Ship It Conversations: Rob Koch of Slalom on Accessible Cloud Native, Sign Language Recognition with Kubeflow, Overhyped AI, Conference Accessibility, and Why Teams Should Hire Deaf Engineers

Transcript

Most accessibility work gets planned at the very

end, which is a lot like building a piece of

furniture in your garage and only then checking

whether it fits through the door. My guest today

is Deaf. He's an AWS Hero, and he's worked on

using Kubernetes and Kubeflow to train models

that recognize sign language. And he jokes that

we don't really have AGI until AI can understand

sign language. Because sign language happens

in three dimensions, and a camera still can't

reliably tell which way a hand is facing. I'm

Brian Teller from Teller's Tech, and this is

Ship It Weekly. Welcome back to Ship It Weekly,

where I filter the noise and focus on what matters

when you are the one running infrastructure and

owning reliability. Most weeks it's a quick DevOps, SRE, platform, cloud,

and security news recap. In between those, I

do conversation episodes with people building

and operating these systems. And I've got a bunch

of these recorded and lined up, so expect a lot

more conversations over the next few weeks. Today,

I'm joined by Rob Koch, Senior Principal at Slalom,

AWS Hero, and co-chair of the CNCF

Deaf and Hard of Hearing Working Group. Rob is

Deaf, so the voice you'll hear answering is his

ASL interpreter, Amelia, voicing Rob's words. If

you're watching on YouTube, you'll see both of

them on screen. We talk about cloud native as

the toolbox and platform engineering as the workbench,

which parts of AI are actually overhyped, what

conferences get wrong about accessibility, and

what it really takes to recognize sign language

on Kubernetes. All right, let's jump in. Rob,

thank you for joining me. Yeah, you're welcome.

Happy to be here. So I wanted to start out...

Is this on? No, just teasing. Just, yeah, just

happy to be invited to the podcast and talking

about cloud native and accessibility. So yeah,

just pretty sure that... Hope I have some insights

that I can share to give out to folks out there

listening today. So we'll see. Hopefully we can

benefit folks out there in the community. Awesome.

Appreciate it. Thank you. So I wanted to start

out by asking, what do most teams miss when they

talk about cloud native and platform engineering?

I would say many, many times the terminology

is vague or there's some overlap. And I think

splitting out the cloud native part and what

the platform engineering part looks like. So

thinking about things like going to CNCF and

their particular, you know, map, their landscape

map that they have, there's so many tools out

there, right? And so many projects that are being

conducted and so many computing projects and

things that are... you know people constantly

coming up with new projects and all these different

things that are going on too many to name really

and I think that like too many to name so much

so that I can't even think of one right it's

like hard of no hard to know where to start so

I think for me it is like if I go to my garage

for example and I've got my toolbox and I open

it up And I see all the different tools within

it, you know, screwdrivers and, you know, pliers

and different wire strippers or, you know, all

the different things, hammers, whatever you may

have. Well, I guess not everything is a nail,

right? So just looking through some of these

things, that's where I feel like cloud native

is my toolbox, right? have all of these different

things and resources to help me build something

that I want to build and to build something cool.

And also, you know, to bring that like back to

your question about the difference and talking

about like platform engineering, I'm trying to

use the analogy back to my garage. So maybe that's

like my workbench, setting up everything with

a nice clean slate and something that I have

all these tools behind me that I can pull in

and assemble all of those things together, the

different parts and the different pieces that

might, if you want to call it that, right, that

might make the craft of the working functional

product that I'm trying to build. I feel like

the platform engineering part is the workbench.

It gives me the level space to be able to work

on. Also having some of those things like maybe

There's a table saw in place, or maybe I put,

you know, all of my other screws and nails in

one corner, and then I have easily accessible

tools that I can bring in to my workspace and

my workbench to have that there as what I would

call the platform engineering space. That makes

sense. Is there a part that you think's overhyped

right now, in general? I'm afraid to say AI.

