40.589
can then run and check for regressions. It allows
00:22:40.589 --> 00:22:46.130
you to really scale the... scale the the experience
00:22:46.130 --> 00:22:48.509
and the amount of testing that you can do while
00:22:48.509 --> 00:22:50.730
at the same time ending up with a deterministic
00:22:50.730 --> 00:22:53.150
endpoint so you actually know that you're getting
00:22:53.150 --> 00:22:55.490
what you should be getting yeah yeah that's it's
00:22:55.490 --> 00:22:59.349
interesting using it to to run these scenarios
00:22:59.349 --> 00:23:01.769
and capturing the day but the scenarios and then
00:23:01.769 --> 00:23:05.230
running the scenarios in playwright yeah no no
00:23:05.230 --> 00:23:10.470
that that sounds like a good idea I mean, obviously
00:23:10.470 --> 00:23:12.630
it needs to be validated by a human, but at the
00:23:12.630 --> 00:23:15.869
same time, it gives you a starting point to be
00:23:15.869 --> 00:23:19.690
able to do these things. And the reality is we
00:23:19.690 --> 00:23:24.589
are producing so many interfaces that there aren't
00:23:24.589 --> 00:23:26.670
enough accessibility testers out there to be
00:23:26.670 --> 00:23:29.549
able to hire them all to do the number of scenarios
00:23:29.549 --> 00:23:35.329
that we need to do. And especially at the speed
00:23:35.329 --> 00:23:37.799
at which we could. could do it as well when well
00:23:37.799 --> 00:23:40.700
that's the the problem being there is that there's
00:23:40.700 --> 00:23:45.279
a huge more amounts of code i mean i can i can
00:23:45.279 --> 00:23:49.440
create code with gpt and and i mean i'm aware
00:23:49.440 --> 00:23:54.039
that the of the user interface aspects but it
00:23:54.039 --> 00:23:56.380
just it takes you moments to create something
00:23:56.380 --> 00:23:59.650
and then you know put it out into the world,
00:23:59.670 --> 00:24:01.609
then I can create something else. So there's
00:24:01.609 --> 00:24:05.170
lots and lots of code being generated. And this
00:24:05.170 --> 00:24:07.750
is the same for large companies. I know that
00:24:07.750 --> 00:24:11.609
people are going mad on it. But unfortunately,
00:24:11.789 --> 00:24:20.029
the amount of code coming out is, as you say,
00:24:20.210 --> 00:24:24.210
I mean, testing of it is limited by the amount
00:24:24.210 --> 00:24:27.890
of people currently. Automation will hopefully
00:24:27.890 --> 00:24:32.329
fill that gap, but you just have to wait and
00:24:32.329 --> 00:24:35.069
see. What I was going to ask you, though, was
00:24:35.069 --> 00:24:40.410
that you were talking about agents doing stuff.
00:24:40.890 --> 00:24:48.569
Now, just the practicality of handing my agency
00:24:48.569 --> 00:24:53.069
off to a third party to go and buy me a ticket,
00:24:53.150 --> 00:24:57.960
I wouldn't trust. Oh, no. Yeah. No, no, I'm with
00:24:57.960 --> 00:25:00.460
you. I'm not signing up to have an agent do shopping
00:25:00.460 --> 00:25:03.519
for me and stuff like that. But to run scenarios.
00:25:03.900 --> 00:25:07.880
Yeah. It's a potential thing for people with
00:25:07.880 --> 00:25:11.720
disabilities. Yeah, I was going to say, one of
00:25:11.720 --> 00:25:13.799
my colleagues has been building himself utilities
00:25:13.799 --> 00:25:17.400
within Windows to do different things for him
00:25:17.400 --> 00:25:19.619
to make it easier. And so I think that's actually,
00:25:19.839 --> 00:25:23.099
you're talking about the ability to create coded
00:25:23.099 --> 00:25:25.259
scale. I think one of the really interesting
00:25:25.259 --> 00:25:27.940
opportunities with coding agents is being able
00:25:27.940 --> 00:25:31.480
to create tools for yourself where the audience
00:25:31.480 --> 00:25:35.839
is one. So I have a colleague who's blind who's
00:25:35.839 --> 00:25:39.380
made himself his own email reader. And so he
00:25:39.380 --> 00:25:42.259
built it to be totally the experience that he
00:25:42.259 --> 00:25:43.960
wants to have with his screen reader of reading
00:25:43.960 --> 00:25:47.839
emails. And yes, he also has to use Outlook because
00:25:47.839 --> 00:25:51.700
he's at Microsoft. But for his experience, what
00:25:51.700 --> 00:25:54.799
he needed was just that. So the idea of being
00:25:54.799 --> 00:25:57.220
able to create really bespoke software for yourself
00:25:57.220 --> 00:26:01.119
where you're the only audience is a really interesting
00:26:01.119 --> 00:26:03.400
opportunity, especially for people with disabilities
00:26:03.400 --> 00:26:07.539
who honestly have been the hackers for a really
00:26:07.539 --> 00:26:10.339
long time, both the built environment and the
00:26:10.339 --> 00:26:12.480
digital environment. And so I think that's a
00:26:12.480 --> 00:26:16.339
really interesting opportunity because... much
00:26:16.339 --> 00:26:18.859
like the web democratize the ability to publish
00:26:18.859 --> 00:26:20.619
your thoughts and share those for good in that,
00:26:20.740 --> 00:26:24.240
right? You know, I think this is really putting
00:26:24.240 --> 00:26:26.059
tools in the hands of people with disabilities
00:26:26.059 --> 00:26:28.000
to be able to create the tools that they need
00:26:28.000 --> 00:26:30.839
for themselves. And, you know, who knows what
00:26:30.839 --> 00:26:32.400
that's going to end up creating that ends up
00:26:32.400 --> 00:26:34.400
benefiting everyone, right? I get there's, you
00:26:34.400 --> 00:26:37.250
know, the more... there are out there, the more
00:26:37.250 --> 00:26:39.269
opportunities there are for people to create.
00:26:39.309 --> 00:26:42.670
Yeah, I mean, I can see that aspect to me. It's
00:26:42.670 --> 00:26:45.490
such a, you know, topic that's fraught with or
00:26:45.490 --> 00:26:49.690
such a subject that's fraught with concerns about
00:26:49.690 --> 00:26:54.490
its fuel use, you know, its effect upon. So it's
00:26:54.490 --> 00:27:00.910
really, I mean, I think I've described it. recently
00:27:00.910 --> 00:27:03.849
is a triple ed sword i mean it's just difficult
00:27:03.849 --> 00:27:07.470
because i i find it really interesting and and
00:27:07.470 --> 00:27:13.130
um inquisitive but at the same time i yeah yeah
00:27:13.130 --> 00:27:15.849
i mean my my hope in that stuff is that we eventually
00:27:15.849 --> 00:27:17.990
end up in a place where the models are running
00:27:17.990 --> 00:27:21.589
locally on my own machines yes and that that
00:27:21.589 --> 00:27:24.490
to me is sort of the that's the end all be all
00:27:24.490 --> 00:27:27.759
and like that's To take us in a total other direction
00:27:27.759 --> 00:27:30.519
for a moment, another area that I've been working
00:27:30.519 --> 00:27:34.559
in is the synthetic speech space. So for the
00:27:34.559 --> 00:27:37.019
last three years, I've been working with Team
00:27:37.019 --> 00:27:39.579
Gleason, which is an ALS support organization,
00:27:39.759 --> 00:27:43.500
or MND, as they know in the UK. And we've been
00:27:43.500 --> 00:27:47.039
working on taking recordings of someone's voice
00:27:47.039 --> 00:27:49.720
from before they lost it to dysarthria and being
00:27:49.720 --> 00:27:52.940
able to create an AI voice or a synthetic version
00:27:52.940 --> 00:27:56.339
of their voice. So for people who are trying
00:27:56.339 --> 00:27:58.500
to imagine this, you know, think Stephen Hawking,
00:27:58.599 --> 00:28:00.599
but much better quality, something that actually
00:28:00.599 --> 00:28:04.740
sounds like them. Right. And so we started working
00:28:04.740 --> 00:28:09.900
on that about three years ago. We were working
00:28:09.900 --> 00:28:11.920
with them. We did a pilot with about 30 individuals.
