Disclosure is a component now, so put it in the system
Once telling users what is machine-made is an obligation rather than a courtesy, it stops being a one-off screen and becomes a primitive. Most design systems do not have one.
Journal
Notes on design, engineering and shipping products, from the allan.ltd studio.
Ten models in two weeks is an architecture warning
The trackers counted roughly ten new model releases from six providers in the first half of this month. If your product has one model's name written into it in more than one place, that pace is a problem you already have.
Being citable is a different job from being crawlable
Getting read by an answer engine is access. Getting named as the source is attribution. We spent a day on the second one and most of it was deciding what identity we wanted to have.
Your single-page app is invisible to the systems answering questions about you
We measured what a crawler without JavaScript could read from this site. Every article came back as zero words. Google was fine. Everything else was reading a blank page.
AI fluency became a hiring filter, and portfolios have not caught up
73 percent of design hiring managers now want proficiency with AI tools and 79 percent want experience designing AI products. Those are two different skills and most portfolios show neither.
Designing for the agent that acts, not the one that answers
Gartner expects 40 percent of enterprise apps to carry task-specific agents by the end of this year, up from under 5 percent last year. Chat patterns do not cover any of the hard parts.
Components an agent can reason about
Spotify rebuilt their component architecture into independent layers so an AI has less to hold in its head at once. It is the most practical design system idea I have heard this year.
Crawled a lot, referred nothing
Automated requests are now about 57 percent of web traffic, and the agents doing the clicking grew roughly 8,000 percent in a year. The number that should worry you is not the traffic. It is the referrals that did not come back.
I found out my own site was blocking AI crawlers, and I never chose it
We spent a day making this site legible to answer engines. Then I read the robots.txt actually being served and found half the work was already cancelled by a setting nobody on our side had touched.
The AI Act stopped being a legal problem and became a UI requirement
Article 50 of the EU AI Act became enforceable on 2 August. The obligations it creates are not paperwork. They are screens somebody has to design, and most teams have not drawn them yet.
Testing a feature that answers differently every time
Traditional QA asks whether the output matches the expected value. With a model behind the feature there is no expected value, and most teams respond by not testing it at all.
AI features have a unit cost, and design decides it
Every other interface decision you make is free to run. This one is not. A single default setting can multiply what a feature costs to operate, and the person choosing it is usually not looking at the bill.
Why AI products lose people in week two
Signups look excellent, week-one usage looks excellent, and then it falls off a cliff. The pattern is consistent enough across products to suggest the cause is structural rather than a marketing problem.
The accessibility deadline that already passed
The European Accessibility Act became enforceable in June 2025. Most product teams I speak to still treat accessibility as a backlog item, and the penalties in some member states are calculated as a share of turnover.
Writing documentation that an agent will read
Your README was written for a person who can skim, infer and ask a colleague. It is now being consumed by something that does none of those, and the result lands in your codebase at volume.
What actually happened to design roles in the last two years
Adoption of AI design tools is near universal, but the headcount collapse people predicted did not arrive. Something else happened instead, and it shows up clearly if you look at which briefs are getting paid for.
Your design system isn't ready for AI agents
A system built for humans encodes what things look like. Agents need to know why. Here are the four gaps that turn up in almost every audit, and what closing them actually involves.
When visual style stops being a moat
For a decade a recognisable look was something clients paid a premium for and competitors needed months to copy. That gap closed. What is left to sell is the part that was always harder to put in a deck.
Designing for generative UI: constraints instead of screens
When the model composes the layout at runtime, the screen stops being the deliverable. What replaces it is a set of rules about what the interface is allowed to become, including the worst version it can produce.
Reviewing AI-generated code: what to keep and what to delete
Agents write working code quickly and generously. The generosity is the problem. Most of what comes back is speculative, and the skill that matters now is deciding which parts were load-bearing.
How to tell whether your AI tools actually made the team faster
The average designer's toolstack went from three to seven in a year, and nearly half of teams have not settled. Everyone reports feeling faster. Very few have measured whether the work moves through the studio quicker.
Pricing design work when AI made you faster
Nearly a third of agencies are getting pushback on hourly rates, with clients naming AI as the reason. They have a point, and arguing the point is the losing move. The problem is upstream, in the unit you sell.
Designing interfaces that AI agents can actually use
Gartner expects around 40% of enterprise apps to embed task-specific agents by the end of this year. Those agents will use your product without looking at it, and several common design habits stop working entirely.
Designing for uncertainty in AI features
Most AI features ship with one visual state: confident. The model is wrong some of the time and the interface looks identical either way, which quietly transfers the whole verification burden to the user.
The case for calm interfaces in AI products
Shimmer borders, streamed text, sparkle icons. Most AI products are currently signalling their own cleverness on every interaction. That reads well in a demo and badly on the fortieth use of the day.
Why interaction detail matters more than layout now
A model will produce a competent pricing page in seconds, and it will look like every other competent pricing page. What does not come out of the box is how the thing behaves when you touch it.
The production tier of design work is gone
Not design jobs in general. One specific tier: the work where the brief arrived fully specified and the task was to render it. That tier paid a lot of rent and taught a lot of people their craft.
Prototyping to make a decision, not to get approval
Most prototypes are built to survive a review meeting: perfect data, happy path, nothing that could go wrong while someone senior is watching. They always get approved and they teach nobody anything.
Naming design tokens for machine consumers
Tokens used to be an internal convenience. A badly named one cost a moment of squinting. Now agents read them, and a name that describes appearance instead of purpose gets applied wrongly at scale.
What belongs in a design portfolio now that polish is free
The visual quality floor rose sharply and made most portfolios indistinguishable. A beautiful final screen now demonstrates something a model does in a minute. It says almost nothing about the person who made it.
Moving a studio off hourly billing: what broke
Around 38% of agencies have shifted at least one service line off hourly. The advice is right. The difficulty is undersold, and it lands in one specific place that nobody warns you about.
The limits of adaptive interfaces
Personalisation is the trend of the year and it has a failure mode that rarely makes it onto the slide. When an interface reshapes itself per person, it stops being something two people can talk about.
Why small senior teams now ship like large ones
The gap between what three senior people deliver and what fifteen deliver has narrowed sharply. The cause is not that individuals got faster. It is that the coordination work justifying the larger team mostly disappeared.
The visual signature of generated work, and how to break it
Users cannot articulate what is wrong with the page. They just trust it slightly less. The tells are consistent enough to list, and none of them are mistakes, which is what makes them hard to argue about.
Design handoff: the specification was never the problem
A decade of tools promised to fix design-to-development handoff, and they largely did fix the file transfer. Builds still come back wrong, because the thing that was missing was never in the file.