Because I know it's a little overhyped right

now and it's, you know, becoming dependent on

it. I've heard a lot of stories and people's

comments and people haven't touched code in months,

right? Because of AI. And so I'm hesitant to

say that, but it is, I don't think it's considered

in reality very much of a hype. I don't necessarily

think that. I think it's here, but I think overhyped,

I would say maybe something of where we're like,

Hmm. Like one-off systems, maybe? Things where,

you know, oh, AI is going to replace software

or AI is going to replace your tools or, you

know, things that you already have or it can

do, you know, something over the weekend that

can be done and out the door. I think we still

need the structure and we still need the guardrails

and we still need A lot of these components,

the working CI/CD platform, as well as still needing

things like, you know, human review and tools

to stop AI from going off the rails, right? And

so I think it's key to making AI not, you know,

not hype, but reality. I guess hype to reality,

to have a great platform engineering workbench.

To go back to the analogy is kind of like having

that level bench is important, right? And having

everything that you need to make sure that the

deployment is stable and versions, etc. So yeah,

I think that's where we're at. That makes sense.

So you've done a lot. You've done a ton of accessibility

work through the CNCF. What does accessible cloud

native actually mean in practice? It could mean

many different things to many different kinds

of people. So I think my group, the Deaf and

Hard of Hearing Working Group that we have within

CNCF is very different than some of the other

groups. And so I can talk about the needs of

my particular group that I'm involved with. Excuse

me, there might be other people talking about

their needs and within their groups. Our needs

are more focused on communication, whereas others

maybe are focused on physical access or there

may be, I don't want to call it social interaction,

perspectives and things like that. So I feel

like there's different needs depending on the

group. And of course, there's a lot of what our

wants and needs are tend to overlap with each

other within those groups. So like, for example,

let's see, I would say something like pushing

for captions for accessibility needs and, you

know, different talks and keynotes and things

like that, where once you push that out and the

captions are available, it's amazing how much

response we get, not only from the Deaf and Hard

of Hearing Working Group, but people in, you

know, the Neurodiversity Group. And now they have access to information that

is their preference to how they process that

information, right? And so auditorial, you know,

maybe they're processing the information differently,

or they prefer to read the captions so that their

brain can process how that works. And there's

so many amazingly different types of the way

people process information, right? So I think

in the spirit of deaf and hard of hearing and

our community is you know, inspiring other groups

to then set up their own. We've got eight different

working groups within the CNCF umbrella that

is part of what we call Merge Forward. And so

that has been created as an umbrella within.

So all of these different working groups are

within that umbrella of Merge Forward. And so

we've all come together and shared our experiences,

our mission. Things that we're trying to accomplish

more together with less because of that overlap.

So I feel like there's something that we can

be proud of in terms of the community. It's very,

very accepting of what we're doing within CNCF.

There are ways to provide access for the various

groups. And I think that they're already starting

to provide all of that to all of the different

groups out there as well. So that's been really,

really great. CNCF and the community, really,

they're all about community and supporting the

different needs of people. So I think the approach

here is really to hopefully kind of what can

rub off from the CNCF community could maybe get

back out to where people are working in their

day-to-day jobs. And some of those things where

people get a lot of pushback in places of their

workplace. What they can see being practiced

out there, then maybe envisioning now to get

a better understanding of what they can bring

back to their workplaces and within their communities

locally and self-identify. However, you know,

things get accessed in the workplace, sharing

stories, sharing tips and strategies, and then

bringing that back to where they are in their

workplace, hopefully making it a better situation

for all of them. So I feel like seeing is believing,

if you want to say it that way. People see something

that other people are doing and they think, oh,

okay, that is how a deaf person works with an

interpreter. Or many times, you know, I meet

many people that say, wow it's my first time

interacting with a deaf person and so yeah I

mean there's deaf people everywhere you just

might not realize it but we you know by having

these experiences and identifying working with

a deaf person maybe it's not as bad as they thought

I hope anyway I give people a hard time but you

know it's it's like those type of experiences

right and so I think yeah that's that's what

the goal is and of course You know, we're having

the many different tools within the CNCF and

that toolbox, like I spoke about, having that

and needing contributors to make that product

more accessible, right? More colors that we have

in the palette, making sure that the videos have

captions, making sure that the font size is appropriate

or can be changed from different sizes. There's

a lot of visual noise. And auditorial noise.