00:28:14.690 --> 00:28:18.190
The impact on them and their loved ones of being
00:28:18.190 --> 00:28:20.789
able to hear their voice again after, in some
00:28:20.789 --> 00:28:23.609
cases, like a decade of not hearing them speak
00:28:23.609 --> 00:28:26.289
was pretty transformative and really powerful.
00:28:26.470 --> 00:28:30.269
But to me, the real opportunity there is not
00:28:30.269 --> 00:28:32.549
cloud -based voices being created, but actually
00:28:32.549 --> 00:28:34.250
it needs to be running on the device because
00:28:34.250 --> 00:28:37.309
you can't have somebody go to get on an airplane
00:28:37.309 --> 00:28:39.769
or something like that and not be able to communicate,
00:28:40.109 --> 00:28:44.299
right? Yeah. So, you know, I think the potential
00:28:44.299 --> 00:28:46.819
of these onboard models and being able to run
00:28:46.819 --> 00:28:49.460
things quickly, run very, like, computationally
00:28:49.460 --> 00:28:52.579
intensive processes locally on a machine is really
00:28:52.579 --> 00:28:55.480
compelling and really going to change a lot of
00:28:55.480 --> 00:29:00.220
people's lives. Yeah, yeah. Well, I was watching
00:29:00.220 --> 00:29:03.880
something, Patrick Lauco, my colleague and friend,
00:29:03.980 --> 00:29:08.299
shared something from W3C the other day. It was
00:29:08.299 --> 00:29:12.920
a... It was a talk that happened at TPAC in China.
00:29:13.559 --> 00:29:19.180
It was in Chinese, but the slides were in English.
00:29:19.220 --> 00:29:22.640
But essentially, it's about a Gen Tech AI, like
00:29:22.640 --> 00:29:27.200
tiny, it's called tiny Gen UI or something. It's
00:29:27.200 --> 00:29:32.559
some sort of where they create, you know, customised
00:29:32.559 --> 00:29:37.880
interfaces on the fly. Yeah. And I thought, well,
00:29:38.000 --> 00:29:40.579
I'll be just in trying that. But then, yeah,
00:29:40.700 --> 00:29:47.079
the process of installing a local LLM and those
00:29:47.079 --> 00:29:51.500
sort of things, it's, yeah, it's a bit of a little
00:29:51.500 --> 00:29:54.420
cumbersome right now. Yeah. Yeah. And it can
00:29:54.420 --> 00:29:56.420
really, if you don't know what you're doing,
00:29:56.440 --> 00:29:59.079
it can really also like. hosier machine your
00:29:59.079 --> 00:30:01.759
fan spinning up really really bad and stuff yeah
00:30:01.759 --> 00:30:04.740
yeah yeah so but that'll that'll improve and
00:30:04.740 --> 00:30:06.400
people are working on making smaller and smaller
00:30:06.400 --> 00:30:10.220
models and um you know the efficiency will improve
00:30:10.220 --> 00:30:16.940
over time so sure um let's just do the uh fireside
00:30:16.940 --> 00:30:22.339
i actually had this fireside uh i produced it
00:30:22.339 --> 00:30:25.819
when when we we uh initially agreed to meet and
00:30:25.819 --> 00:30:30.089
then uh We've had subsequent... Yeah, sorry about
00:30:30.089 --> 00:30:34.109
last week. My daughter needed a haircut and I
00:30:34.109 --> 00:30:37.309
had to... 30 miles, that's all I had to drive.
00:30:37.569 --> 00:30:41.769
Took me almost two hours to get there. The traffic
00:30:41.769 --> 00:30:46.569
in the UK is horrible. It's an overpopulated
00:30:46.569 --> 00:30:50.670
island. Well, overpopulated cars at least. Okay,
00:30:50.890 --> 00:31:01.250
so let me just share this now. I can? Okay. So
00:31:01.250 --> 00:31:04.509
you should see the fireside. You haven't got
00:31:04.509 --> 00:31:06.789
a huge round of people, and three of them are
00:31:06.789 --> 00:31:12.250
you, so you can't. So first up, we've got a double
00:31:12.250 --> 00:31:17.930
O. Do you know either of these people? Oh, gosh.
00:31:19.710 --> 00:31:24.089
It's all right. No, I don't. I don't. It's hard
00:31:24.089 --> 00:31:26.609
to see them because they're so small. Oh, yeah.
00:31:28.349 --> 00:31:30.690
Oh, there we go. Let's see. Yeah, it still doesn't
00:31:30.690 --> 00:31:36.029
help, does it? Yeah, it's a little pixely. These
00:31:36.029 --> 00:31:40.730
people are always first and second on the Fireside
00:31:40.730 --> 00:31:45.609
Rose Gallery. I was going to say, is that Patrick
00:31:45.609 --> 00:31:48.410
on the left? Yeah, Patrick on the left and Dr.
00:31:48.509 --> 00:31:50.869
Swallow on the right. Okay, I don't think I know
00:31:50.869 --> 00:31:54.869
Dr. Swallow. Yeah, I meet up with him every couple
00:31:54.869 --> 00:31:58.779
of months. This is in Preston. So where Pat lives.
00:31:59.039 --> 00:32:01.960
Gotcha. So we get together for just hang out.
00:32:02.220 --> 00:32:06.460
So the three of us used to all work at TPG together.
00:32:06.740 --> 00:32:10.259
Yep. Dave still works at TPG. Obviously, Pat
00:32:10.259 --> 00:32:16.859
and I don't. Yeah, yeah. Yeah, just an old friendship.
00:32:17.839 --> 00:32:23.589
Who's up next? I don't know. In the middle. So
00:32:23.589 --> 00:32:25.769
you're talking about the photo with me with the
00:32:25.769 --> 00:32:28.589
coffee cup in my hand? Yes. So you've got Chris
00:32:28.589 --> 00:32:32.809
Wilson and then Douglas Crockford. Who's the
00:32:32.809 --> 00:32:36.609
guy in the middle? Douglas Crockford. Oh, that's
00:32:36.609 --> 00:32:40.089
Crockford. He's famous. Yeah, yeah. The curmudgeon
00:32:40.089 --> 00:32:43.710
of the JavaScript world. Yes, that's right. Yes,
00:32:43.710 --> 00:32:46.349
that's right. And Chris Wilson, of course, and
00:32:46.349 --> 00:32:49.690
he's recently retired, I believe. Yep, yep. Formerly
00:32:49.690 --> 00:32:53.569
of the IE team and then over to Chrome. Yeah.
00:32:54.029 --> 00:32:57.630
Yeah. Worked a lot on web standards, W3C. Yeah,
00:32:57.710 --> 00:33:03.589
yeah. Love Chris. Stalwart. Yeah. So when was
00:33:03.589 --> 00:33:07.569
that? So that looks like the Ajax experience.
00:33:07.869 --> 00:33:14.869
I'm guessing probably 2007, maybe. Yeah, you
00:33:14.869 --> 00:33:18.789
look like you're still just out of nappies there.
