Design Process in the now

For most of my career, "design process" meant a slide. Discover, Define, Develop, Deliver. Research, synthesis, ideation, workshops, prototype, test. It was a story you told a client to justify eight weeks of work, and it was also, mostly, true — that's roughly how the good projects actually went. The diamond shapes weren't decoration. They were the schedule.

That schedule doesn't exist anymore, not in the same way. AI didn't kill process — it collapsed the parts of it that were really just labor. A synthesis pass that used to take three days of sticky notes now takes an afternoon. A first-draft flow that used to require a junior designer and a week now exists in minutes, in ten variations instead of one. What's left, once you strip out the labor, is a smaller and more honest question: what was the process actually for?

I've landed on an answer, and it's the same one I'd have given fifteen years ago, just with the fat trimmed off: process was never about the artifacts. It was about forcing yourself to understand a problem before you started solving it. Everything else — the workshops, the personas, the beautifully tabbed Miro boards — was scaffolding to make sure that understanding actually happened, because understanding is the step people skip when they're in a hurry. AI makes it easier than ever to skip it faster.

The future of "TV" is the clearest version of this I've worked on. A major telecom wanted to reimagine what "watching TV" even means when the box under the set is basically irrelevant. The temptation, with tools that can generate a plausible interior UI in seconds, is to start there — mock up ten home-screen concepts and let the client pick a favorite. We didn't. The ask stayed the same as it would have in 2016: what job is this product actually doing for a household, and where does the current experience quietly fail that job. AI entered later, and only where it earned its place — compressing the exploration phase once we knew what we were exploring, generating variation once the constraints were real. The brief didn't get faster. The distance between having a brief and having options did.

Reimagining the Assortment Process, for a major US retailer, ran the same way from the other direction. That project was never going to be about interface polish — it was about a merchandising workflow buried in spreadsheets and tribal knowledge, used by people who'd tell you, correctly, that no dashboard was going to fix a process problem. The research there couldn't be compressed at all. You cannot AI your way into understanding why a category manager distrusts a system; you have to sit with them while they use the old one. What compressed was everything downstream of that understanding — once we knew the actual decision points in their workflow, drafting the interaction models and testing variations against real data happened at a pace that would have been unthinkable a few years ago.

That's the pattern across the current work, mining and healthcare and banking included: the front end of the process — the part that requires being present with actual humans and genuinely not knowing the answer yet — hasn't gotten one bit faster, and I don't think it should. That's not nostalgia for whiteboards. It's that understanding a problem is a human-paced activity almost by definition; you're limited by how fast people can articulate what they actually do and why. The back end — synthesizing what you heard, generating options, iterating on craft — has gotten dramatically faster, and that's unambiguously good, because it means more of the schedule is spent testing ideas against reality instead of producing the next deliverable on the plan.

The risk, and I see it constantly, is running the process backwards: using AI's speed at generation to skip the slow, human part, because the slow part is the part clients are most willing to cut from the timeline. A workshop that "just" produces alignment doesn't look like progress. Ten AI-generated concept screens look like enormous progress. But concepts built on an assumption instead of an understanding are just handsome guesses, and they cost more to unwind later than they saved up front.

So the process I actually run now looks lopsided compared to the old diamond-shaped diagram. It's front-loaded on the human, unglamorous, can't-be-automated work of figuring out what's true, and back-loaded on fast, AI-assisted iteration once that's settled. Fewer workshops than a Fjord deck from 2019 would have prescribed, but each one working harder, because I'm not padding the schedule with busywork anymore — every hour I ask someone to spend with me has to earn its place, the same way the tools do.

Twenty years in, the honest version of "design process" isn't a methodology I follow. It's a discipline about sequencing: understand slowly, build fast, and don't let the tool's speed talk you into reversing the order.