My biggest takeaways from @OpenAI's Head of Design @iansilber:
1. Designers are the unhappiest people in tech right now, and Ian has a theory for why. My tech workforce survey found that of all roles, designers are the most overwhelmed, anxious, tired—and least likely to recommend their role to others. Ian's explanation is that engineers have found incredible productivity gains from AI, sometimes 10X-ing or 100X-ing their work, while designers haven't. In part, because the design process is inherently messy—you try something, it sucks, you try again, you gather feedback, you realign a large org.
2. Nobody has the new design process figured out yet—which means nobody is behind. Ian’s message to designers feeling behind—whether they’re early-career, late-career, or have never designed anything—is that nobody has this figured out, including people inside OpenAI. Whatever seems cutting-edge today will look archaic within a year, let alone a month. In Ian’s words, “I strongly believe this is the best time in the world to be a designer.”
3. The designers who are thriving are constantly experimenting with new ways of working. The differentiator is whether AI has made someone feel more capable. Ian sees this at OpenAI: designers who approach tools with curiosity and a willingness to experiment find they can move through far more ideas, express themselves more fully, and raise the quality of their thinking. Throw an idea into an agent at the very start (to think it through, to prototype it), try new ways of working without needing them to be the final answer, and re-test constantly, because what didn’t work yesterday might work today.
4. Ian’s most counterintuitive advice for overwhelmed design teams is: do less. Instead of designing something new—extend an existing feature, build on an existing component, or question whether the thing needs to exist at all. That instinct connects to the skill he says is becoming most important: systems thinking. Rather than shipping one-off siloed experiences, OpenAI tries to build primitives that compose (the way built building blocks early on), so the product feels like one cohesive thing and the model can reason about how to pull the pieces together.
5. ChatGPT may have the widest audience spectrum in history. The same product serves someone picking tonight’s recipe and someone automating their factory in Japan. Ian’s team operates on the idea of capability overhang—the vast majority of users tap a sliver of its value—so they keep the mainline experience extremely simple, put cutting-edge features where the people who care will find them (the desktop app, Codex), and only distill something into the default once it no longer requires a mode or a switch.
6. The PM/eng/design roles are blurring, but they’re not going away. Marc Andreessen described PMs, engineers, and designers as a three-way standoff, each convinced the other roles will disappear. Ian’s take is that the lines have always been blurry (he’s seen product-manager-archetype designers and polish-archetype designers his whole career) and they’re getting blurrier, but at any real scale you still need people responsible for different things, because one person who can design, engineer, align a company, and rally teams is vanishingly rare.
7. Ian’s hot take: AI is already an incredible product designer. AI isn’t better than humans at visual design yet—but is, broadly, genuinely great. And, unlike any designer in history, it’s accessible to every single person. Humans still have the edge where there’s no training data: understanding what people need, watching them use things, and inventing new paradigms that people haven’t asked for. Multi-touch on the iPhone, Instagram feeds, Snapchat’s disappearing messages—none of that could be pattern-matched from what came before.
8. OpenAI designs at two speeds. For parts of ChatGPT, which is used by over 1 billion people, they obsess—“We try 100 things, and we throw out 99, and we finally ship one,” often with full A/B testing and user research behind it. For other things, like much of Codex, they can go from idea to ship in four hours, building in public for instant feedback.
9. The “super app” vision is one universal input that knows when to answer quickly, or chat, or just go do the work. Ian’s north star for ChatGPT’s broad consumer base: you shouldn’t have to think about how it works, which model you’re on, or what speed you’re using. Expose those controls for the people who need them; make the default great for most. The experience gets more proactive, with richer voice input, more interactive visual outputs, and durable workflows you reuse instead of starting from scratch each session. ChatGPT Work already hints at this with suggestions from calendar and Slack. Voice just took a step-change toward feeling natural. “The design job is simplifying while the capability surface explodes.”
10. Even the head of design at OpenAI spent his first stretch feeling like he was bad at the job. Ian arrived intimidated—GPT-4 had just come out and he didn’t feel he truly understood it—and stepped into a role scaling faster than anything he’d managed before. When he talked to friends who had led large design teams, they all felt exactly the same way, and an impartial mentor or buddy outside the company turned imposter syndrome into shared, solvable challenges. His broader advice for everyone in the field right now points in the same direction: be humble about how early this all is, have fun with it, and focus on the outcome—“We’re here to make great software that people love.”
★ Required reading for makers of websites: Matuzo’s HTML boilerplate in 2026.
Every element world-class front-end dev, author, a11y auditor & web standards speaker Manuel Matuzo uses for the basic structure of an HTML document … with explanations.
Updated from his classic 2021 boilerplate, ’cuz a lot has changed in HTML since then.
https://t.co/jkKOleEFc4
The CSS Working Group has just published the first public draft of the CSS Values and Units Level 5 Module. Although still in its early stages, this release introduces many new features and exciting improvements.
Here's a summary of some of the changes.
NEW: I wrote about the 200,000 American children orphaned by COVID, and why we’re not on track to prioritize the children who never get to leave the pandemic behind. @TheAtlantic 1/
https://t.co/TrVKQCWa4O
"We have very talented front-end developers, however, they are not React experts - nor should they need to be. [..] front-end should be built as standards-compliant HTML/CSS with JavaScript used to enrich functionality where necessary and appropriate."👏🏻💯 https://t.co/EMiLLXZko8
"You can't make a baby in 1 month by throwing 9 people at the task."
I heard this many years ago, and it's really stuck with me. It always blows my mind how often people try to accelerate projects by throwing more people at it. Very often it has a counterproductive effect.
Very clever CSS trickery here from Carter Li. If you declare a CSS Custom Property as an <integer>, then it can be interpolated as an integer. Meaning it can be animated!
https://t.co/vQyKCxnwsl
MDN as a website isn't going anywhere right now. The team is smaller, but the site exists and isn't going away. We will be working with partners and community members to find the right ways to move it forward given our new structure at Mozilla.
The number of people who have told me that MDN is a wiki, therefore the community will keep it up to date tells me two things. People do not get the value of professional tech writers. Folk are incredibly optimistic about what "the community" will do for free.
It’s free!
You want to take my HTML course for free?
Along with a whole series of software development courses? Here you go.
Thanks, LinkedIn and Microsoft. Thanks for recognizing people need this kind of opportunity right now.
I'm old-fashioned.
I like my CSS seperated from my HTML; my HTML seperated from my JS; my JS separated from my CSS. I like my JS layer only added when I need it, usually progressively. CSS added progressively on top of semantic markup. I don't fight the C in CSS, I embrace it.