There's a lot of different ways where we can

contribute to help build better tools for our

toolbox. Very well said. What do conferences

and communities get wrong when they try to be

accessible? Or is there commonality? Yeah, sure.

So before we had a web page on CNCF, there's

probably... You can use your search back to, you

know, something that's buried in layers and layers.

CNCF is a big community, right? What we did expect.

And so we have what we call the conference best

practices. And so if you look that up on the

CNCF website, there is everything pretty much

laid out to what you'd want to set as setting

up for success for the deaf and hard of hearing.

but there's also other groups that will contribute

to that. Someday we'll have all of that, like

I said, with that umbrella of Merge Forward and

we'll make a checklist for that. But basically

the accessibility is what mistakes that have

been experienced in the past and at the very

end of planning and accessibility should really

be the forefront instead of an afterthought.

And so I think when we're trying to build out

these you know if you have like a new furniture

on my workbench right I am designing furniture

to be able to fit through the doorway for example

right I'm going to make something that is maybe

disassembled uh that could be brought in maybe

it's too big I can't get into the house if you

think of that like ahead of time when you're

building something or maybe like the materials

that I'm using, that I want to make sure that

those are robust and, you know, the materials

are going to last and be useful for at least,

you know, 20, 30, 40 years if I'm looking ahead

at that. So I think thinking about all of those

things in the forefront, and it's very parallel

to accessibility in a lot of ways. So setting

up a screen that It's maybe a big screen. How

will the different groups of people be viewing

that screen at a conference, right? So having

the information that they need and how they're

taking it in and how they're looking at it and

watching it. So that's one thing. The speaker,

right? If they're talking, will people be able

to understand them? No, you have to add a mic,

right? Like just talking about in generally,

if you think about adding a microphone and they've

got like the AV system and the The lines and

the voices of what's being able to be carried

across the room and so people can hear them.

And so thinking about the auditorial part, but

what about the visual part, right? So maybe you're

having things that are set up like a screen in

a room with a lot of sunlight. Now you can't

see the screen because sunlight's too bright

and it's reflecting off the screen. And so there's

a lot of things to think about in terms of accessibility.

So that's just one of those things. I think the

major impact in conferences is cost. And so the

website page that we have for CNCF with the best

practices, a lot of them are free, yes, but interpreters

aren't expensive. They cost money and they have

to make a living too, right? So that is part

of it. And, you know, we take their time for

that. And so that costs money. So there's some

situations where it's, you know, accepting of

captions only at like a smaller conference because

that can be something that can be free to implement.

And I think, you know, midsize conferences where

they have a little bit of extra funds for things

like that, having sign language interpreters.

For the keynote stage, maybe only. And then,

you know, for the major conferences where each

deaf individual will want to have their interpreter

with them all the time for different interactions.

So the majors, you know, those have higher budgets

and can contribute to those things. But in the

grand scheme of things, it's really coming down

to costs and all of that, the organization. The

accessibility is such a small, very, very minor

cost in the grand scheme of all of it when you're

talking about a big conference. So it's things

like that where we see some conferences have

the accessibility coordinator, let's say, and

that is a huge help to the major conferences,

the bigger ones that we have. They have a specific

coordinator who has experience with accessibility

and maybe they have Something like, you know,

a first year coordinator, which is fine. They'll

make mistakes and then that's okay. Learning

from it and then being able to get better and

better as they go along with the same conference.

And that can permeate, you know, out to other

conferences and those have, you know, their own

community accessibility coordinators sometimes

and stories and experiences to share that make

conferences better for everyone, really. So not

only deaf and hard of hearing again, but for

many of those different groups that I spoke about

before. Yeah, that makes sense. Thank you for

that overview. So we spoke earlier a little bit

about AI. Everyone wants AI-powered observability.

What's useful? What's garbage? What's risky?

So with that, going way, way back in the days

of the implementation of AI, joking aside, right?

