UXCon Vienna 2026 Journal

“Design teams must review agent payloads with the exact same rigor as visual components.”

Intelligence originally meant "to choose between", and the talk traces how computing lost that meaning (Babbage's deterministic engine, IQ as a single number, the bicycle-for-the-mind era) before agents brought choice back into the machine. Once agents plan, run for hours and fail by semantic drift instead of crashing, fixed screens stop working. The design answer is supervisory UX: mission-control views with dry runs, live confidence, mid-flight steering and rollback. And every product now has a second user, the agent, which reads schemas instead of screens, so the tidy data layer carries "semantic debt" and needs the same design rigor as the visual one.

“Skills: problem framing, epistemic judgement, taste.”

The talk opened on a UX Magazine article, "The New Makers", and then showed a landscape cross-section of what makes a designer or researcher: tools on the surface, processes just below, then skills, values and personality as the deeper strata. The point of the picture is that tools and processes are the thin, changeable topsoil, while skills like problem framing, epistemic judgement and taste, and the values and personality underneath, are what actually carry the work.

“We design behavior (not outputs).”

The old mental model was a straight line owned by data science: requirements go into a prompt, output comes out. Kofler's replacement is a loop where the prompt is only the beginning: product requirements and experience goals feed behavior design context and data, the output goes through human review and an LLM judge working from shared evaluation criteria, and what was learned flows back into the goals. Designers own the definition of "good" (understanding, substance, clarity, usefulness, with safety as the baseline), humans calibrate the judgment, and the LLM judge applies it at scale so each evaluation result turns into a hypothesis and a change.

“Lead with value, not the technology.”

Google's Gmail surveys split users into AI Innovators, AI Receptive and AI Skeptics, and the skeptics are the majority, growing, and skewing younger. Their top reason is forced presence: AI features shoved in their face with no control, which is a UX problem rather than an ideological one. The fix is to lead with value instead of the technology and to give real user agency; copy experiments that reframed "AI inbox" as "never miss another important email" and "AI overview" as "prep for this meeting" raised interest sharply, most of all among skeptics and young users.

“If you haven't evaluated a system on all language varieties, you can't be sure it will function as required.”

Speech technology is built around a default English, and everyone whose accent, dialect or life stage sits outside it gets worse results. Kibedi frames this through "technocreep", the slow accumulation of unseen technological relations that hide racialised and gendered histories, and through Bender and Hanna's warning about automatic speech processing in 911 call handling: a system not evaluated on all language varieties cannot be trusted to work as required.

“Your design judgement is more important than ever.”

When anyone can code, ideas are free and speed is the default, the old handoff pipeline with clean PM, design and engineering swimlanes collapses into a tangle. Choi's answer is that designers should claim the two decisions only they can make well: deciding what to build (is this a real problem for a real person, should it exist at all) and deciding when to ship (scoping, reviewing PRs, defining the launch bar, design review as a CI check). Amid the jargon, the thing that matters is taste; design judgement, meaning creative solutions, simplifying complexity, product cohesion and fit and finish, is more important than ever.

“The valuable stuff wasn't in the report in the first place.”

Research used to be slow and expensive, so every report was a special pebble; now teams point AI at everything and produce a cloud of decks, summaries and wikis that pollutes rather than informs. Rosala's argument is that the report was never where the value lived. Knowledge is built by the effort of doing the work (System 2, the self-generation effect) and by shared experience and stories, and AI's illusion of learning removes exactly that effort. Research therefore has to be both the artist who produces evidence and the gallery curator who decides what matters, how it is presented, how findings connect and how people experience them.

“Moving the bar isn't sloppy. Refusing to think about it is.”

How long an experiment takes is set by three dials: traffic (nobody controls it), the size of the change (the designer controls it) and how sure you need to be (the experimenter controls it). A timid design at a 95% habit ran six weeks and came back flat; a bolder design at a deliberately chosen 90% ran four weeks and lifted conversion, revenue and account visits. The confidence bar should move with the cost of being wrong: a microcopy tweak can live with 90%, removing a payment method deserves 99%. Research before testing also matters: tests backed by research win far more often.

“Make critical actions unmistakable.”

Typography is usability. The Oscars envelope mix-up becomes a checklist: clear visual hierarchy, most important information first, never below 14 to 16 px, test under pressure with the squint test, and make critical actions unmistakable. A body typeface should be understated and get out of the reader's way; a before-and-after of a newsletter signup shows how small type changes clean up a component.

“Agents like blueprints too, especially when they're written in markdown.”

The double diamond used to be the designer's whole territory; now the last diamond, Deliver, is something anyone can build, so the designer's edge moves to the front. Greene borrows the architect's way of working: plan what to build and for whom, align the crew on the plan, know the site, then build in order from foundation to details. The plan lives in a markdown blueprint that carries your vision, judgment and taste plus the functionality of the thing you are building, and the first exercise is to shape that plan with the agent before building anything.

“Remember that delight fades away - find new pain points!”

Delight is not decoration. It comes either from brand experiences, which suit niche audiences, or from fixing pain points, which has wide appeal and gives people a story to share. Colborne's recipe is one pattern repeated in variants: find a point of anxiety (experienced, remembered, or deliberately heightened), then resolve it effortlessly, cleverly, with a happy ending, or with a result superior to what peers get. Pick one pain point and fix it completely, measure through user tests and word of mouth, and expect the delight to fade, so keep hunting for the next pain point.