The Interface Illusion: Why AI can't (yet) replace user control
Edition No. 26: Few brewing thoughts, two curated links, and one good read.
If you’re here for the first time, thanks for checking out Thoughts Brewed—a monthly newsletter sharing my learnings and real-time musings on startups, design, leadership, business building, and the what impacts it all. And as with most things of mine, it’s without a filter.
Before we dive in, I’m kicking off the year with a few things on deck that you may benefit from…
📚 I’m teaching a free Lightning Lesson on Maven focused on Driving measurable business impact with accessibility.
📚 I’m joining Mindaugus Petrutis (CEO of Coho) for an exclusive session on the Truth About Designer Portfolios: A Founder's Perspective. It’s for Coho customers but they’ve reserved a few seats under the “Friends of Cat Noone” ticket type at sign up!
💬 I’m officially an expert on Intro.co where you can get personalized consultation and advice from me on a variety of topics. Just an FYI: an initial 1:1 advisory is cheaper with a Founding Member plan via my newsletter — as it’s included.
One Good Read
The Interface Illusion: Why AI can't (yet) replace user control
In my last post I mentioned A.I. in design as another way to say Adjustable Interfaces. I was asked whether or not I consider adjustable interfaces the same as probabilistic interfaces. For starters, 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.
Both deal with the idea of adapting the interface to the user. However, adjustable interfaces rely on direct user input, while probabilistic interfaces rely on predictions and machine learning. They aren't just different approaches; they represent fundamentally different philosophies about user agency.
When accessibility comes into the mix, the pros and cons are even more critical to consider given how the end [disabled] user will either benefit or be negatively impacted. And given the drastic ways probabilistic interfaces can transform how we interact with technology and make it more accessible for everyone, we need to talk about sequencing to ensure we once again don’t design a [digital] society that excludes.
So how do we get from where we are now (working our way toward adjustable interfaces) to accurate predictability for what a user needs? Something I’ve been spending my time thinking about is how we ensure we capitalize on the speed at which we’re transforming these interfaces in way that ensures we collect enough accessibility data to truly ensure disabled people benefit.
To explain the differences, let’s think of it like an assistant rearranging the furniture in your home or doing a nice interior re-design…
Adjustable Interfaces: The user is in command
Adjustable interfaces are what you might call the “classic” approach. They allow users to customize their experience—choosing font sizes, color contrasts, layouts, and so on.
Want to rearrange the furniture? Go for it. Need to adjust the lighting? It's your call. These interfaces put users in the driver's seat, giving them the tools to shape their digital environment to match their needs. The structure of the house doesn’t change, but you make it work for how you live.
This approach is foundational for accessibility. It ensures that users, regardless of their abilities, have control over their interaction with software. Customization empowers people to adapt their environments to suit their unique needs and preferences.
Predictable, reliable, and simple.
Probabilistic Interfaces: The machines predict your needs
On the other hand, probabilistic interfaces rely on predictions. Powered by algorithms, they aim to understand what a user might need next and adjust dynamically.
That’s where things get tricky for accessibility.
Probabilistic systems often lean on historical data to make decisions. But for accessibility (and in turn disabled users) there simply isn’t enough historical data to make those predictions accurate. This isn’t just a data gap; it’s a window into how often disabled users have been excluded from the design process.
In this case, sometimes you get an overeager assistant that nails it and it feels like magic. But often times…they’re constantly trying to anticipate your needs and end up putting the couch in the kitchen. And anybody who has used any of the AI models for long enough (be it for blog post writing, in Cursor building an app, etc.) knows this happens more often than we'd like to admit.
For someone simply coding, it can be annoying at best — but you put up with it. Innovation! When it impacts your day-to-day living though? It becomes downright disruptive — creating new barriers and cognitive load, making the experience frustrating or even broken.
Sequencing tomorrow’s AI (accessible interfaces)
With the introduction of AI agents—think the latest introduction of Operator by OpenAI (so damn cool!)—there’s no doubt that automation is getting more powerful. Fast forward to this recent attempt where an AI agent struggled to navigate a CAPTCHA though and you realize we’ve got a bit of ways to go.
Does this mean there’s no future for probabilistic interfaces in accessibility? Not at all. Now this is relative to the speed at which we’re moving in tech these days. So I’m not saying we’re even half a decade away; I’m just saying it ain’t tomorrow. And for them to work effectively (soon), we need to get the sequence right:
First, accessibility needs to be the default. If your software isn’t accessible at its core, it’s game over. Prioritize creating interfaces that meet the needs of all users, with a focus on predictability and consistency. And no, this doesn’t mean your product won’t be beautiful or easy-to-use.
Next, let users adjust. Once you’ve built an accessible foundation, empower users to tailor their experiences with a robust adjustable interface. This step not only improves usability but also generates valuable insights about how people interact with your software in real-world scenarios. And yes, you should actually talk to them about to how they use it.
Finally, layer on probabilistic features. With a strong dataset—rooted in user-driven customization—probabilistic interfaces can step in as an enhancement, not a replacement. They’ll have the context needed to make accurate, helpful predictions without disrupting the user’s control or trust.
