Every deck that cuts a list cites one jam display from one grocery store. Pooled across 63 conditions and 5,036 people, the effect is 0.02. Here is what the evidence really supports, and the four things to check before you cut anything.
NASA had a full design system on paper in 1975, and it is better written than most of the ones I see in Figma. Not because of the grid. Because it said no in public, and said why.
Teams can watch every step their AI takes, but barely half test it against a real set of inputs. Deciding what counts as a good answer is a design job. Here is how to write it down.
Scroll driven animation is a few lines of CSS now, and all three engines support it. The defaults are the problem: no reduced motion guard, and motion that runs the whole time an element is on screen. Five fixes and a four question audit.
The way most teams run critique is close to the setup psychologists use to produce conformity on purpose. The fix is not better intentions. It is a different running order.
Designers quote the IKEA effect to justify long signup flows. The original study points the other way. Value appeared only when people were allowed to finish something, and that changes what the first five minutes of your product should do.
Everyone is selling prompting as the new design skill. It is one level too low. The transferable skill is describing done precisely enough to judge what comes back. Five ways designers write adjectives when they should be writing tests.
Users are not scoring your options. They are rehearsing an argument they will have to make to someone who was not in the room. Five places your comparison table arms the wrong choice, and what to ship instead.
In May I wrote that Liquid Glass changed everything. I had the shape of it wrong. iOS 27 hands users a continuous slider over the material and stops honouring the key that let apps stay out of it. Liquid Glass is not an appearance you adopt. It is a range you have to survive.
State of Prototyping asked 1,478 designers how AI is changing their work, and one finding went everywhere: researchers feel most at risk. Then I called the survey's own API and checked the row count. It is 23 people, nine of whom are the frightening number.
Every few years a team strips the labels off its toolbar and calls it clean. Vienna ran that experiment properly in 1925, and the answer has not changed: icons are a helping language, not a replacement for words.
Teams read cognitive load as screen density, so they strip the screen and the hard tasks get harder. The split that matters is between the load the task carries and the load you added.
The 2026 interop list reads like a permission slip to delete code. I checked all four features against Baseline this morning. One is ready, one has no Safari, one is still partial in Firefox, and one is safe to ship anyway for a reason worth knowing.
73% of designers say the bar went up this year. 13% of leaders changed how designers are reviewed. When screens are nearly free to produce, counting them grades the tool, not the designer.
More than 100 alarms went off at Three Mile Island and none of them were ranked. The fix already existed and had for decades. A 2026 survey says most on-call teams have rebuilt the same control room in software.
Pretty prototypes test well. A meta-analysis of 31 studies says they also work a little better, but nowhere near as well as the ratings suggest. Here is how to tell what your last test actually measured.
Your interface can now run as a guest inside Claude or ChatGPT. The host owns the font, the colour, the navigation and the network. Five things you lose, what to do about each one, and a ten minute audit of your own product.
Agents build UI from what they can read. One audit of 37 public design systems found six publishing tokens in the standard format and three shipping Code Connect. Here is what to publish, and in what order.
Beck's diagram has been copied by transit systems all over the world, and most of the copies took the half that does not matter. What he actually deleted, what it cost London, and the test for when you are allowed to get the data wrong on purpose.
People judge an experience by its sharpest moment and its ending. Design writing turned that into permission to skip the middle. Five things the research really says, and a ten minute audit of your own product.
Code layers are still behind a waitlist eleven weeks after Config. The habit they promise does not need them. Five moves you can make today, and a ten minute audit of your product file.
You can spot an agent-built screen from across the room. Six tells, what each one is really missing, and why writing a longer prompt is the wrong lever.
The average designer runs seven AI tools, up from three. The top two things slowing designers down are now learning the tools and juggling them, ahead of output quality. Cut to two, and judge a tool by whether it survived ninety days.
AI gave design a new gear, and one number in this year's 900-designer report moves the wrong way. Four times as many designers say collaboration has got worse. Five things that are actually breaking, and what to change on Monday.
A sparkle no longer counts as telling anyone anything. Five things your interface has to do now that the EU rules are enforceable, and the one exemption every team will misuse.
Most UX metrics look good on dashboards but fail to reflect real user success, so teams should focus on measuring actual outcomes, effort, and task completion instead.
Generative UI is about designing systems that dynamically build and adapt interfaces in real time based on user intent, context, and behavior instead of fixed layouts.
Users are exhausted by AI. And most products are making it worse. The trust deficit is real: high trial, steep drop-off after day 3, low 30-day retention.