The vibe coding and things like that, a lot of

that stuff was pretty terrible in ways to come

up with a product or maintain a system with that

particular approach. Finding those things out

and AI integration getting more and more mature,

we're noticing that a lot of the vendors out

there are starting to add tools to it and plugins.

and plugins for the tools and skills and all

of those types of things that are propagating

now as far as that realizing, remembering when

we were starting with AI and having all the hallucinations

that we were having oftentimes, which was quite

frustrating. And, you know, we were making fun

of AI at that point because of the process and

becoming a little bit more skeptical of AI. My,

how, we have, how times have changed, right?

AI is, it's a great tool to have in your workbench.

We go back to that, right? So we're having all

of the AI writing the YAML files, whether you

love it or hate it. YAML being that. You know,

writing all of the Java writing thinking of

other, you know, just the other stuff. But having

all of that scripting and having it do all of

that is pretty amazing. It's making our work

faster, more efficient, and we can deploy more.

We can do PRs. Like nowadays, we're at the point

where we can type something out on Slack, for

example, right? And we type something like, please

create this PR for me. And in the backend, there's

something magical happening where if we've got

a good platform, good platform and platform engineered

system where we can actually go through all of

the code and review all of the things and come

up with the user stories and it can kick those

all off for us. It can do the coding. The code

changes, the reviews, you can do the unit testing.

It can do some of the regression testing as well

with the browser plugin and playing around with

that type of thing. Then it can go through everything,

making it look really good. And then you have

another AI agent maybe to review the code at

that point, say whether it's good or bad, and

then put it forth to be ready for a human to

approve it. And being able to go from basically

idea and concept to a finalized, ready to go

tooling approach in literally a few minutes,

it's pretty amazing. Normally that would take

a day or a half a day or, you know, all these

things. And now we can do this like very rapidly.

And I think, what was the question again? I know

I'm going on and on. I just forgot. Did I get

to the actual point of your question? What's

useful? What's garbage? What's risky when it

comes to AI observability? Observability, right.

So adding to that, then to kind of round it out.

So AI observability, yeah, I think it's always

nice to have something that's auditable, something

that we can look at tool-wise to make sure that

AI is on the right track. And it's doing what

we want it to do. It's using the logging and

things to validate and verify what we're trying

to achieve. And so it doesn't go too wild, especially

with things like new versions of, let's say,

LLMs, right? When they're coming in and a new

one comes in and a new version, something maybe

can go wrong, something maybe they don't have

the right settings. Maybe there's something in

terms of the confidence and the effort and, you

know, all the things, the temperature, all of

those type of things. When we're talking about

LLMs, they might not get it right off the bat

and you deploy something that's not great. And

then, you know, you blame the LLM, but there's

always nice to have something to audit. And then,

of course, you know, doing the AI audit as well,

reviewing that. So it's like the double audit,

right? Things along that line, I think. But yeah,

anyway, I think that's what I have to say about

it in AI observability from that lens. Sounds

good. I read online you had spoken about using

Kubernetes to enable real-time sign language

interpretation. What's the system story there?

I'm kind of curious why Kubernetes and what was

the problems that you were solving with Kubernetes

there? So yeah, Kubernetes is really cool to

work with, right? I can't avoid it. At least

I can't. I think in reality, when we're talking

about wanting the system to have the scalability

and the influx of training of the data and things

coming in, having the ability to go through all

of the training of the data and creating the

models and being able to train the data that

we have. Let's say a person is signing a word

or a letter, anything along those lines. We're

going to want to be able to put that into the

system, like a picture or a video. Similarly,

something that's like broken out into a series

of images from a video maybe, and then having

it be able to pick out the motions that are happening.

Of course, everyone signs differently. And the

size of their fingers are different and how they

gesture and how they move is different. So how

they sign regional signs, of course, are different

even within the U.S. ASL is a little different

from other sign languages and terminology and

things like that. So there's a lot of that where

like, you know, 30 minutes away from you, somebody

might sign something different meaning the same

thing. It's similar to an accent in spoken language.

