2026-11-03 22:00:00
No plan survives contact with the battlefield and no design system survives contact with the product. Products are full of variants and combinations most design systems don't account for. So when AI agents are the ones applying design systems to production code, how does this real world battle hardening happen?
In Design systems for AI agents, I outlined the techniques we use to ensure coding agents make use of design specifications and components each time they make changes or create new features. This has the advantage of continually steering software towards a cohesive product design and brand. But no matter how robust a design system is, it can't take into account all the potential combinations of layout and functionality a software application needs to work.
Nor should it. Over the years we've intentionally scaled back the amount of content in our design systems so they can stay relevant and manageable. Without significant restraint, design system systems have a tendency to spiral out of control in both resource requirements (an ever increasing design system team) and complexity (more components, more states, more documentation).
But if design systems for AI agents are not enough, how do you close the gap on coherent, integrated products? In our recent projects, we've relied on a rapid and lightweight design feedback loop. The first pass of a feature or change is driven by the design system and agent instructions, which gets us 90% of the way there. (it's worth noting earlier AI model capabilities could only get us 50 to 60% of the way.)
We then use a process of visually showing agents where things don't lineup or work and having them correct each one until all the variants and composite layout issues are taken care of. We've tried a number of ways to provide this feedback including marking it up inside Figma with comments, annotating screenshots, and just plain old typing each change into a chat box.
These experiments led to a more a tightly integrated feature in Intent for providing visual feedback to AI agents. In the latest release, the notch on macOS devices provides a way to use your entire screen as feedback. Not only can you type or speak instructions but you can also annotate anything on your screen, take screenshots, make videos, or any combination of these to provide last mile feedback on user interface issues.
This approach streamlines bridging the gap between a high-level component and design system and its real world application in a variety of screens, edge cases, and features. The logical next step would be using all this feedback to further improve the design system. After all... intelligent systems should learn right?
Alas today's current AI models still lack the judgment needed to know what to integrate and how. If given free license today's models will err on the side of over documentation instead of the pattern simplification and cohesion needed in an effective design system. So for now, a designers judgment still matters a lot.
2026-10-26 22:00:00
Years ago I was invited to Netflix to give a talk on why we should design software for mobile and how. In the Q&A I was told that people wouldn't watch movies or TV on small screens. When I gave the same talk at eBay, I heard people would never buy expensive items without reviewing them on large displays. Obviously neither is true today. So why the disconnect?
In short, many people fixate on the limitations of new technologies instead of their long-term potential. It's happened over and over again.
The Commodore 64 had just 64 kilobytes of memory. Programs could take several minutes to start up. A complete personal computer in the 80s could set you back $4,000 of today’s money. Today's $600 MacBook Neo has none of these problems.
The early Web ran on dial-up connections. Designers were stuck with a 256 color "Web-safe" palette and hacked together page layouts using table and font tags. Today's Web runs applications, TV shows, games, and more.
The first iPhone only connected to 2G networks, had no copy and paste, only had 16 built-in apps and no way to add more. I won't belabor the point but you'd think with all these examples behind us, we'd know early technology limitations don't last. And the fixation continues with AI:
These limitations have largely disappeared in short order. And the folks that focused on what AI models couldn't do are now grappling with the larger impact of what they can do. Designers, writers, and developers all of a sudden face capabilities they actively bet against, leaving them vulnerable to job disruptions and more.
If the last forty years are any guide... technology limits are temporary. Plan accordingly.
2026-10-20 22:00:00
The Ask LukeW feature on my website has been answering people's questions about digital product design for over three years. How well it answers depends on the content available to it. So in addition to improving content retrieval recently, I've also improved content indexing: a key part of making personal AI work.
Before any of my articles, presentations, or files can be used to answer people's questions, they need to be cleaned up and turned into a format that makes semantic search possible. This happens through a content ingestion pipeline, which I had to kick off manually for every piece of new content.
Now the system automatically checks for anything new or changed each hour. Audio, video, PDFs, and images it finds get queued, processed in the background, and added to the corpus. How that happens depends on the types of content.
Audio and video files get transcribed into text spelling out what was said and who said it. The service doing this recognizes my voice so it can consistently label me in each transcript. An AI model then reads the transcript and generates a title, summary, list of topics, and date if it can find one in the text. Each transcript gets split into distinct passages so an answer can reference the exact moment in a talk where I said something, and you can jump straight to it in the embeddable video and audio players.
PDF files get read by a vision model that not only extracts all the text but also describes charts and diagrams in words. Since many of my presentation slides are visual, this process matters a lot. Each PDF page is individually saved so answers can show slides inline resulting in less walls of text.
Of course, images help with that as well, so an AI model analyzes each new image added, describes what's in it, and uses my captions as additional context if the image appears in a blog post.
That's all well and good, but I'm not writing that many articles and definitely not giving enough talks to keep content recent and relevant. Thankfully, there's Twitter (X now) and I've posted there nearly 20,000 times since 2008. Damn, I need hobbies.
Despite the volume, none of those tweets were available to Ask LukeW. Now when I post, X notifies our server which pulls in the full post or deletes it if I had a half-baked idea that I removed. Older posts were added through a one-time import though I skipped replies to other people, reposts, and anything linking to lukew.com (that content was already there).
For each post, shortened links are expanded so the text reads naturally, and quoted posts are included for context. Any photo, video, or GIF also gets described by an AI model.
So why does all this matter? Making content ingestion automatic, especially Twitter/X posts as they happen, allowed me to start the Ask LukeW experience with an answer vs. requiring a question. Which addresses one of the most challenging bits of AI products: knowing what they can do. Plus Ask LukeW answers are now better because they have more content to draw from. So that's cool.
