What are key considerations when designing AI-driven products to ensure they remain accessible to all users?
Edition No. 25: A few brewing thoughts, two curated links, and one good read.
In my regular fashion, I’m here to promise you this newsletter will become more regular. Saying this though, I feel the same way I did back in high school where I’d do my homework but forget to hand it in because I was too busy sketching out ideas in my math notebook. 🙈
In hindsight though, I guess that behavior paid off in one way or another. Confirmation Bias is a hell of a drug…
I spent a lot of time thinking and being asked about the role that AI plays in accessibility, my thoughts on AI in general, and taking the company through a few monumental stages that are worth talking about. The next few posts will reflect that, but don’t hold me to it in when those get delivered.
Let’s dive in…
If you’re here for the first time, thanks for checking out Thoughts Brewed—a monthly newsletter sharing my learnings in startups, leadership, design, and life in general. And as with most things of mine, it’s without a filter.
One Good Read
What are key considerations when designing AI-driven products to ensure they remain accessible to all users, especially those with disabilities?
This may sound silly, but a lot of the considerations that come to mind for an AI-Driven product are the same ones I’d give you if you weren’t utilizing an AI.
There are two sides to every coin. The reality is the only difference between AI-driven and non-AI driven products is that AI-driven products ensure the end-users experience is augmented, assisted, or accelerated based on whatever they’re looking to achieve. While it certainly makes our jobs easier, it also means you have to be a lot more considerate, accounting for the nefarious ways it can silently show up in the product experience. With that said, a few considerations that come to mind are…
Everyone vs Anyone
Accessible design isn't "design for everyone". It's "design for anyone". Segmentation and target users are obviously important given you have to start somewhere. However, anyone should be able to use what you design and build. Right? Accessibility is a proxy for user experience. So if your general default as a design team is to build for the best case scenario and a very specific user while completely ignoring the disabled population, it’s safe to assume you will do so when building an AI-driven product.
A.I. in design is another way to say Adjustable Interfaces
AI makes personalization or customization easier than ever before which is an absolute shortcut to ensuring anyone can independently, easily, and delightfully use your product.
Twitter used to have a fantastic “increase contrast” mode on mobile. Aside from the fact that it really was a nice accessibility feature that people could flip on and off, it was also beautiful. And it allowed different color templates you could select from if blue wasn’t a color you could see because of colorblindness or other vision deficiencies.
Adjustable interfaces1 can mean alternating the visual interface or ensuring that individuals who can’t see have an auditory experience that’s equally as lovely. Ensuring the user can set their preferences at the onset, or easily in settings is a no-brainer — especially with the use of agents that can “follow along” a user’s journey and adjust that experience with input.
And in the age of A.I, a great question is: “do we consider ‘adjustable interfaces’ the same as ‘probabilistic interfaces’? The answer is no though. While both deal with the idea of adapting the interface to the user, adjustable interfaces rely on direct user input, while probabilistic interfaces rely on predictions and machine learning.
Note: I have a post in the pipeline coming soon on why adjustable interfaces are the much more effective solution for accessibility in the near term.
How we label is critical
Disabilities are diverse, nuanced and dynamic; they don’t fit within the formulaic structure of AI, which is programmed to find patterns and form groups. Our tools are a product of their environment. They reflect their creators’ worldview and subjective lens. For too long, the same groups of people have been overseeing faulty data systems. It’s a closed loop, where underlying biases are perpetuated and groups that were already invisible remain unseen. But as data progresses, that loop becomes a snowball. We’re dealing with machine-learning models — if they’re taught long enough that “not being X” (read: white, able-bodied, cisgendered) means not being “normal,” they will evolve by building on that foundation.
So when using these data sets that are popping up left and right there needs to be this form of “scrubbing” to ensure you train it on diverse data sets that include voices, images, and behaviors from users with disabilities. Doing so avoids bias and ensures AI systems recognize and serve diverse populations effectively. Especially as a healthcare or education focused organization.
In the end, designing AI-driven products that are accessible to anyone isn’t just about implementing the latest technologies—it’s about fundamentally shifting the way we approach user experience. Whether through adjustable interfaces that empower users to tailor their interactions or through more equitable data practices that account for all populations, including those often left behind, the goal remains the same: to ensure that anyone, regardless of ability, can engage with your product with ease, independence, and delight. AI has the potential to revolutionize accessibility, but it requires intention, vigilance, and constant iteration to ensure that the systems we build are truly inclusive.