Like if you have a Boston, Massachusetts accent

or different things like that, right? You might

have different regionalities as well in sign

language. So people sign differently across the

US. So like my last name, for example, a lot

of people, there's probably like four or five

different ways to pronounce my last name, even

though it's only four letters, like, come on,

but it's true, right? And so just things like

that, where we have Kubernetes helping us to

have different things that are maybe tough to

bring into the data and then having it being

labeled appropriately and that is a lot of human

initiative and an intuitive process that we have

right where sometimes things are kind of dumb

when it comes to language or sign language when

you're having like the contextual data and everything's

being pulled in from the different places different

books and things from all over the place and

things that haven't been trained necessarily

and of course having enough content within sign

language that's been recorded out there there's

not a not as much so that's a challenge a lot

of companies that are deaf run and have been

started require a lot of time invested and capital

and the question of the ROI on their, you know,

their return on their investment and things like

that. So I feel like it's a lot of, there's a

lot of money and benefit to having sign language

recognition in terms of the capability. For example,

like a deaf person signs something. It's quiet.

There's no auditory there. But if you were to

go to Home Depot and buy a tool, let's say, and

you're wondering where the tool is, like what

aisle it's in at Home Depot, that could be something,

for those of you who don't know Home Depot, that's

basically a hardware store based in the U.S.

And so if you're going into a store like that

and you need to be able to get all of that information

out there, So this is the inference of the process

and the interpretation of the data. So if we

have something like going back to the example

that you have an ASL, an American Sign Language

interpreter, maybe. So like you go to a store,

you say store go, the grammar gets changed around

and reverted. So store I go or store I go or

went like me go store, you can say it in a lot

of different ways in ASL. The LLM could figure

out like, okay, I'm going to the store, right?

It can interpret that part just fine. But the

recognition of the sign language part is the

hard part. That's the tough part because the

different sides of the fingers and how people

sign, is it a dark room? Is it a bright room?

What color of people's skin or shirt or with

them? Their background, all of the visual noise,

all of those things, you have to train the data

on. And so being able to be in a position of

where their hands are on the letter D, I can

sign D like this, I can sign D like this, I can

sign, you know, this is a three-dimensional

X, Y, Z, Z, right? Like this way, it's pretty

easy, you know, 0, 1, 0, 1. It can pick up these

things from the palm in this fashion but if

you're this is my perspective right and the way

I'm seeing it many times the camera cannot figure

out which direction my hand is facing and so

it misses a lot of these essential components

and so when we talk about the future and the

past and current times those are on the z-axis

which are very important to the perspective of

what's being signed and so the compute part of

that is hard to figure out as well with the scale

and the motion and all of the things that are

part of it I always tease that AGI artificial

general intelligence you know is not there until

it got until they until it recognizes sign language

right so it's I don't believe AI will pass that

type of a test until it can understand sign language

really I mean it's especially crucial with that

3D spatial component right so going back to you

know Kubernetes and having the amount of scalability

and being able to save costs and things like

that right there's money in there's not much

money in this type of a space so to be able to

scale up scale back down to one or zero when

you need to, things like that. Whereas managed

services can be quite expensive to keep them

up and running. And so, yeah, I feel like there's

a lot involved. I could go on and on to talk

about the challenges in the industry, of course,

but also just back to the Kubernetes world, I

think that we were using Kubeflow to set this

up for sign language recognition. MLOps and the

pipeline that we were using to be able to keep

the data flowing. And the more data that we feed

it, of course, the more it's running and then

the more enriched the data can become for the

model and getting it ready for the inference,

etc. That makes sense. Is there a lesson that

you think listeners can take from that with your

specific use case that... maybe can be abstracted

out to other platform teams? I haven't thought

about how they could necessarily take that to

other platforms, but I always dream of or get

ideas of being capable to sign to a computer

or gesture. Like, hey, do you mind turning this

up or different, you know, three nodes instead

of two? Or, you know, can you imagine like signing

that? Through Kubernetes and leveraging that,

like if there's noise in an office environment,

maybe you can create your own sign language to

communicate. Just say like, I want to put this

up to level two or level 10 or, you know, something

that's not actually like sign language. ASL,

we sign 10, but like, you know, maybe for a hearing

person, we can gesture like 10 with two hands.