2026-10-17 22:00:00
Several years ago I read Jeff Hawkins’s A Thousand Brains, which dug into his theories on how our brain works. In it he argued intelligence comes from thousands of cortical columns that each learn their own model of the world and then vote to reach agreement. Then AI engulfed the tech world, and along the way I kept finding myself thinking about Jeff's theory.
How Jeff described our brains and how today’s AI models are built seem to have a lot of parallels. Specifically, a common set of building blocks for intelligence. How so?
Hawkins estimates the part of the brain that handles seeing, hearing, language, and thinking contains about 150,000 cortical columns. Each of these columns effectively runs the same program on nearly identical biology. But they differ in what they're connected to (eyes, ears, fingertips). Each column learns things by combining what it senses with where it senses it.
For example, you learn the shape of a coffee cup in the dark when you touch the rim here and the handle there. Hawkins claims the same process works for everything we learn, even math. In addition, cortical columns work together by sharing their guesses until they agree on what’s there. That’s why you recognize the cup faster by grabbing it with your whole hand.
The Transformers architecture behind many of today’s AI models is also built from a repeated block. It handles text, images, audio, and video once they’re broken into tokens. Each token carries a “what” (the word “cup”) and a “where” (third token in the sentence). An attention mechanism lets each token look at the others and work out how the pieces fit. It's kind of like the brain's voting between columns. Many tokens share information to settle on one interpretation.
During training, language models improve by predicting what comes next and adjusting after wrong guesses. This seems to mirror how Hawkins says the brain learns from surprise. But while Transformers learn most of what they know during training, cortical columns keep rewiring themselves as they experience the world. AI researchers are working on models that keep learning after deployment too.
So in both our brains and AI, intelligence may come from one building block, repeated many times and shaped by experience. Of course, that's at a high level and lots of details differ. But the similarities do make me think about intelligence of both types differently.
2026-10-09 22:00:00
After seeing lots of posts over the years arguing that design has become homogenized, I decided to pull together historical and modern designs for a few different product categories: TVs, phones, cars, etc. You don't have to squint to see the pattern... but why?
Do we eventually reach a final form? Does everything just trend toward average as mediocrity seeps in? Is it driven by what's possible or cost-effective in the supply chain? Or does everybody just copy everybody else? Here's the reasons I see people argue most often.
Once a product's purpose becomes clear, design can get out of the way. A TV is for watching a picture, so the ideal TV is mostly picture. Same for smartphones, cash registers, etc.
The more mature a technology, the more it competes on price and specs. Standing out is risky and expensive so less companies do it. The ones that do tend to be smaller so they have less resources and reach, making it hard to go beyond niche markets.
Nature does it too. Crustaceans keep independently evolving into crab-like shapes: a pattern we call carcinisation. Someone suggested the consumer product version should be called "crabitalism".
Aerodynamics, thermals, and materials push cars and planes toward the same shapes. Constraints do the same. Planes are designed around airports, and cars around safety standards. Change the job (an F1 car, a fighter jet) and the design changes with it.
Software is even more susceptible. Every app is already being chat-ified (video below) even though we're only a few years out from ChatGPT's release. Hardware convergence often took decades.
On that note, AI hardware is currently a bunch of experiments: pins, glasses, earbuds, and even pendants. We'll see which of them (if any) make it into the right-hand column, but for now it's a design playground.
2026-10-05 22:00:00
In today's AI-powered software, you manage agents, they manage more agents, and those agents manage even more agents. Chat threads break with that scale. We need to know what agents are planning, what they've already done, and how best to steer them. In other words we need different ways of seeing things.
When most people think about staying on top of agent activity, they jump to management views: an inbox like those found in Slack and email; a dashboard like those found in monitoring apps. Effectively these answer the question "what are my agents doing right now?"
I took a look at inboxes, Kanban boards, task lists, dashboards, and calendars in a post last year about agent management interface patterns. But while these kinds of agent views currently get all the attention, more specific and dynamic views may actually be more useful.
That's why in Intent, our large scale agent coordination tool for developers, we've built in multiple ways to see what's going on. Intent lets you create any number of documents (we call them notes) when working on a task, so you can have the specific views you need when you need them. A few examples:
Agents can draw architecture maps, flowcharts, sequence diagrams, and state machines inside notes and chats so you know what they are planning to do, and how, before spending the tokens.
Stepped walk-throughs move you through a diagram one stage at a time, each with a short explanation and the relevant parts highlighted. This not only makes plans easier to follow but also helps you quickly understand the current state of things. Boxes that reference files or notes open them when you click so can go deeper if needed.
Agents can take before and after screenshots of their UI changes and place them side by side in a note letting you judge results without reading any code. These notes can update with each change or stick around as a historical record. Agents in Intent can also pin a screenshot to show progress at a glance.
A app or website's design system can also be viewed as a note: colors, type styles, spacing, corner rounding, animation timings, and components with their variants, all in one place. Keeping a view like this open while you work makes "which padding?" one glance away. Agents read the same notes, so their UI work lines up with your references.
Of course, lots of views aren't much help if you can't set them up in a way that's useful. Every panel in Intent can hold a stack of agents, notes, files, terminals, and browser tabs. Drag things between panels, drop one on an edge to split off a new panel, and zoom any panel to full screen if you need focus.
Yes, chat still has a place for kicking off, moving along, and redirecting work. But as agents keep multiplying, a configurable workspace of useful views goes a long way toward steerings thing toward the outcomes you want.