It's like building a house - you don't install a smart home system before you've got solid walls and working plumbing. Yet somehow, in tech, we keep trying to skip to the fancy stuff before we've laid the foundation. With an increase in expectations from users, those decisions will start to quickly compound as people turn away from products that feel like “it doesn’t know us”.
Optimizing for the bottom line
For users, disabled or not, predictability and trust are non-negotiable. Probabilistic systems, by nature, introduce a level of uncertainty that can be harmful if not handled with care. We can’t rush into flashy AI-driven interfaces without ensuring they’re grounded in reliability.
Thoughtful design is the co-pilot of innovation.
Let's be clear on something: I'm all in on probabilistic interfaces. The potential for AI to create truly adaptive, personalized experiences is incredible. Adjustable and probabilistic interfaces each have their place in the design landscape.
For disabled users, interface reliability isn't just a preference - it's critical to their ability to navigate digital spaces much like a wheelchair ramp, braille, an elevator in a physical space, etc. Before we can have AI making smart predictions about how to adapt interfaces, we need robust data about how different users actually interact with technology. That data has to come from somewhere, and right now, we're missing crucial pieces of the puzzle.
It’s not about choosing one over the other. It’s about recognizing that the journey to accessibility is iterative, and we’re finally at a place where more companies than ever are adjusting their software to have proactive and continuous accessibility baked in to ensure their customers have usable software.
By prioritizing adjustable interfaces today, we're not just solving immediate accessibility needs - we're creating the dataset that will power the next generation of adaptive technology. That’s where real innovation happens.
Two Links
How to Think About Product-Led Growth, Bootstrapping vs VC, and Early Exits with Jason Lemkin
“I could never raise venture capital until I was so profitable we didn’t need it.” — @benchestnut
This is a great video and list of topics covered by Jason that I've spent way too many hours thinking about as a founder (we're a venture backed startup, PLG motion, etc).
I particularly love when you can hear (in the silence) everybody's brain break as Jason details why VCs will never back the $35M ARR company asking the question — but PE will.
Is the Chief Design Officer dead? (No.) The Great Creative Awakening: On Business leadership in the AI era
Think about it this way: If AI is handling the execution and optimization of ideas, then the real competitive advantage lies in creating environments where breakthrough thinking becomes not just possible, but inevitable. The Creative Leader becomes the architect of what I like to call “organizational ingenuity” – designing the conditions, systems, and experiences that unlock a company’s collective creative potential at scale.
Wrapping up his time with Stark this week as our Chief Design Officer, Benedikt Lehnert shared a fantastic and timely post on the evolving role of design leadership within organizations, the influence of design (finally) becoming more integrated across various leadership roles, and where design becomes embedded into the core strategy and operations of companies — rather than being siloed within a single executive position. The integration signifies the maturation of design as a critical component of business success, moving beyond titles and into an intrinsic part of organizational culture.
Few Brewing Thoughts
→ Hiring is as much about intuition as it is about on paper qualifications
One of the most critical skills in running a company is knowing how to hire. It’s also one of the hardest to master. And lord have I made my fair share of mistakes over the years. But while books, frameworks, and consulting firms may have you believe it’s a science, the truth is that effective hiring—the kind that strengthens your company’s DNA—is as much about intuition and observation as it is about qualifications.
Here’s the thing about people who don’t know what they’re talking about: they become absent of details the more you dig in.
There’s always going to be data—it's the internet after all. We each decode our experiences in different ways. So the best way to assess how someone thinks, not just what they’ve done means multi-layered "how" questions, and a lot of reps for you as the founder, become your best tools.
→ The omission of optionality when funding your company
One of the biggest problems founders in tech are made to believe is if they didn't go to [insert fancy school], if they don't fundraise, don't live in the valley, etc. that the idea has no path. We don't hear the fantastic success stories of Mailchimp, Atlassian, etc.
You're not taught optionality (e.g. Family offices, VC, PE, Bootstrap FT, Bootstrap as a side project until it takes off, etc.) — mainly because we're all led to believe that "if you just have the numbers, VC is for you! You don't have to be making money!". There's a reason most startups fail.
→ The white collar spaces that need disrupting were handed to us on a job board like the modern day Gold Rush & Claim Maps.
I was reading Tomasz Tunguz around the “White Collar Revolution” and found myself in disbelief at the number of ideas that were just handed to founders on a platter. Whether you’re exploring with AI or looking for your next idea, the Bureau of Labor Statistics' job board has become a roadmap for founders in identifying the next wave of automation opportunities. The companies that recognize where AI can drive efficiency in extremely archaic markets will define the next era of work. Not to mention, as part of Software Engineering, it obviously includes the spectrum of sub-ideas and immersive experiences ripe to be created with AI in gaming and entertainment — and are continually flocked to in good times and bad.
It feels like a modern day Gold Rush with Claim Maps! During the 19th-century Gold Rush, so many prospectors flocked to California, but the ones who struck it rich weren’t just aimlessly digging—they followed government-issued geological surveys and claim maps that highlighted areas with high gold deposits. Today, founders (focused on utilizing and stretching AI to its limits) are in a similar position. The BLS job board is a modern-day claim map, marking where inefficiencies exist and where AI can automate white-collar work.
The opportunity isn’t hidden—it’s just waiting for someone to dig in with the right tools.
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.
Until next time…