Designing AI-driven products with accessibility in mind isn’t just about adhering to best practices; it’s about setting a standard for how we imagine and build the future. This is a moment to go beyond checking the box. AI isn’t static—it evolves with every interaction, every user, and every piece of data it ingests. It can amplify inequities or become a tool for equity. The choice lies in how you approach its design.
By prioritizing accessibility, you’re not just ensuring that more people can use your product; you’re creating a foundation for innovation that benefits everyone. Because when you build with anyone in mind, you end up building something extraordinary for everyone.
Two Links
They've heard no for so long on these missions that starting a business seems like a piece of cake because now they know that all they need to do is withstand those no's.
And no happens to be the most important word that an entrepreneur will ever hear.
[…]
…you sat there, your dignity was challenged, your self-worth was challenged, your ego was challenged. They said no to you repeatedly. And he said, ego, dignity, self-worth, all that stuff is nonsense, pride.
If that's what you care about, you will never succeed. The minute somebody says no to you is the minute the negotiation begins. And that's how I run my business.
Very rare do I come across content that I consider evergreen and want to consume over and over, but this episode of Diary of a CEO with Guy Raz (How I Built This) is officially going in my Top 10 for founders.
“No” is the most important word that an entrepreneur will ever hear. And boy does it do something to the soul once you’ve heard it for the 100th time. I’ve come to find my biggest competitor in business is who I look at in the mirror, and learning how to get out of my own way continues to be a task I take on at each stage of the company which enables me to grow with the challenges of that next phase.
Burn the Playbooks in Not Boring by Packy McCormick
There is a pattern that repeats itself over and over. Maybe it’s always been this way. I wasn’t alive always, so I don’t know. I know what it looks like today:
Someone does something for love of the game or out of pure curiosity.
They have success doing that thing.
Others see the success, deconstruct how they did it.
Others still repeat those same things, joylessly, without the original spark.
Create a cheap shadow of the original.
Nothing great was ever created from a playbook.
Playbooks are the death of creativity and joy.
Few Brewing Thoughts
→ There’s a 99% chance your startup will fail
Success in any industry is based on luck — under the pretense that you view luck as hard work + opportunity colliding. A number of ingredients must come together: quality customer experience that solves the problem, timing of a market, resources to help maintain momentum and velocity, taking (the right) risks (that can’t always be quantified), having the best people (read: founder[s]) at the helm. All with relatively perfect timing.
Your job is to simply stay alive long enough to beat the incumbents. Each stage de-risks it a little more as you go along, but even then, there’s no guarantee.
Even giants fall sometimes.
→ “Startups don’t need more processes as they scale. They need to hire people smart enough to work comfortably with ambiguity.”
I saw this from someone the other day and have been thinking about it a decent amount. While it isn't wrong, it's also not right. These two can end up not having anything to do with each other. There's a misconception that if someone is "smart enough" they’ll have the ability to sit comfortably with ambiguity.
Your goal as a founder is to find someone with the smarts plus experience. I've worked with brilliant folks that hadn't [yet] built the muscle to handle ambiguity no matter how much brains or ownership mentality they had.
Wider convo at play here about 0 → 1 people and processes versus Scale stage.
→ Educating people in a paradigm shift is incredibly important.
There’s so much inertia that has to occur. The comparison I use frequently on this topic is like moving an elephant. You can’t just assume you can push it into the existing reality; you have to give them a better model. Often that means waving peanuts to encourage movement forward and peanuts in this case means money (lost or gained). By looking at the comparison between the past method and future state, it makes it easier to show the arc and illustrate that this new way can’t carry over — it worked for it’s time but now it’s not effective, it’s too expensive, and it doesn’t actually solve your problem.
As always, thanks for reading! I appreciate you. If you found it insightful, please give it a share on the socials. If you have any questions, go ahead and AMA by dropping a comment or pinging me on Twitter. If I don’t have the answer, we’ll deep dive together and I’ll more than likely turn it into a post.
Until next time…
Cat.
I was asked if I consider “adjustable interfaces” the same as “probabilistic interfaces”. While both deal with the idea of adapting the interface to the user, adjustable interfaces rely on direct user input, while probabilistic interfaces rely on predictions and machine learning. Blog post coming next on this…