Every deck that cuts a list cites one jam display from one grocery store. Pooled across 63 conditions and 5,036 people, the effect is 0.02. Here is what the evidence really supports, and the four things to check before you cut anything.
NASA had a full design system on paper in 1975, and it is better written than most of the ones I see in Figma. Not because of the grid. Because it said no in public, and said why.
Teams can watch every step their AI takes, but barely half test it against a real set of inputs. Deciding what counts as a good answer is a design job. Here is how to write it down.
Scroll driven animation is a few lines of CSS now, and all three engines support it. The defaults are the problem: no reduced motion guard, and motion that runs the whole time an element is on screen. Five fixes and a four question audit.
The way most teams run critique is close to the setup psychologists use to produce conformity on purpose. The fix is not better intentions. It is a different running order.
Designers quote the IKEA effect to justify long signup flows. The original study points the other way. Value appeared only when people were allowed to finish something, and that changes what the first five minutes of your product should do.
Everyone is selling prompting as the new design skill. It is one level too low. The transferable skill is describing done precisely enough to judge what comes back. Five ways designers write adjectives when they should be writing tests.
Users are not scoring your options. They are rehearsing an argument they will have to make to someone who was not in the room. Five places your comparison table arms the wrong choice, and what to ship instead.
In May I wrote that Liquid Glass changed everything. I had the shape of it wrong. iOS 27 hands users a continuous slider over the material and stops honouring the key that let apps stay out of it. Liquid Glass is not an appearance you adopt. It is a range you have to survive.
State of Prototyping asked 1,478 designers how AI is changing their work, and one finding went everywhere: researchers feel most at risk. Then I called the survey's own API and checked the row count. It is 23 people, nine of whom are the frightening number.
Every few years a team strips the labels off its toolbar and calls it clean. Vienna ran that experiment properly in 1925, and the answer has not changed: icons are a helping language, not a replacement for words.
Teams read cognitive load as screen density, so they strip the screen and the hard tasks get harder. The split that matters is between the load the task carries and the load you added.
The 2026 interop list reads like a permission slip to delete code. I checked all four features against Baseline this morning. One is ready, one has no Safari, one is still partial in Firefox, and one is safe to ship anyway for a reason worth knowing.
73% of designers say the bar went up this year. 13% of leaders changed how designers are reviewed. When screens are nearly free to produce, counting them grades the tool, not the designer.
More than 100 alarms went off at Three Mile Island and none of them were ranked. The fix already existed and had for decades. A 2026 survey says most on-call teams have rebuilt the same control room in software.
Pretty prototypes test well. A meta-analysis of 31 studies says they also work a little better, but nowhere near as well as the ratings suggest. Here is how to tell what your last test actually measured.
Your interface can now run as a guest inside Claude or ChatGPT. The host owns the font, the colour, the navigation and the network. Five things you lose, what to do about each one, and a ten minute audit of your own product.
Agents build UI from what they can read. One audit of 37 public design systems found six publishing tokens in the standard format and three shipping Code Connect. Here is what to publish, and in what order.
Beck's diagram has been copied by transit systems all over the world, and most of the copies took the half that does not matter. What he actually deleted, what it cost London, and the test for when you are allowed to get the data wrong on purpose.
People judge an experience by its sharpest moment and its ending. Design writing turned that into permission to skip the middle. Five things the research really says, and a ten minute audit of your own product.
Code layers are still behind a waitlist eleven weeks after Config. The habit they promise does not need them. Five moves you can make today, and a ten minute audit of your product file.
You can spot an agent-built screen from across the room. Six tells, what each one is really missing, and why writing a longer prompt is the wrong lever.
The average designer runs seven AI tools, up from three. The top two things slowing designers down are now learning the tools and juggling them, ahead of output quality. Cut to two, and judge a tool by whether it survived ninety days.
AI gave design a new gear, and one number in this year's 900-designer report moves the wrong way. Four times as many designers say collaboration has got worse. Five things that are actually breaking, and what to change on Monday.
A sparkle no longer counts as telling anyone anything. Five things your interface has to do now that the EU rules are enforceable, and the one exemption every team will misuse.
Most UX metrics look good on dashboards but fail to reflect real user success, so teams should focus on measuring actual outcomes, effort, and task completion instead.
Generative UI is about designing systems that dynamically build and adapt interfaces in real time based on user intent, context, and behavior instead of fixed layouts.
Users are exhausted by AI. And most products are making it worse. The trust deficit is real: high trial, steep drop-off after day 3, low 30-day retention.