00:33:19.269 --> 00:33:22.970
Yeah, yeah. That was before I had a kid and got
00:33:22.970 --> 00:33:27.470
all the gray. Yeah, that's what happens. Bottom
00:33:27.470 --> 00:33:33.690
left. Bottom left is Jenny. Jenny. Now, it's
00:33:33.690 --> 00:33:40.789
funny because is she still the... The CIO of
00:33:40.789 --> 00:33:44.890
Microsoft? No, so Jenny recently moved up a level
00:33:44.890 --> 00:33:48.190
at Microsoft. Oh, did she? Yeah, she now runs
00:33:48.190 --> 00:33:51.549
the Trusted Technology Group. Oh, okay. So the
00:33:51.549 --> 00:33:55.549
CAO now reports to her, which is awesome. So
00:33:55.549 --> 00:33:59.049
she's overseeing a lot of different aspects of
00:33:59.049 --> 00:34:03.609
Trusted Technology at Microsoft. And Neil Barnett
00:34:03.609 --> 00:34:06.369
is our new chief accessibility officer who was
00:34:06.369 --> 00:34:09.449
one of Jenny's direct reports and recently got
00:34:09.449 --> 00:34:12.489
hired into that position. Oh, cool. Yeah, because
00:34:12.489 --> 00:34:16.429
I was talking to Dennis Ledby last week and I
00:34:16.429 --> 00:34:20.469
had the same photo. But yeah, I didn't realize
00:34:20.469 --> 00:34:23.369
that you changed jobs. But then I was talking
00:34:23.369 --> 00:34:27.909
to, I work with a guy called Dean Holden, who
00:34:27.909 --> 00:34:30.849
I used to work with at TPG as well. It's a story
00:34:30.849 --> 00:34:34.420
of my life. But he was, apparently he was saying
00:34:34.420 --> 00:34:38.920
his wife and Jenny used to, went to university
00:34:38.920 --> 00:34:41.380
together and used to share, Bristol University,
00:34:41.699 --> 00:34:44.360
she studied music or whatever. And he went to
00:34:44.360 --> 00:34:46.960
her first wedding. So there you go. I never realised
00:34:46.960 --> 00:34:52.139
things that come up. Well, I didn't realise that
00:34:52.139 --> 00:34:58.159
she was, that she'd changed jobs. Anyway, that's
00:34:58.159 --> 00:35:03.539
Jenny. Every time I think of a name, I always
00:35:03.539 --> 00:35:10.019
think Jenny McFlurry. Jenny Leigh Fluffy, which
00:35:10.019 --> 00:35:14.820
is her Twitter handle. Oh, nice. The next person.
00:35:15.320 --> 00:35:20.960
Snook? Yes. I'd forgotten who it was, and then
00:35:20.960 --> 00:35:27.760
Dennis told me. So what's the Jonathan Snook,
00:35:27.760 --> 00:35:31.329
is it? Yeah, Snook. Yeah. And what's his story?
00:35:31.389 --> 00:35:34.329
Where is he? I haven't seen. I'm not sure what
00:35:34.329 --> 00:35:36.329
Jonathan's doing now. I mean, he was probably
00:35:36.329 --> 00:35:40.610
most famous for the SMACS approach to CSS. He
00:35:40.610 --> 00:35:43.690
did that as a self -published book. Gosh, the
00:35:43.690 --> 00:35:49.670
mid -teens, I think. So I worked with Jonathan
00:35:49.670 --> 00:35:52.690
on, I don't remember if it was one or two books
00:35:52.690 --> 00:35:56.449
on JavaScript. Really came out of the JavaScript
00:35:56.449 --> 00:35:59.650
space. well -respected in the design space too.
00:36:00.190 --> 00:36:03.769
Had a really good dual brain for coding and design.
00:36:03.949 --> 00:36:06.369
Yeah, he's just somebody I remember, but I haven't
00:36:06.369 --> 00:36:11.289
seen him pop up for a while. And next to him...
00:36:11.289 --> 00:36:14.710
Next to me is Molly. Yeah, it's Molly, yeah.
00:36:15.170 --> 00:36:18.789
Yep, yep. So that was in Pasadena, California.
00:36:19.510 --> 00:36:24.650
And Molly and I were doing a talk on web standards
00:36:24.650 --> 00:36:29.230
at... Gosh, what was it? It was an event for
00:36:29.230 --> 00:36:32.869
education, for EDUs, people working in the education
00:36:32.869 --> 00:36:37.769
space. Yeah, that was 2004. Wow. That's a long
00:36:37.769 --> 00:36:41.869
time ago. Yeah, a long time ago. Miss you, Molly.
00:36:42.670 --> 00:36:47.809
Yeah, yeah. I had quite a bit to do with Molly
00:36:47.809 --> 00:36:51.750
over the years. She was a nice person. Unfortunately,
00:36:51.769 --> 00:36:56.190
I remember one TPAC I was... Before I went to
00:36:56.190 --> 00:37:01.710
bed, Molly was out in the sort of bar area drinking.
00:37:02.190 --> 00:37:04.389
I woke up at like six o 'clock in the morning
00:37:04.389 --> 00:37:06.409
because I couldn't sleep. I got up to get a coffee.
00:37:07.090 --> 00:37:10.570
Molly was still there, still partying on. She
00:37:10.570 --> 00:37:15.829
definitely lived life to the full. And lastly,
00:37:16.130 --> 00:37:21.469
on the - Kimmy Blessing. Who is this? Kimberly
00:37:21.469 --> 00:37:25.840
Blessing. Kimberly, yes, I know. Is one of these
00:37:25.840 --> 00:37:29.139
people I know, like I've, you know, sort of known
00:37:29.139 --> 00:37:34.519
her on the socials for years and years and years.
00:37:34.639 --> 00:37:37.820
And I saw her face and I thought, I couldn't
00:37:37.820 --> 00:37:39.820
remember, because she used to really like a band
00:37:39.820 --> 00:37:43.380
and follow them around all the time. Yeah, Duran
00:37:43.380 --> 00:37:46.530
Duran. yeah if i'm remembering correctly yeah
00:37:46.530 --> 00:37:48.730
it was something like that yeah yeah she had
00:37:48.730 --> 00:37:51.269
the like the first uh the first duran duran website
00:37:51.269 --> 00:37:54.210
like fan site she had created oh really and then
00:37:54.210 --> 00:37:56.730
yeah and so she if i'm remembering correctly
00:37:56.730 --> 00:37:58.949
and she like met the band a bunch of times and
00:37:58.949 --> 00:38:01.869
got to hang out with them and stuff so she when
00:38:01.869 --> 00:38:04.889
i first met her she was at paypal and she was
00:38:04.889 --> 00:38:08.050
the the person in charge of their design system
00:38:08.050 --> 00:38:12.519
this was in like 2000 2004 maybe 2005 somewhere
00:38:12.519 --> 00:38:17.579
in there um and then she eventually left there
00:38:17.579 --> 00:38:20.920
she was uh moved back she was at the time i want
00:38:20.920 --> 00:38:22.480
to say some living somewhere in the midwest i'm
00:38:22.480 --> 00:38:25.440
forgetting where midwest us and then moved back
00:38:25.440 --> 00:38:30.150
to um the philadelphia area um and was there
00:38:30.150 --> 00:38:32.809
for a while and then uh last time i talked to
00:38:32.809 --> 00:38:36.650
her she had moved to scotland i believe and taken
00:38:36.650 --> 00:38:41.750
a position at an agency in scotland yeah i haven't
00:38:41.750 --> 00:38:46.309
seen much because when i was on twitter i used
00:38:46.309 --> 00:38:50.150
to i think we followed each other but i don't
00:38:50.150 --> 00:38:52.929
know if she's moved across to the various uh
00:38:52.929 --> 00:38:56.349
other social media yeah when twitter sort of
00:38:56.349 --> 00:38:58.769
fell apart it became a became interesting to
00:38:58.769 --> 00:39:00.849
try and follow people a little bit more. Yeah.
