Jon explores a challenge many of us are running into with AI. Language has become incredibly cheap to produce, but fluent words are no longer reliable evidence that real thinking happened.
His main argument is that language is an expression of thinking, not thinking itself. AI can generate polished documents, prototypes, code, and presentations that look like careful reasoning, but polish doesn’t guarantee anyone wrestled with the tradeoffs, formed a point of view, or made meaningful decisions
In his article, Jon describes how this changes the way we work. Leaders can mistake polished writing for strong judgment. Designers can also mistake completed prototypes for completed thinking. Teams can end up debating positions that no one actually formed because AI filled in the gaps. The thinking doesn’t disappear. It just gets pushed downstream to the next person reviewing the work.
Let’s jump into the discussion
The big idea to me is that AI hasn’t replaced thinking. It has separated thinking from production, which means our job in product and design is no longer just creating artifacts. It’s making sure those artifacts actually express judgment, reasoning, and intent.
Here’s the question I’d like to open up:
As AI makes it easier to produce polished work, how do you tell the difference between good output and good thinking?
Jon is a featured Helio author, and I’m looking forward to hearing everyone’s perspective as we dig into Jon’s ideas together.
Great to chat with you today, Jon. This is a great topic to dive into as we search for the meaning in our work. This stuck out
“Then, very recently, words got cheap. Words can now come from a system with no understanding, no message, no thinking behind them.”
I think this statement puts a spotlight on the divide I see right now with product and design people- you either believe this is horrible, and you continue to run into a feed of AI-generated content that makes you feel discouraged, or you see a new path forward to capitalize on all this junk.
I’m curious where you are in this journey, and how did you reconcile this thinking in the article!
It’s easy to fall into despair. I’ve been there and continue to fight the negative view, so I’m choosing to look for positive thinking to capitalize on this.
The first step for me has been understanding how these systems work. That’s helped me realize where their limits are and when it’s best to use them, and how to engage with someone else’s AI creation.
This resonates with me. I just started using tools to figure out what worked best and how they worked. I’ve always just looked at AI as a tool.
Ok, let’s jump into the human side- you always have fun imagery and sketches, what got you motivated to sketch your ideas? Sketching ties back into your article.
First off, thanks for noticing the hand drawn sketches in every post.
There’s a few reasons I firmly believe in sketching:
I heard this argument once: “A pencil and paper are by far the cheapest mode of idea expression and critique.” I still agree with it. If I ask a designer to go explore 3 more approaches, they can do that in a few minutes or hours on a whiteboard or piece of paper, without getting bogged in details like border radius, type face, etc. It helps focus on the core idea, rather than the polish of the idea.
Several incredibly respected design leaders Nigel Cross, Christopher Alexander, and Bill Buxton all highlight the importance of it as a cognitive function. Expression of ideas with the ability for those expressions to “talk back”. Sketching is the fastest way to get something to “talk back”.
Personal choice. I want to “practice what I preach”. So, with every single article, I explore how to visually express ideas in sketches and metaphors to help people better understand the content.
I don’t think of myself as “a good drawer”, but it’s been fun to explore the struggle of visually expressing different concepts.
Totally agree. Sketching is a way to think. Love it.
I taught design conceptualization at Stanford for 10 years, before it was in the D School. Product design was an offshoot of Mechanical Engineering and I was helping all kinds of people express their ideas.
Sketching can lead you to deeper ideas and help to articulate complex ideas. Love your progression across these two. These are fairly complex ideas, but the two work together.
How do you iterate on these? I’ve found it fun to ask Claude or ChatGPT what they think my sketches mean. It can be very helpful to see potential gaps.
I’ve tried a variety of things, nothing has really “stuck yet” but I’ll share a few of my experiments.
Sometimes an article idea starts with a metaphor I invent on the fly in the flow of my day-to-day work as I’m explaining something. So, I jot the idea down, then come back to sketch the visualization and see if it has merit. I will typically try 2-3 different compositions to see what helps emphasize my point and what doesn’t.
