Judgment, not production, is now the bottleneck. Producing variations costs next to nothing, but deciding which is right costs what it always did, and that judgment runs on evidence as much as instinct. As the loop shortens, teams that learn fastest don't just win. They pull further ahead.
I Excellence
Every design team uses AI now, and AI takes all of them to the same place: roughly eighty percent, fast. Design closes the remaining twenty percent, the gap between an experience people trust and one that merely functions.
Polish hides the flaw. Output arrives finished-looking, which makes what is wrong in it harder to find rather than easier.
Research run against synthetic personas reads as completely convincing: full profiles, direct quotes, recommendations. Calibrating it against real studies put it at eighty to eighty-five percent accurate, which is enough to point a direction and not enough to settle a decision. Knowing which of the two you are doing is the whole skill.
Speed is only worth what you spend it on. Every team's cycle got faster. The difference is what you do with the time it gives back.
Research coverage went from one market to three without the team growing. The efficiency went into scope rather than into a smaller headcount, and the expansion was only possible because of it.
Consistency is built into the system. An end-to-end AI-native pipeline captures design intent and removes what is not AI-native, so human review can concentrate where there is no precedent to inherit.
A design file describes an intention. Code describes what shipped. The design system I built on a recent rebrand lives entirely in code for that reason, and the client's own designer shipped their pages with it unaided.
Craft that does not wait on a calendar. Content and research expertise gets encoded, so a team can reach a good standard without booking the person who owns it.
Tone of voice, glossaries and a content design system were encoded into AI skills that any team could run directly. Content designers moved from writing every line to setting the standard and taking the hard problems.
II Becoming AI-native
Tools are easy to buy and easy to hand out. Changing how an organisation works is neither, and that gap is the difference between a team that uses AI and one that is AI-native. Most programmes stall there.
The hard part is operationalising, not experimenting. The work is building processes nobody has built before, persisting when the result is not what you needed, working out why it was not, and going again.
Failure still has to be safe, but what it takes now is stamina rather than permission.
Lead with a trial, not a policy. Every new capability runs into the same wall: it needs some other function to work differently. Asking a whole function to change starts a negotiation. Asking for one exception starts on Monday.
Designers shipping their own fixes to production meant engineering changing how it reviewed and released, and one fix used to take two weeks of asking to get reviewed. The trial was a single designer paired with one engineer who would clear blockers of any size. She went from copy corrections to running 74 variants against each other in live production, and the arrangement widened from there.
Change requires protected time. The constraint on adoption is not curiosity or tooling. It is that nobody has time, and only a leader can create it and then defend it.
We protected four hours every second week for every designer and researcher across a team of around twenty, named it the team's first priority for six months, and I took the escalation myself whenever anyone came for those hours.
III People
A tool arrives ready to use. The capability to use it well has to be grown, and it does not happen by accident.
Double down on the people already doing it. Willingness cannot be mandated. Put the weight behind the people who already have it, and let their results do the persuading.
What spread was the way of working itself: reserving time for exploration, then investing in defining, building and operationalising new processes. Brand and marketing took it up first, then product management, then most of the company. Nobody was instructed to.
Ensure everyone knows which skills to develop. People develop deliberately when they can see what is expected of them. A vague instruction to use AI more is not something anyone can act on, so I rewrote what good looks like at every level, across product design, content design and research.
IV AI-native leadership
Judgment cannot be automated, and it has always been what a design leader is for. The role has not moved. What changed is that you can no longer do it from a distance.
Leaders cannot set direction on work they have never done. A leader who has not used the tools cannot tell what is actually possible, which constraints are permanent, or which bets are worth making. Direction moves to whoever has the most recent experience rather than the best judgment.
A design system for a startup, a brand tool that generates a dynamic identity, this site. Every one of my projects starts with AI now. Doing the work is the only way to know what the tools are good at this month rather than last quarter.
Knowing which altitude a problem needs. Staying close to the craft is not the same as taking the team's work. Some weeks the right altitude is a pricing conversation with the executive team, some weeks it is building the thing myself.
I work through managers, including when the idea is mine, and go direct only on the projects that decide the year.