You know, that'd be cool if we can just say like

close window and do this or something like that

to the computer or stop, you know, or something

to have a gesture-based computing approach.

The reason why I say that with sign language

is because we are starting to recognize, the

computer's starting to detect some of the words

or the things that I'm doing or hand positions

could be a little bit better detected. And then

now that you have, you know, maybe like a gesture-

based type thing. Maybe I had a surgery or something

on my throat and I can't speak. So you, you know,

can't rely on speaking to AI anymore. Dang, I'm

out of it, right? So I used to talk to my computer

and now I can't tell it what to do. And so being

able to have the option of gesture-based communication

with the computer would be cool. And also just

like noticing a lot of deaf and hard of hearing

folks as well who make videos. The captions.

they're not making it right because they don't

you know that you're signing and the automatically

generated captions while you know hearing people

like it's easy right you just speak like we're

speaking now and the speech to text is automatically

generated and it then you can just implement

it on your video but a deaf person would have

to then manually go in and put that there right

because the computer can't detect what they're

signing and so Then there's the other aspect

of maybe paying a sign language interpreter who

can voice and or check the captions and things

like that. And that's an added cost. So there's

a lot of nuances to that where I'd love to be

able to be able to sign to the computer. I know

there's one company that's doing a pretty good

job out there with doing a little bit of that

sign language recognition. I'm a fan of that

company, so they've been doing a great job with

it. But basically, it's... Things are starting

to get there and it's happening, but it's slower.

I think the speech to text, what we've had like

30 years head start. Remember, you know, back

when like there was a company called Dragon NaturallySpeaking,

where you could speak into it, right?

And we remember how bad it was back then for

the recognition. So things like that were. I

don't know, there might have been like Apple,

was it like, was it Newton or something? But

I just remember like it being really bad at recognizing

all the speech to text. And so now, I mean, 30

years in of that, we're leaps and bounds ahead.

But now, you know, as deaf and sign language

users, you know, we're getting there, but we

just have a lot of catching up to do. Yeah. Okay,

so wrapping up. If a team wants to make their

systems and their org more accessible this year,

what should they do first? Where do they start?

I wouldn't look at it necessarily systematically

or tech related. I would say the human aspect

of it is probably the first place to go. A lot

of deaf and hard of hearing people are highly

skilled. And I would say hire them and bring

them into your org and give them an opportunity.

to get through an interview to, you know, maybe

it's not a good fit. Maybe it is a good fit,

whichever, but give them an interview. There's

some challenges with bringing in interpreters

and it's, you know, deaf, deaf people have, they're

very resourceful and how to, you know, get accommodations

or access to sign language interpreters. I would

say, just ask them if you're not sure, like it

would just be like anybody who is, you know,

like BIPOC or, you know, anybody, it wouldn't

necessarily shut down the communication. It's

just like, hey, I'm not sure how to interact

with you because ours is communication based.

You know, I've never interacted with a deaf person.

How do I do that? Just ask them and let them

explain. They will help you. They will help you,

you know, and work back and forth with you. So

I think that first is hire the deaf and hard

of hearing people. Once they get into the organization,

then you'll know what the accommodations look

like for that particular individual or how to

make your tooling more accessible because they'll

bring that perspective compared to the general

public. Whereas like, you know, we use tools

and there are things that, you know, different

types of tools used in different ways to get

different information, right? So I think it's

more like, more or less like bringing in a deaf

person and then things just happen because of

their perspectives like my co-workers see a

deaf person and they say oh okay like my tool's

not accessible to you so shoot I better start

you know developing that type of accessibility

so that it can be useful for that other person

who I didn't think about it's you know kind of

contagious in terms of like making a product

more accessible when you have that exposure to

a person in your organization so I feel like

what I'm thinking of is primarily, another thing

I would say is like, you know, the WCAG, the

standard that we have for accessibility, those

guidelines, the federally implemented ones that

are, you know, by the community and whatnot,

that is important to adhere to, of course, but

the biggest impact I would say is to hire a deaf

person or somebody in those groups that can help

bring that perspective to the organization. Awesome.