00:39:01.150 --> 00:39:03.570
Yeah. Well, that's what I did is when I look
00:39:03.570 --> 00:39:07.769
for people that are already new. Um, yeah, but
00:39:07.769 --> 00:39:10.550
it was, it was, it was hard. Like I had like
00:39:10.550 --> 00:39:14.130
30 ,000 followers or something. So, or some huge
00:39:14.130 --> 00:39:16.809
amount. So it was a bit hard to give them away,
00:39:16.929 --> 00:39:18.829
but it just came to the point where I had to
00:39:18.829 --> 00:39:23.050
say, fuck this enough. Yeah. That was a couple
00:39:23.050 --> 00:39:27.380
of years ago. What social media do you, use prefer
00:39:27.380 --> 00:39:32.460
now um i kind of bounce between um between blue
00:39:32.460 --> 00:39:37.860
sky and uh and mastodon and uh and then a bit
00:39:37.860 --> 00:39:39.980
of stuff on linkedin as well i try to cross post
00:39:39.980 --> 00:39:42.940
to a bunch of places um but that's that's yeah
00:39:42.940 --> 00:39:45.860
kind of what i end up i cross post i always cross
00:39:45.860 --> 00:39:50.940
post as a dentist library i said i you know i
00:39:50.940 --> 00:39:54.400
i use linkedin more than i used to but i just
00:39:54.400 --> 00:39:58.989
yes same old crap that i posted for also yeah
00:39:58.989 --> 00:40:01.369
i still have my twitter account but that's mainly
00:40:01.369 --> 00:40:05.090
because i have um i have a bunch of direct messages
00:40:05.090 --> 00:40:07.170
from molly in there and i don't want to lose
00:40:07.170 --> 00:40:10.190
them yeah i'm not not willing to not willing
00:40:10.190 --> 00:40:13.230
to give that up so yeah no no that's perfectly
00:40:13.230 --> 00:40:16.869
understandable well that is the uh rose gallery
00:40:16.869 --> 00:40:20.889
it wasn't too uh yeah i didn't do too bad no
00:40:20.889 --> 00:40:24.590
so all lots of lots of Lots of faces I recognize.
00:40:24.909 --> 00:40:27.949
Yeah, well, that's why I didn't want to put you
00:40:27.949 --> 00:40:34.429
off. So, okay, so what are you working on right
00:40:34.429 --> 00:40:36.949
at the moment then, apart from what you already
00:40:36.949 --> 00:40:43.130
told us? Yeah. What's the future, the near future
00:40:43.130 --> 00:40:47.909
hold for you, Aaron? So it's going to be a lot
00:40:47.909 --> 00:40:50.650
of continuing to work in the shift left accessibility
00:40:50.650 --> 00:40:53.190
space, seeing how we can improve AI systems.
00:40:54.630 --> 00:40:58.329
Like I said, we've been doing a bunch of benchmarking
00:40:58.329 --> 00:41:00.230
work. So I mentioned that I had done some initial
00:41:00.230 --> 00:41:03.769
analysis of like what were, I spoke specifically
00:41:03.769 --> 00:41:05.809
looked at forms, but like what are coding models
00:41:05.809 --> 00:41:08.150
good at for HTML forms? What do they suck at?
00:41:08.389 --> 00:41:11.380
Basically, what are the knowledge gaps? because
00:41:11.380 --> 00:41:15.320
I kind of know what they are for real human developers,
00:41:15.440 --> 00:41:17.360
but I wanted to find out what they were from
00:41:17.360 --> 00:41:20.679
a model standpoint as well. That dovetailed into
00:41:20.679 --> 00:41:23.099
some work with the University of Washington,
00:41:23.199 --> 00:41:26.800
and we just did a... a research paper that will
00:41:26.800 --> 00:41:30.639
be at the Assets Conference in October that is
00:41:30.639 --> 00:41:34.019
basically analyzing a bunch of different models
00:41:34.019 --> 00:41:39.159
across several benchmarks related to form creation
00:41:39.159 --> 00:41:41.619
or form field creation and different types of
00:41:41.619 --> 00:41:44.780
controls and the relative amount of accessibility
00:41:44.780 --> 00:41:48.579
and how different approaches can sort of boost
00:41:48.579 --> 00:41:53.159
the accessibility of those models. And then sort
00:41:53.159 --> 00:41:55.340
of parallel to that, I've been working with Michael
00:41:55.340 --> 00:41:58.099
Fairchild, who is at Microsoft now, used to be
00:41:58.099 --> 00:42:04.519
at Deque. And he's been doing analysis into LLMs
00:42:04.519 --> 00:42:07.739
as well, kind of on a parallel track and looking
00:42:07.739 --> 00:42:10.679
at, you know, where do models have strengths,
00:42:10.760 --> 00:42:13.460
but also what tools can we use to improve the
00:42:13.460 --> 00:42:16.199
quality of the code that's being output? So that
00:42:16.199 --> 00:42:21.130
comes down to instruction files and skills. deterministic
00:42:21.130 --> 00:42:24.010
tests and stuff like that. So I'm expecting that
00:42:24.010 --> 00:42:26.230
I will continue kind of in that research space.
00:42:26.610 --> 00:42:30.369
And my goal, just being who I am and kind of
00:42:30.369 --> 00:42:32.329
my background is to try and open source as much
00:42:32.329 --> 00:42:34.250
of this stuff as I can. I don't want it to be
00:42:34.250 --> 00:42:36.269
something that we just hold close to the vest
00:42:36.269 --> 00:42:38.570
in Microsoft, but that we can actually help to
00:42:38.570 --> 00:42:43.110
improve the quality of CodeGen across the market,
00:42:43.269 --> 00:42:45.949
including the actual base models too. Like I
00:42:45.949 --> 00:42:48.329
would love to be able to influence the base models.
00:42:49.909 --> 00:42:54.269
My dream is to create a benchmark similar to
00:42:54.269 --> 00:42:56.030
what you had worked on for HTML5 accessibility,
00:42:56.449 --> 00:42:58.469
where you had kind of like where the browsers
00:42:58.469 --> 00:43:01.670
sit on a certain set of benchmark standards,
00:43:01.909 --> 00:43:05.530
to have that same sort of thing for the various
00:43:05.530 --> 00:43:07.349
coding models and to be able to see where they're
00:43:07.349 --> 00:43:10.570
at and how they're doing against that benchmark,
00:43:10.730 --> 00:43:13.750
but to do it in a very public, transparent, everybody
00:43:13.750 --> 00:43:16.989
collaborating on it and sort of everybody generally
00:43:16.989 --> 00:43:21.929
agreed upon. Yeah. I've always been a big proponent
00:43:21.929 --> 00:43:25.190
of sharing knowledge and, you know, and getting
00:43:25.190 --> 00:43:27.929
the data out there. And that's really important.
00:43:28.230 --> 00:43:31.829
100%. You know, you can't, well, you can lead
00:43:31.829 --> 00:43:34.190
a horse to water, blah, blah, blah. But at least,
00:43:34.190 --> 00:43:36.630
you know, if you say we've got the data there,
00:43:36.769 --> 00:43:40.489
you know, and analyze the data and provide, you
00:43:40.489 --> 00:43:45.929
know, sort of expert sort of analysis of it,
00:43:45.949 --> 00:43:48.940
then. you can't say that people aren't, you don't
00:43:48.940 --> 00:43:51.679
have the opportunity of being forked. Yeah, for
00:43:51.679 --> 00:43:54.699
sure. And so, yeah, so, you know, working on
00:43:54.699 --> 00:43:57.780
a lot of that stuff, I started a project at our
00:43:57.780 --> 00:44:02.619
company -wide hackathon last year to build a
00:44:02.619 --> 00:44:06.409
whole new suite of... of unit tests that could
00:44:06.409 --> 00:44:09.510
be dropped in and that would be an open source
00:44:09.510 --> 00:44:11.949
project. I'm hoping to get the opportunity to
00:44:11.949 --> 00:44:14.909
continue working on that. Like I said, the heuristic
00:44:14.909 --> 00:44:17.550
evaluation stuff is an area that I'm really interested
00:44:17.550 --> 00:44:20.650
in as well. You know, how can we improve and
00:44:20.650 --> 00:44:23.809
scale not just the compliance, but the actual
00:44:23.809 --> 00:44:26.789
usability for people with disabilities of interfaces?
