#16 | The blinking cursor problem, A-Corps, and sculptural flower arrangements

What our frustration with empty text boxes reveals about the design gap in AI

I always spell color with a u. Colour, not color. You’d never know, because an invisible army of machine learning tools correct my writing across devices. Adjusting my British spelling to American standards. 

These little helpers are everywhere. When you set out on a journey with your Maps app, there’s a little robot somewhere in the code ready to change the route based on real-time traffic.

If you take photos on your smartphone one of these helpers is adding tags to every image. The date, the time, what it can see in the photo.

We don’t ask for these actions, they’re examples of AI that operate unprompted. I’m old enough to remember using the F7 key to run a spellcheck on my documents. It was a process that felt like it required my attention, nothing like the background tidying of spelling and grammar that exists today.

This background tooling that we’re all so happy to ignore raises some interesting UX questions for the future of AI interactions. Because I’m pretty sure nobody wants to always have to type their request into a small, empty box full of endless possibilities.

Filling the void

People who design things for a living deal with empty spaces all the time. A blank piece of paper, a fresh Figma file, an empty room, or a plot of land. They all make demands of us. They ask us to pour part of ourselves into a process that fills the empty space with something useful and beautiful. 

This act of creation is tough. You have to know what you want, what you really, really want (it was tough not to link to the Spice Girls there).

I think there’s a strong argument to say that the entire industry of User Experience design is built on the fundamental concept that people do not know, intuitively, what they want to do with technology. They need to be guided, prompted, reassured, and kept safe. I don’t mean this to sound patronizing. That’s where the value is—for me as much as anyone else. I don’t want to write Wireframe by starting in a command line. I want to be guided by the invisible hand of someone who understands my needs and desires, who’s built an exquisite game for me to play, that leads me to my intended goal. UX can make doing things easy, sure, but it’s at its best when it makes doing things enjoyable, repeatable, emotional. These aren’t creative indulgences—they’re the reason design is such a formidable economic force.

With that in mind. Consider the launch of ChatGPT5 last week and the brutality of the blinking cursor. 

I know for some people it’s rationalized as a limitless start-point for human creativity—but I bet plenty of people feel, if only for a fraction of a second, either (a) stuck—not knowing what to write, or (b) disappointed that they need to type the whole thing out.

Not automation

There’s a way to interpret the AI helpers I mentioned above as automated. My British spelling of colour is automatically changed to color when AI spots it. But it’s not the right way to think about a future ideal for AI interaction design.

If you take automation and expand it into more sophisticated scenarios, it gets complicated. It’s now technically feasible to make the automation work in some pretty complex problems. The average dumpster-fire inbox is a good example, but that type of automation is unlikely to feel good. 

Consider for a minute, I design a new AI hammer. My AI hammer has safety features that mean it will never hammer anything but a nail. All the nails in your life will disappear, hammered into material to bind whatever it’s intended to bind. 

If you’re anything like me, you find this notion terrifying. Not because automated hammers are in and of themselves frightening (although, I suppose they are), but the nature of a nail is such that once it has been hammered, it’s very hard to un-hammer it. 

The same is true of email. It’s not the sending of the email that’s the hard bit. It’s the consequences of sending the email. I suspect ChatGPT could run my inbox better than me. I bet it could adopt my style and tone, I’m sure it could even improve my emails, but every email it sends has real-world consequences that are as hard to un-hammer as a nail. 

Unfortunately, it’s not as simple as asking every time you want a nail hammered or an email sent. Imagine how quickly you’d get bored of that interaction. 

Collaboration

Perhaps the space we don’t talk about enough is how humans collaborate with each other. The patterns, exchanges, interactions and underpinning assumptions that power human collaboration are incredibly complex. But they’re not beyond analysis and application to AI. 

To highlight the absurdity, imagine you didn’t know ChatGPT was an AI. Imagine someone had told you at work one day that you had a new colleague. They’re smart, earnest, enthusiastic. Keen to help. Good at data processing and always willing to complete complex or menial tasks. The catch is that you can’t meet with them. They only communicate through chat. 

That would be frustrating. You’d want to meet with them. You’d want to invite them along to meetings you were attending. You’d want them to watch a big presentation you were delivering so they had additional context for an important project. 

The odd thing is, all of this is possible with AI right now. We just haven’t figured out the interaction patterns. 

Take meetings. We invite AI along to do a very specific thing. We say, transcribe the whole meeting, summarize it into notes for me, then share it with all the attendees. Very useful.

What if we said to the AI, can you come to this meeting? It’s important. Not to take notes, not to summarize actions, just to be present and witness what’s going on. Just absorb the context that will help our collaboration next week. Imagine how much more powerful that tool would be with the context of the meetings you’ve attended. 

There are plenty of people working on products that will solve this problem. They’ll make AI part of an ambient computing movement, where the intelligence recesses into the background and observes everything. Humane’s AI pin is a good (and terrible) early example. But in making it omnipresent, are we running past an obvious opportunity to treat AI like a direct collaborator? Inviting it to meetings. Asking it to join presentations. Having 1:1s with it?

The design gap

I started this article by referring to my little AI helpers as tools. They correct my spelling, find me a faster route while I’m driving, keep my photos organized—but that’s the wrong mental model.

I think it’s better to think of them as collaborators. They’re independent of me. They’re doing their thing, and when it makes sense, they overlap with my thing. I never prompt them, instruct them, or even ask. That might be the critical test we should be applying as designers.

Plenty of people walked away from the GPT-5 launch feeling underwhelmed.

I walked away with one clear thought: the relentless excitement driven by model improvements is losing momentum. The next phase of growth for all of the big AI players is going to come from better product thinking.

We need smarter feature design. Innovative interaction patterns. Proactive, rather than reactive AI. We need AI that fits seamlessly into our lives, AI that knows when to step in and when to stay quiet.

We don’t need AI with bigger brains, we need AI with better manners.

Yancey Strickler, who you may know as a Kickstarter founder, has an incredibly interesting idea for a new type of corporation. The Artist’s Corporation, or A-Corp would “allow artists to take full advantage of all the benefits of capitalism”.

If you haven’t seen the sculptural flower arrangements of September Studio on Instagram, you really should.

This wonderful interview with Katie Dill touches on so many important parts of design leadership—her vulnerability is so inspiring and I love her articulation of how beauty has an inherent value for design [shoutout to Lenny’s Newsletter for recording such an amazing interview].

Google AI Storybooks is cute. I think many people were already doing this, but Google put a UX wrapper around it. Very interesting tool if you’re building customer journeys.