Every deck that cuts a list cites one jam display from one grocery store. Pooled across 63 conditions and 5,036 people, the effect is 0.02. Here is what the evidence really supports, and the four things to check before you cut anything.
NASA had a full design system on paper in 1975, and it is better written than most of the ones I see in Figma. Not because of the grid. Because it said no in public, and said why.
Teams can watch every step their AI takes, but barely half test it against a real set of inputs. Deciding what counts as a good answer is a design job. Here is how to write it down.
Scroll driven animation is a few lines of CSS now, and all three engines support it. The defaults are the problem: no reduced motion guard, and motion that runs the whole time an element is on screen. Five fixes and a four question audit.
The way most teams run critique is close to the setup psychologists use to produce conformity on purpose. The fix is not better intentions. It is a different running order.
Designers quote the IKEA effect to justify long signup flows. The original study points the other way. Value appeared only when people were allowed to finish something, and that changes what the first five minutes of your product should do.
Everyone is selling prompting as the new design skill. It is one level too low. The transferable skill is describing done precisely enough to judge what comes back. Five ways designers write adjectives when they should be writing tests.
Users are not scoring your options. They are rehearsing an argument they will have to make to someone who was not in the room. Five places your comparison table arms the wrong choice, and what to ship instead.
In May I wrote that Liquid Glass changed everything. I had the shape of it wrong. iOS 27 hands users a continuous slider over the material and stops honouring the key that let apps stay out of it. Liquid Glass is not an appearance you adopt. It is a range you have to survive.
State of Prototyping asked 1,478 designers how AI is changing their work, and one finding went everywhere: researchers feel most at risk. Then I called the survey's own API and checked the row count. It is 23 people, nine of whom are the frightening number.
Every few years a team strips the labels off its toolbar and calls it clean. Vienna ran that experiment properly in 1925, and the answer has not changed: icons are a helping language, not a replacement for words.
Teams read cognitive load as screen density, so they strip the screen and the hard tasks get harder. The split that matters is between the load the task carries and the load you added.
The 2026 interop list reads like a permission slip to delete code. I checked all four features against Baseline this morning. One is ready, one has no Safari, one is still partial in Firefox, and one is safe to ship anyway for a reason worth knowing.
73% of designers say the bar went up this year. 13% of leaders changed how designers are reviewed. When screens are nearly free to produce, counting them grades the tool, not the designer.
More than 100 alarms went off at Three Mile Island and none of them were ranked. The fix already existed and had for decades. A 2026 survey says most on-call teams have rebuilt the same control room in software.
Pretty prototypes test well. A meta-analysis of 31 studies says they also work a little better, but nowhere near as well as the ratings suggest. Here is how to tell what your last test actually measured.
Your interface can now run as a guest inside Claude or ChatGPT. The host owns the font, the colour, the navigation and the network. Five things you lose, what to do about each one, and a ten minute audit of your own product.
Agents build UI from what they can read. One audit of 37 public design systems found six publishing tokens in the standard format and three shipping Code Connect. Here is what to publish, and in what order.
Beck's diagram has been copied by transit systems all over the world, and most of the copies took the half that does not matter. What he actually deleted, what it cost London, and the test for when you are allowed to get the data wrong on purpose.
People judge an experience by its sharpest moment and its ending. Design writing turned that into permission to skip the middle. Five things the research really says, and a ten minute audit of your own product.
Code layers are still behind a waitlist eleven weeks after Config. The habit they promise does not need them. Five moves you can make today, and a ten minute audit of your product file.
You can spot an agent-built screen from across the room. Six tells, what each one is really missing, and why writing a longer prompt is the wrong lever.
The average designer runs seven AI tools, up from three. The top two things slowing designers down are now learning the tools and juggling them, ahead of output quality. Cut to two, and judge a tool by whether it survived ninety days.
AI gave design a new gear, and one number in this year's 900-designer report moves the wrong way. Four times as many designers say collaboration has got worse. Five things that are actually breaking, and what to change on Monday.
A sparkle no longer counts as telling anyone anything. Five things your interface has to do now that the EU rules are enforceable, and the one exemption every team will misuse.
Most UX metrics look good on dashboards but fail to reflect real user success, so teams should focus on measuring actual outcomes, effort, and task completion instead.
Generative UI is about designing systems that dynamically build and adapt interfaces in real time based on user intent, context, and behavior instead of fixed layouts.
Users are exhausted by AI. And most products are making it worse. The trust deficit is real: high trial, steep drop-off after day 3, low 30-day retention.