00:44:28.469 --> 00:44:30.190
Because I think that gives us an opportunity
00:44:30.190 --> 00:44:34.099
to look at the intersectionality of disabilities.
00:44:34.280 --> 00:44:37.260
So like, for instance, I have a colleague who
00:44:37.260 --> 00:44:42.500
has a son that is both deaf and, or sorry, no,
00:44:42.599 --> 00:44:45.559
is blind and non -speaking autistic. And so he's
00:44:45.559 --> 00:44:47.679
got an interesting set of disability intersections
00:44:47.679 --> 00:44:52.300
that require communication. So he has AAC software,
00:44:52.599 --> 00:44:54.619
which helps him to communicate, but at the same
00:44:54.619 --> 00:44:56.340
time, he's also using a screen reader. And so
00:44:56.340 --> 00:44:58.300
you obviously don't want those voices to be the
00:44:58.300 --> 00:45:01.659
same, you know, those sorts of things. And then
00:45:01.659 --> 00:45:04.579
outside of the coding arena, I've been continuing
00:45:04.579 --> 00:45:08.159
to work in the communication space, doing a lot
00:45:08.159 --> 00:45:13.400
of work to get the technology that we used for
00:45:13.400 --> 00:45:14.940
the pilot that I was telling you about with Team
00:45:14.940 --> 00:45:18.079
Gleason out to the public. And so for the last
00:45:18.079 --> 00:45:19.519
two years, I've been working with Team Gleason
00:45:19.519 --> 00:45:21.340
on making an open source voice banking dashboard
00:45:21.340 --> 00:45:25.280
where people can come in and upload recordings
00:45:25.280 --> 00:45:27.780
of their voice or make recordings of their voice.
00:45:28.489 --> 00:45:30.429
keep that in something that is private to them.
00:45:30.570 --> 00:45:33.469
So right now it's hooked up to somebody's OneDrive.
00:45:33.610 --> 00:45:35.349
It just uses their Microsoft account. So Team
00:45:35.349 --> 00:45:38.190
Gleason never sees your recordings, but you can
00:45:38.190 --> 00:45:40.090
record your voice. It sort of facilitates that.
00:45:40.230 --> 00:45:41.929
And then you can look at a variety of different
00:45:41.929 --> 00:45:45.090
providers for voice models and then choose the
00:45:45.090 --> 00:45:46.929
provider you want to go with to use with the
00:45:46.929 --> 00:45:51.789
AAC software that you use. So really democratizing
00:45:51.789 --> 00:45:55.750
access to these voice technologies and then simultaneously
00:45:55.750 --> 00:45:59.829
working with various AAC vendors like your Tobii
00:45:59.829 --> 00:46:05.289
Donabox, your Smartbox, those folks to really
00:46:05.289 --> 00:46:10.789
improve the support for AI models in those AAC
00:46:10.789 --> 00:46:13.989
softwares. And then also looking, you know, what
00:46:13.989 --> 00:46:16.090
is the future of AAC? You know, what opportunities
00:46:16.090 --> 00:46:20.150
do we have? Prior to all of my focus on shift
00:46:20.150 --> 00:46:22.670
left at Microsoft, I was running our AI for Accessibility
00:46:22.670 --> 00:46:24.989
grant program, which was basically I got to spend
00:46:24.989 --> 00:46:27.489
Microsoft's money on or invest Microsoft's money
00:46:27.489 --> 00:46:30.750
in advancing technology for people with disabilities
00:46:30.750 --> 00:46:35.469
globally. And I did a couple of targeted investments
00:46:35.469 --> 00:46:38.469
into the AAC space. So there's an open source
00:46:38.469 --> 00:46:41.929
AAC software called Seaboard, the letter C, B
00:46:41.929 --> 00:46:46.309
-O -A -R -D. And it's available in over 40 languages
00:46:46.309 --> 00:46:49.510
and is a web app. It's actually a progressive
00:46:49.510 --> 00:46:51.730
web app. And so people can install it. It'll
00:46:51.730 --> 00:46:54.010
work offline. We've been working with them to
00:46:54.010 --> 00:46:56.829
improve the speed with which people can compose
00:46:56.829 --> 00:47:00.789
sentences. So adding... So most AAC software
00:47:00.789 --> 00:47:03.289
traditionally is like symbol sets and you compose
00:47:03.289 --> 00:47:07.090
sentences using those symbols, which can be very
00:47:07.090 --> 00:47:09.690
slow and tedious if you're using eye tracking
00:47:09.690 --> 00:47:11.889
to do so. So if you're somebody with ALS, for
00:47:11.889 --> 00:47:14.449
instance. So we worked with them to incorporate
00:47:14.449 --> 00:47:16.730
predictive typing into that process to speed
00:47:16.730 --> 00:47:21.070
up the composition. But then we funded a secondary
00:47:21.070 --> 00:47:24.300
project for them, which created a... project
00:47:24.300 --> 00:47:26.820
called C fluent, which actually listens to a
00:47:26.820 --> 00:47:28.760
conversation in real time that like you and I
00:47:28.760 --> 00:47:31.179
would be having. And then if you're posing a
00:47:31.179 --> 00:47:32.780
question or something like that, it is listening
00:47:32.780 --> 00:47:35.880
to that and able to propose responses based on
00:47:35.880 --> 00:47:38.679
my, you know, my personality, my interests, those
00:47:38.679 --> 00:47:41.980
sorts of things that I can then tailor and really
00:47:41.980 --> 00:47:45.099
quickly have a larger response and stay engaged
00:47:45.099 --> 00:47:47.260
in the conversation in real time. And that, that
00:47:47.260 --> 00:47:50.099
to me is another really powerful AI opportunity,
00:47:50.360 --> 00:47:52.780
especially if it's running offline on a device
00:47:52.780 --> 00:47:55.739
to be able. to you know the traditional aac the
00:47:55.739 --> 00:47:58.280
composition is like six to nine words a minute
00:47:58.280 --> 00:48:01.019
um but to actually be able to keep up with the
00:48:01.019 --> 00:48:03.460
pace of conversation and and be a real participant
00:48:03.460 --> 00:48:07.179
is such a just total game changer for for people
00:48:07.179 --> 00:48:10.739
who use aac so and so this this is probably publicly
00:48:10.739 --> 00:48:14.119
available? Yeah, C -Fluent is a project that's
00:48:14.119 --> 00:48:16.639
been shown off at a couple of different conferences.
00:48:16.980 --> 00:48:19.539
I think they may be in beta now or they may be
00:48:19.539 --> 00:48:23.119
coming into beta soon, but they were at ATIA.
00:48:23.440 --> 00:48:29.179
Yeah, it's really amazing stuff. you know i start
00:48:29.179 --> 00:48:30.539
to think with all that stuff though you know
00:48:30.539 --> 00:48:32.679
privacy and such and that's why the onboard models
00:48:32.679 --> 00:48:34.739
is so important like being able to have that
00:48:34.739 --> 00:48:37.659
that speech to text happen on the device being
00:48:37.659 --> 00:48:39.380
able to have that composition happen on device
00:48:39.380 --> 00:48:44.500
have it be really personal is super key so going
00:48:44.500 --> 00:48:49.119
changing the subject subject slightly um how
00:48:49.119 --> 00:48:51.679
do you put because i'm interested in this uh
00:48:51.679 --> 00:48:55.380
digital sovereignty and and the way that the
00:48:55.380 --> 00:49:00.530
european countries are building their own infrastructure
00:49:00.530 --> 00:49:06.429
and stuff because they, well, they feel to an
00:49:06.429 --> 00:49:10.929
extent that the America and thus American companies
00:49:10.929 --> 00:49:14.670
are no longer as safe business partners as they
00:49:14.670 --> 00:49:17.550
used to be. Is that something that resonates
00:49:17.550 --> 00:49:21.630
within Microsoft in the States or is it just...