I use Claude as part of my writing tool stack. One of the skills I’ve built is a “visualization idea generator”. So, after I have a solid draft of an article, I will have the Claude skill provide a list of ideas. I haven’t been overly impressed with the results. I try hard to connect my design ideas to very distant metaphors (like trees, weeds, materials, constructions, riding bikes, etc). I believe it helps connect closer to real life with people. most of the time, any LLM I use has less than novel-ideas.
I wanted to see if I could use AI to create my sketches for me. I took all of my drawings from over 50 articles and worked on creating a style that Gemini with Nano Banana can replicate. It was underwhelming, but not entirely useless. I’ve used a few shapes that it’s created and redrew them with my own flair. Ultimately, I’ve committed to hand drawing my sketches (using 11" Apple iPad Pro, Apple Pencil 2, and Apple Freeform) for authenticity’s sake.
I haven’t tried asking for an evaluation from any of the vision models…but that’s a great idea. Especially since I have such terrible handwriting
I tend to find it’s not great at this. ChatGPT has better image rendering, but I use it as a thought partner to see where ideas are vague. I create all mine as a rough idea in my head, then create them in illustrator. Sometimes I start with the visual, other times I combine the thinking mid post.
Let’s switch over to this new mindset. I’ve thought about this as well, you hightlight this in your article:
So now, we can no longer rely on what was produced as evidence of thinking. There’s a new burden imposed when interacting with any content or artifact. The cognitive burden to ensure things make sense increases. In some cases, the work doubles. The producer has to think harder about what was produced, because so much was created so fast.
Many people think AI saves them time. In many low level cases, that is true. But it also shifts the work. In some cases it takes longer if you are interested in higher quality. For many product and designer peeps, this shift to editor mode is not natural. I’ve also found both parties have to want to play the same game.
Honestly, it’s overwhelming to think about how this could play out in all the wrong ways over the next few years. But, again, I try to focus on positive futures and work hard at moving in that direction. Here’s a few big impacts that I’ve been thinking about.
Since so many people can quickly create something without knowing what makes the end artifact a good fit to solve the problem, we have a lot of responsibility…but also influence.
If you’ve ever been a designer that jumps in mid-project you know how long it can take to get up to speed. That catch up effect is made worse when everyone is creating with AI, and even worse if they are just accepting AI output as the answer (that’s called slop).
Why? Because people have been involved in the intentional decisions during the creation process. When AI creates something, it will not and cannot make decisions on the fly the way we as humans do. LLMs are always solving a mathematical problem of “probability”. Sometimes (not always) that probability helps match what’s produced to what solves the problem. Now, when a human designer comes in to evaluate what someone produced, those decisions weren’t made, and can’t be interrogated. They have to be discovered and teased out. That takes time. And, it’s not just one person making with AI now…it’s potentially dozens of partners all trying to produce the solution quickly. Herding all of that will be a lot. But, I’m thinking that we can find ways to use AI to help with that monumental amount of work.
Hidden in that last paragraph is a lot of deep thinking. And, it’s compressed into an expectation that it happens much faster.
Lastly, that deep thinking needs to be expressible. As a designer or design leader, I need to be able to communication to another human what’s good and what’s bad and why. It can’t be based on feeling or gut. It has to be concrete, external frameworks that are time-tested and widely agreed on (Elements of UX by Garrett, UX Value Honeycomb by Morville, Laws of UX, etc)
I hope you notice I avoid the word “taste”. I think taste is an unhelpful word at this moment in design history. It’s ambiguous and ill-defined.
I think what I’m most excited about with AI is how digital product designers can now get much closer to the materials, like generations of past designers.
For far too long designers were separated from code. With the emergence of advanced javascript frameworks, it was really tough for designers to break in and make something with the end materials.
LLMs make it much more feasible for a designer to “code” without becoming a React or web component expert. A designer can provide a solid feature blueprint (architecture, user flow, goal, etc) and produce a functional demo that’s very real, built with the corporate design system, and one step closer to the end production state. That’s a very different handoff than a Figma UI layout. It’s also catching and solving other hidden issues sooner since you have a more real prototype to evaluate.
It also opens the door for designers to change production code. They can identify that usability issue, open the codebase, describe then changes needed, then submit the change for review by an engineer. AWESOME!
I think the future is bright for design, and AI gives us more opportunities to enhance the work we do. Stay curious and playful. It’s a fun time to discover what’s possible.