Where can people find more information? Where

can they find your writing, your socials, that

sort of thing? You can find me on LinkedIn. I

have a GitHub as well. It's robcube. So you

can find me usually under that name on GitHub

if you find any information there. What else?

So LinkedIn for the most part. Yeah, I would

say. And then part of the CNCF Slack as well.

You can find me there. So there's the CNCF Slack

in general, but then you can get access to finding

my name, Rob Koch, K-O-C-H on there. And so

yeah, you can pretty much find me there. I'm

not hiding. I'm out there. Cool. I'll put links

for all of those as well in the show notes. Rob,

thank you so much for coming on. Really, really

appreciate it. Absolutely. Yeah, you bet. And

happy to be here. If there's any questions or

anything people aren't sure about, just don't

be afraid to reach out. I would just say, share,

I can share resources. There's plenty of things

I could forward on, include more people. Definitely

get, it's always nice to get more perspectives

from different people. You know, I'm one deaf

person, right? And I have one perspective and

I have my own opinions, but there's so many other.

out there, other deaf people, different perspectives

and things like that. So please, I'm happy to,

you know, share my opinion. But if you don't

like mine, I'm happy to share for you to somebody

else. Awesome. Thank you again. Really appreciate

it. Yeah, you bet. Thank you. Good chatting with

you. Yeah, appreciate it so much, Brian. That

was my conversation with Rob Koch from Slalom.

And a big thank you to Amelia for interpreting.

The biggest takeaway for me is that accessibility

works a lot like reliability. Bolted on at the

end and it's expensive and fragile. Designed

for it from the start and most of it is cheap.

Rob's furniture analogy nailed it. You build

the pieces so it fits through the doorway. You

don't find out after it's built. And at conferences,

captions are close to free and even the parts

that cost real money, like interpreters, are

a tiny line item next to everything else. I also

liked how far accessibility work reaches past

the people it was aimed at. When the working

group pushed for captions on talks and keynotes,

some of the strongest response came from the

neurodiverse community, people who just process

information better by reading it. On AI, Rob

was honest. The agents are genuinely fast now.

Idea to reviewed PR in minutes. But that only

works on a level workbench. CI/CD, guardrails,

something auditable, and a human who approves

the change. And his answer for where to start

wasn't a tool. It was hire deaf and hard of hearing

engineers. Give them the interview. If you are

not sure how to work with someone, just ask them.

Their perspective will make your tooling more

accessible faster than any checklist. You can

find Rob on LinkedIn, GitHub, and the CNCF Slack,

plus the CNCF Conference Accessibility Best Practices

through the links in the show notes. Follow or

subscribe to Ship It Weekly wherever you listen

and find previous episodes at shipitweekly.fm.

Thanks for listening, and I'll see you later

this week.

Ship It Conversations: Rob Koch of Slalom on Accessible Cloud Native, Sign Language Recognition with Kubeflow, Overhyped AI, Conference Accessibility, and Why Teams Should Hire Deaf Engineers

Watch this episode here

Chapters

Transcript

Catch This Episode

Host Commentary

Additional Links

Show Notes

👤 Guest

More from Ship It Weekly

Cloudflare Saves 100TB of RAM, AI Drives Server Prices Up, AWS Adds a Fourth London AZ, Route 53 DNS Self-Service, AKS eBPF Routing, Go 1.27, and the Danger of Hidden Infrastructure Assumptions

Ship It Conversations: Justin Garrison of Sidero Labs on Kubernetes, Platform Engineering, AI, Golden Paths, and Knowing What to Say No To

Ship It Conversations: Ned Bellavance of Ned in the Cloud on DevOps Beyond the Buzzwords, Terraform, AI, the Future of Infrastructure as Code, and Why Fundamentals Still Matter

AI Agents Target GitHub, Kubernetes 1.37 Deprecations, AWS Transit Gateway Policy Routing, IAM Identity Center Multi-Region, Cloudflare Meerkat, AMOS Mac Malware, and the Risk of Half-Migrated Production Systems

Get the next episode in your inbox