00:49:22.589 --> 00:49:25.829
I mean, I can't speak to the data center stuff.
00:49:26.650 --> 00:49:28.590
I do know that there are a lot of people who
00:49:28.590 --> 00:49:30.190
really take that stuff seriously and they've
00:49:30.190 --> 00:49:32.210
been working. I mean, we have data centers all
00:49:32.210 --> 00:49:34.309
over the world that have been working on the
00:49:34.309 --> 00:49:37.690
digital sovereignty stuff for a long time. So
00:49:37.690 --> 00:49:40.909
I think it is something that they're, at least
00:49:40.909 --> 00:49:42.530
within the context of Microsoft, really on top
00:49:42.530 --> 00:49:46.090
of and really thinking a lot about. Yeah, I mean,
00:49:46.110 --> 00:49:48.090
I'm just thinking from the terms because I'll
00:49:48.090 --> 00:49:50.539
speak to Robin Bergeon. a couple of weeks ago.
00:49:50.739 --> 00:49:53.599
Robin is based in Europe and he's involved in
00:49:53.599 --> 00:49:58.480
privacy and digital sovereignty and amongst many
00:49:58.480 --> 00:50:00.539
other things. He's one of these people, he's
00:50:00.539 --> 00:50:05.059
a thinker. Do you know him, Robin? No. He used
00:50:05.059 --> 00:50:10.159
to work at W3C. That's how I know him. We co
00:50:10.159 --> 00:50:12.639
-edited the HTML5 spec together, but he worked
00:50:12.639 --> 00:50:16.699
at W3C and then he went to work on... privacy
00:50:16.699 --> 00:50:20.079
stuff at the Washington Post. And yeah, he's
00:50:20.079 --> 00:50:23.019
an interesting character. But he was telling
00:50:23.019 --> 00:50:27.739
me about these two judges in the hay that got
00:50:27.739 --> 00:50:33.440
all their finances cut off because they were
00:50:33.440 --> 00:50:41.719
pursuing America for issues in Gaza. And so then
00:50:41.719 --> 00:50:44.820
they got all the... they're basically unable
00:50:44.820 --> 00:50:49.280
to do any financial transactions. So, and that's
00:50:49.280 --> 00:50:53.280
scary to think that, you know, because what I
00:50:53.280 --> 00:50:55.500
think about my banking or anything, I don't think
00:50:55.500 --> 00:50:58.579
that it all runs through America, but in fact
00:50:58.579 --> 00:51:02.699
it does. So if, you know, I don't see myself
00:51:02.699 --> 00:51:06.679
being shut down yet, but just the, you know,
00:51:06.679 --> 00:51:11.719
the thought of that, I mean, it's just, it's
00:51:11.719 --> 00:51:15.170
scary. That's all. Yeah, I think anytime you've
00:51:15.170 --> 00:51:19.449
got concentration in one particular area, it
00:51:19.449 --> 00:51:24.389
can be problematic. I mean, not to pick on any
00:51:24.389 --> 00:51:26.590
particular company, but I know similar experiences
00:51:26.590 --> 00:51:29.750
where people were running their web conference.
00:51:30.510 --> 00:51:32.150
uh ticketing and everything like that through
00:51:32.150 --> 00:51:35.550
paypal and paypal thought that they were uh you
00:51:35.550 --> 00:51:37.130
know that there might be something shady going
00:51:37.130 --> 00:51:39.650
on and so they locked their account and then
00:51:39.650 --> 00:51:41.670
all of a sudden all of the ticket money that
00:51:41.670 --> 00:51:43.289
had come in they didn't have to pay for their
00:51:43.289 --> 00:51:44.969
venue and all that other stuff and so i think
00:51:44.969 --> 00:51:47.750
anytime when you have uh centralized stuff like
00:51:47.750 --> 00:51:50.610
that it can be uh it can be problematic so yeah
00:51:50.610 --> 00:51:54.269
totally totally understand that yeah so as a
00:51:54.269 --> 00:51:58.630
small business in the uk we've been looking at
00:51:59.280 --> 00:52:02.500
um eu office online as a possibility it's not
00:52:02.500 --> 00:52:06.579
available yet but i'm just uh doing some initial
00:52:06.579 --> 00:52:11.519
yeah i mean i know i know microsoft has like
00:52:11.519 --> 00:52:14.440
we have subsidiaries that exist within the different
00:52:14.440 --> 00:52:16.699
countries that are like microsoft france or microsoft
00:52:16.699 --> 00:52:19.039
this and the other and they they are their own
00:52:19.039 --> 00:52:21.519
autonomous thing like i don't know what the nature
00:52:21.519 --> 00:52:23.619
of the relationship is and all of that but i
00:52:23.619 --> 00:52:26.199
mean you're imagining that they they own different
00:52:27.219 --> 00:52:29.920
Well, hopefully, but you say that, but it's just
00:52:29.920 --> 00:52:32.639
that, yeah, in the situation where you've got
00:52:32.639 --> 00:52:36.860
the sort of things that the government is doing
00:52:36.860 --> 00:52:40.119
or the people in power are doing in the States
00:52:40.119 --> 00:52:45.159
at the moment are just, you know, like you don't
00:52:45.159 --> 00:52:47.940
know what the expectation will be. So you don't
00:52:47.940 --> 00:52:49.679
know what's going to happen next. So you've got
00:52:49.679 --> 00:52:52.760
to be careful. I mean, I'm glad that you're there
00:52:52.760 --> 00:52:54.579
and I'm not there because I wouldn't want to
00:52:54.579 --> 00:52:57.400
be there at this time. And I'm sure a lot of
00:52:57.400 --> 00:52:59.599
people, you know, I mean, I couldn't imagine
00:52:59.599 --> 00:53:03.519
what it's like to live under that regime at the
00:53:03.519 --> 00:53:07.579
time. But I'm sure it's fine for most people.
00:53:07.639 --> 00:53:13.199
But, yeah. I mean, it's not. But, you know, it's
00:53:13.199 --> 00:53:15.980
interesting to look back, though, and see, like,
00:53:16.039 --> 00:53:22.059
I had friends who left the US for Canada when...
00:53:23.000 --> 00:53:25.539
george w bush was elected because they're like
00:53:25.539 --> 00:53:27.400
oh no things are things are going terrible and
00:53:27.400 --> 00:53:30.139
it all it almost seems quaint like not not to
00:53:30.139 --> 00:53:32.480
excuse any of the like awful war crimes and such
00:53:32.480 --> 00:53:34.940
that that he was involved in yeah yeah but still
00:53:34.940 --> 00:53:40.019
like the the levels seem like a reason a safe
00:53:40.019 --> 00:53:43.659
pair of hands compared to bloody oh anyway yeah
00:53:43.659 --> 00:53:52.460
we know um thank you aaron yeah yeah I appreciate
00:53:52.460 --> 00:53:55.559
talking to you and having the discussions. Oh,
00:53:55.659 --> 00:53:58.420
that's what I was going to say. I recently read
00:53:58.420 --> 00:54:02.119
an article by one of your colleagues from GitHub,
00:54:02.239 --> 00:54:04.820
a guy called Eric Bailey. I'm sure you know.
00:54:04.940 --> 00:54:07.639
Yeah. But he was talking about, you know, you've
00:54:07.639 --> 00:54:12.880
been doing a lot of work on AI testing and stuff
00:54:12.880 --> 00:54:16.039
like that. Unfortunately, it's not publicly available
00:54:16.039 --> 00:54:19.760
yet. But I mean, it's sort of like you read it
00:54:19.760 --> 00:54:22.019
while I read it. I thought, hmm, that sounds
00:54:22.019 --> 00:54:24.340
tasty. And then you can't get your hands on it
00:54:24.340 --> 00:54:29.460
yet. But I know that Eric is working like yourself
00:54:29.460 --> 00:54:33.000
because I think you both see the value of open
00:54:33.000 --> 00:54:36.320
source. Yeah, no, absolutely. And I mean, I would
00:54:36.320 --> 00:54:38.460
say that GitHub has been doing some stuff to
00:54:38.460 --> 00:54:42.679
get that stuff out there. So the GitHub accessibility
00:54:42.679 --> 00:54:44.559
scanner, I don't know if you've looked at that
00:54:44.559 --> 00:54:48.559
yet, but... I've heard of it. Yeah, so you can
00:54:48.559 --> 00:54:51.880
set that up. It's available in the GitHub marketplace.
00:54:51.980 --> 00:54:54.179
It's available for free and you can run it against
00:54:54.179 --> 00:54:57.659
your own website and it will do... We did a demo
00:54:57.659 --> 00:55:00.400
at Microsoft Build, which is our annual developer
00:55:00.400 --> 00:55:04.699
conference. And the demo is basically taking,
00:55:04.940 --> 00:55:08.480
let's say somebody sends you a customer feedback
00:55:08.480 --> 00:55:10.480
response that has to do with accessibility that
00:55:10.480 --> 00:55:13.420
comes in as a GitHub issue. You can basically
00:55:13.420 --> 00:55:19.440
assign that GitHub issue to the scanner and it
00:55:19.440 --> 00:55:21.699
will take that feedback that it got. Like, let's
00:55:21.699 --> 00:55:24.320
say it's somebody who says, you know, I am colorblind
00:55:24.320 --> 00:55:26.900
and I was having an issue with X, Y, and Z. It
00:55:26.900 --> 00:55:30.139
will then go through and parse out that person's
00:55:30.139 --> 00:55:34.309
feedback. scan the site or the pages that they
00:55:34.309 --> 00:55:36.909
were talking about and create a set of specific
00:55:36.909 --> 00:55:41.030
issues that are associated with that feedback
00:55:41.030 --> 00:55:43.929
request. And then you can review those issues,
00:55:43.989 --> 00:55:46.170
say like, does this issue make sense? Does this
00:55:46.170 --> 00:55:48.050
issue not make sense? So that's something where
00:55:48.050 --> 00:55:50.630
a subject matter expert would look at that and
00:55:50.630 --> 00:55:53.829
determine what is worth fixing, what's not. what's
00:55:53.829 --> 00:55:56.429
actually a bug. And then, you know, you could
00:55:56.429 --> 00:55:59.769
submit those to GitHub Copilot to fix or remediate
00:55:59.769 --> 00:56:03.869
them yourself. And so it creates this like virtuous
00:56:03.869 --> 00:56:06.730
cycle, basically, of being able to quickly find
00:56:06.730 --> 00:56:10.210
and address accessibility. So when you review
00:56:10.210 --> 00:56:14.429
it, are you only reviewing what the summary from
00:56:14.429 --> 00:56:16.789
the AI or do you actually get to see? Yeah, so
00:56:16.789 --> 00:56:20.170
what it does is it's actually got like, here's
00:56:20.170 --> 00:56:23.199
what the issue is that it... that it found here's
00:56:23.199 --> 00:56:25.960
why it matters here's like a suggested code change
00:56:25.960 --> 00:56:28.000
here's you know basically all of the details
00:56:28.000 --> 00:56:30.079
that you would need as a human to to be able
00:56:30.079 --> 00:56:32.719
to diagnose whether this is actually a problem
00:56:32.719 --> 00:56:35.880
um or whether it's it's something that is potentially
00:56:35.880 --> 00:56:39.559
like taking just right now on your repo as a
00:56:39.559 --> 00:56:43.079
like when you do commits or whatever and and
00:56:43.079 --> 00:56:45.059
it can be set up in cicd and stuff like that
00:56:45.059 --> 00:56:47.639
as well to do scanning i'll have to i'll have
00:56:47.639 --> 00:56:50.320
to look at that And it's got a plugin architecture
00:56:50.320 --> 00:56:53.019
as well so that people can add their own functionality
00:56:53.019 --> 00:56:56.139
into it. And so there's currently the first plugin
00:56:56.139 --> 00:57:02.639
is for determining reflow issues. And so it's
00:57:02.639 --> 00:57:04.840
not super accurate yet, but it's getting there.
00:57:05.019 --> 00:57:08.500
And we're looking at whether we can bring some
00:57:08.500 --> 00:57:10.920
of the heuristic evaluation stuff that we've
00:57:10.920 --> 00:57:14.000
been working on internally to GitHub. So we've
00:57:14.000 --> 00:57:18.650
been working with Gary Fisher on that. Out. say
00:57:18.650 --> 00:57:21.250
to more Carrie Fisher. Yeah, Carrie's amazing.
00:57:21.530 --> 00:57:25.769
Yeah, I mean, because I think that people might
00:57:25.769 --> 00:57:30.570
think I'm really negative about AI, which I am
00:57:30.570 --> 00:57:35.250
concerned because of the consequences of its
00:57:35.250 --> 00:57:39.289
use. But I mean, mainly I'm interested in seeing
00:57:39.289 --> 00:57:41.909
some runs on the board. If I could see a usefulness
00:57:41.909 --> 00:57:46.409
for it, then, you know. Exactly. Yeah. I consider
00:57:46.409 --> 00:57:49.469
myself a skeptic still. I approach all of it.
00:57:49.510 --> 00:57:51.929
I use AI a lot, but I'm also super skeptical
00:57:51.929 --> 00:57:54.030
about what I get back from it. And definitely
00:57:54.030 --> 00:57:57.710
I'm not rose -colored glasses. I mean, one of
00:57:57.710 --> 00:58:02.610
the things that I really dislike in myself and
00:58:02.610 --> 00:58:10.489
in the whole process of using it is the anthropomorphization
00:58:10.489 --> 00:58:14.679
of it. Like, I don't see it as intelligent. I
00:58:14.679 --> 00:58:16.920
don't see it as a human being. I don't see it
00:58:16.920 --> 00:58:20.119
as an entity. I see it as another fucking tool
00:58:20.119 --> 00:58:23.780
that we're using. Exactly. It's like a fuzzy
00:58:23.780 --> 00:58:26.579
logic indicator to, I mean, in its broadest sense.
00:58:26.699 --> 00:58:30.239
But, I mean, obviously, because it's like, well,
00:58:30.260 --> 00:58:33.000
it does things that you would think, wow, and
00:58:33.000 --> 00:58:35.400
therefore it's magic, and therefore magic means
00:58:35.400 --> 00:58:38.840
intelligence, and it don't. No. Yeah. There's
00:58:38.840 --> 00:58:41.510
a, somebody has created. I haven't looked into
00:58:41.510 --> 00:58:43.500
it yet, but I heard about it. a couple of days
00:58:43.500 --> 00:58:46.820
ago, a caveman mode for Claude code or something
00:58:46.820 --> 00:58:50.159
like that, where it basically does none of the
00:58:50.159 --> 00:58:53.940
niceties and, and like blowing smoke up your
00:58:53.940 --> 00:58:55.880
ass and stuff like that. A lot of the models
00:58:55.880 --> 00:58:59.000
do. And it just like the, the instructions are
00:58:59.000 --> 00:59:00.840
basically like, just give me the response. Like,
00:59:00.880 --> 00:59:02.820
don't show me the in -between stuff. Don't bother
00:59:02.820 --> 00:59:05.099
with all of the pleasantries. Like just focus
00:59:05.099 --> 00:59:07.380
on what the response is. Tell me what it is that
00:59:07.380 --> 00:59:09.500
you want to tell me. And, you know, as straightforward
00:59:09.500 --> 00:59:13.179
a way possible, blah, blah, blah. I think that's
00:59:13.179 --> 00:59:16.139
an interesting idea. And the other thing is like,
00:59:16.239 --> 00:59:19.159
so I think there's some interesting stuff about
00:59:19.159 --> 00:59:22.119
being able to read through chain of thought in
00:59:22.119 --> 00:59:25.219
AI. And so I'll link you to another post that
00:59:25.219 --> 00:59:28.679
I put together that was about thinking about
00:59:28.679 --> 00:59:31.179
how do we make AI systems more efficient? Because
00:59:31.179 --> 00:59:35.900
I was having AI work on a particular task and
00:59:35.900 --> 00:59:38.199
I found that it was just spending a bunch of
00:59:38.199 --> 00:59:40.099
time doing work that it shouldn't have been doing.
00:59:41.739 --> 00:59:44.119
and it really helped me to think more about you
00:59:44.119 --> 00:59:46.780
know making sure that documentation is clear
00:59:46.780 --> 00:59:48.860
that there is not contradictions which is of
00:59:48.860 --> 00:59:52.849
course great for humans as well but the idea
00:59:52.849 --> 00:59:57.050
of building tools that then the agentic systems
00:59:57.050 --> 01:00:00.210
can use to get more deterministic results so
01:00:00.210 --> 01:00:02.730
that they're not, in my particular case, I found
01:00:02.730 --> 01:00:05.190
that it was generating a markdown file. That's
01:00:05.190 --> 01:00:07.110
all it was supposed to do, take certain information
01:00:07.110 --> 01:00:09.389
and generate a markdown file from it. And it
01:00:09.389 --> 01:00:11.409
was actually building the site to validate the
01:00:11.409 --> 01:00:14.449
markdown file. And it was like such a waste of
01:00:14.449 --> 01:00:18.650
time and energy. But reading through that history
01:00:18.650 --> 01:00:21.170
and being able to see the chain of thought for
01:00:21.170 --> 01:00:24.519
it, that enabled me to see what the issue was
01:00:24.519 --> 01:00:27.380
in the process. And then I was able to take steps
01:00:27.380 --> 01:00:29.280
to address that, to take something that was like
01:00:29.280 --> 01:00:31.960
a 20 -minute process and bring it down to like
01:00:31.960 --> 01:00:35.900
a two -minute process was pretty amazing. And
01:00:35.900 --> 01:00:38.519
so, you know, I think there's some real benefit
01:00:38.519 --> 01:00:41.539
in being able to see that. But at the same time,
01:00:41.559 --> 01:00:43.400
another researcher that I know that works in
01:00:43.400 --> 01:00:45.460
the AI space was like, yes, but we've also found
01:00:45.460 --> 01:00:48.039
that they're lying in that. So it's not always
01:00:48.039 --> 01:00:50.739
truthful. So, you know, your mileage may vary.
01:00:51.840 --> 01:00:53.900
but you know, everything with a grain of salt.
01:00:53.980 --> 01:00:56.619
And to me, it's all about how can we, how can
01:00:56.619 --> 01:00:59.079
we learn stuff, you know, sort of like, like
01:00:59.079 --> 01:01:00.719
I was talking about with the top tasks thing,
01:01:00.739 --> 01:01:04.800
how can we. use AI to try and forge ahead, use
01:01:04.800 --> 01:01:07.280
these coding models to try and forge ahead or
01:01:07.280 --> 01:01:09.099
use computer use models to forge ahead. But then
01:01:09.099 --> 01:01:11.500
how can we create backstops that are deterministic,
01:01:11.619 --> 01:01:15.199
that are faster to run, that are covering all
01:01:15.199 --> 01:01:17.000
the edge cases that we need them to cover in
01:01:17.000 --> 01:01:19.400
order to basically put a floor under ourselves
01:01:19.400 --> 01:01:21.300
so that every time we're just getting better
01:01:21.300 --> 01:01:23.440
and better. And that's what really excites me
01:01:23.440 --> 01:01:25.260
about it is because I think it gives us the opportunity
01:01:25.260 --> 01:01:29.539
to explore things and then to create those more.
01:01:31.349 --> 01:01:34.449
deterministic patterns that will help us to not
01:01:34.449 --> 01:01:36.449
use all the energy that we were using on the
01:01:36.449 --> 01:01:40.909
AI piece and really be able to improve the quality
01:01:40.909 --> 01:01:43.550
of code all across the board. That's what I,
01:01:43.630 --> 01:01:49.550
yeah, I think. And I remember recently reading
01:01:49.550 --> 01:01:55.579
that, you know, how they're putting caps. for
01:01:55.579 --> 01:01:58.480
people on on internally and they were finding
01:01:58.480 --> 01:02:02.480
that one of the largest uses of of the lms by
01:02:02.480 --> 01:02:06.699
some company was to uh turn pdf files into slides
01:02:06.699 --> 01:02:10.940
yeah so you know yeah but that's the thing and
01:02:10.940 --> 01:02:16.400
again what i'll say is is that i quite like you
01:02:16.400 --> 01:02:19.599
can i can build things with ai that i can't build
01:02:19.599 --> 01:02:21.980
otherwise because i'm crap at coding but then
01:02:22.599 --> 01:02:25.159
I've got the code and I take it away and I play
01:02:25.159 --> 01:02:28.380
with it, you know, on my local machine and then
01:02:28.380 --> 01:02:32.679
debug it or whatever. So I think that is a reasonable
01:02:32.679 --> 01:02:36.719
method to, you know, but using the LLM, relying
01:02:36.719 --> 01:02:39.440
on them to run all the computation, it's just,
01:02:39.480 --> 01:02:43.920
it's wasteful and expensive and doesn't really
01:02:43.920 --> 01:02:45.920
get you what you want to get. I mean, I like
01:02:45.920 --> 01:02:50.699
that I can get a... a python file over i can
01:02:50.699 --> 01:02:53.699
run that locally and that is deterministic so
01:02:53.699 --> 01:02:56.199
it's not you know i mean i'm not saying it's
01:02:56.199 --> 01:02:58.440
necessarily going to be truthful all the time
01:02:58.440 --> 01:03:02.679
but at least it's some basis where i don't have
01:03:02.679 --> 01:03:04.800
you know i'm not working on shifting sands all
01:03:04.800 --> 01:03:08.159
the time and i think that's a reasonable uh way
01:03:08.159 --> 01:03:13.360
to to to look at the use of ai absolutely yeah
01:03:13.360 --> 01:03:17.659
i've convinced myself i'm off But, yeah, thank
01:03:17.659 --> 01:03:23.280
you again, Aaron, for getting up. What time is
01:03:23.280 --> 01:03:26.320
it there? Is it early? Oh, it's just past nine
01:03:26.320 --> 01:03:28.900
for me. So it's not too bad. No, it's not too
01:03:28.900 --> 01:03:33.340
bad. Hopefully I'll get to see you in person
01:03:33.340 --> 01:03:36.679
again at some point. That would be lovely. I'm
01:03:36.679 --> 01:03:43.780
going to XSVTO in October and TPAC in Ireland.
01:03:45.940 --> 01:03:52.019
End of October. I'll be out and about. All right.
01:03:52.179 --> 01:03:54.699
I hope to run into you at some point. Lovely
01:03:54.699 --> 01:03:59.860
to see you again. Thank you. Bye. Bye.