Working With AI, Designing For AI: How the UX Stack Splits in Two

Most writing about AI and UX picks a side. Either AI is the new tool reshaping how we work, or AI is the new material reshaping what we build. Both arguments are true. But practicing designers don't get to pick – we live in both worlds, often in the same week, sometimes in the same project.

The UX stack is splitting into two operating modes. The designers who will matter most over the next few years are the ones who can move fluently between them, without losing the rigor that made UX worth caring about in the first place.

Before going further: the traditional methods that brought us here are not legacy. Research, benchmarking, personas, journeys, flows, wireframes, and validation are why software stopped being unusable. They are load-bearing, and they will still be load-bearing in five years. What changes is everything around them.

The Traditional UX Stack – Still the Foundation

For two decades, UX teams have followed a recognizable arc. We research users and benchmark competitors. We synthesise findings into personas and user stories. We build information architectures, map journeys, define logic layers, design error states. We wireframe, prototype, and validate. Each step compresses uncertainty into something a team can build.

This is not nostalgia. These methods are how good products get made. They survived the shift to mobile, the rise of design systems, and the move from agency to in-house teams. They will survive AI too – but every one of them now has an AI-accelerated counterpart, and pretending otherwise wastes weeks of work that no longer need to be done by hand.

Working With AI – Accelerating Traditional UX

When AI lives in your toolkit but not in your product, the change is mostly about speed and breadth. The job stays the same; the production tax shrinks.

Research and synthesis. Transcription, affinity mapping, theme extraction, and sentiment analysis used to take a week per study. AI compresses them into hours. What stays with the designer: deciding which themes matter, knowing which quotes are signal and which are noise, recognising when a sample is too thin to draw a conclusion. Speed amplifies judgment – and exposes the absence of it.

Benchmarking. Manual competitive audits and feature comparisons were slow and quickly stale. AI can scan dozens of products, summarise patterns, and surface conventions across an industry in a single afternoon. The designer still has to ask the harder question: which conventions should we follow, and which are inertia we should break?

Personas. This is where the biggest evolution sits. Personas don't disappear – they stop being frozen documents. Instead of "Sarah, 34, marketing manager," teams now work with living behavioural patterns that update as fresh data arrives. The persona becomes less a poster on the wall and more a query against current signal. Designers who clung to the static version will find it harder to defend.

User stories and edge cases. AI is unusually good at generating "what about..." scenarios – the unhappy paths, the weird inputs, the conditions a tired team forgets to write down. Treat it as a co-investigator, not an oracle. It will surface real gaps and invent imaginary ones; the designer's job is to tell which is which.

IA, sitemaps, and logic layers. Content clustering, sitemap validation, and logic-tree review benefit from a tireless second pass. AI catches dead ends, redundant categories, and ambiguous labels faster than a human review can. The structural choices – what concepts deserve their own section, how mental models should be honoured – still belong to the designer.

Flows and error states. Missing-path discovery, scenario expansion, and error-state cataloguing all move faster with AI. But the question of which errors deserve a designed response, and how to communicate them with warmth and clarity, remains a human craft.

Wireframes and prototypes. Variant generation is the most visible AI acceleration. What used to be a three-day exploration becomes thirty minutes of comparing options. The risk is real: more variants does not equal better thinking. Designers who use AI to explore widely and then choose deliberately will pull ahead of those who use it to ship faster without choosing at all.

The pattern across all of these is the same. AI doesn't replace UX thinking. It removes the keystrokes between thought and outcome – which means designers who have nothing to think about will be exposed faster than ever.

AI doesn't replace the designer. It removes the keystrokes between thought and outcome.

Designing For AI – A Different Game

When intelligence moves from your toolkit into the product itself, the rules change. Behaviour becomes adaptive instead of deterministic. Outcomes become probabilistic instead of guaranteed. Trust, transparency, and recovery stop being nice-to-haves and become primary design materials. I covered this terrain in my previous article on designing beyond flows, so I won't re-litigate it here.

What's worth adding is one practical shift the previous piece didn't dwell on: the new error states. Traditional UX taught us to design for predictable failures – empty fields, network drops, invalid inputs. AI products fail in unfamiliar ways.

They hallucinate. They produce confident answers to questions they shouldn't have answered. They drift over time as models change underneath them. They return correct outputs in the wrong tone, or partial outputs that look complete. Designing for these failures means treating uncertainty as a first-class state – surfacing confidence levels, offering "show your work" affordances, making it easy to challenge or correct, and ensuring there is always a clean path back to human judgment.

When AI moves from your toolkit into your product, the designer's job shifts from arranging certainty to managing probability.

When You're In Which Mode

A simple question separates the two worlds: where does the intelligence live?

If intelligence lives in your tools – the research synthesiser, the variant generator, the heuristic checker – you're in Mode 1. The product is conventional; AI is invisible to the user. The stack stays familiar. Your wins come from compressing time and broadening exploration.

If intelligence lives in the product itself – the assistant, the recommender, the agent – you're in Mode 2. The user is collaborating with a system that learns and adapts. The stack still applies, but the principles shift: flows give way to fields of intent, predictability gives way to negotiated outcomes, and design becomes the framing of intelligence rather than the direction of action.

Most real projects mix the two. AI-accelerated research feeding an AI-native product. A traditional checkout wrapped around a single AI-powered recommendation. A long-running platform with one new agentic feature. The stack still holds – you just deploy different parts of it, with different intensity, depending on where the intelligence is at any given moment.

The Designer Who Operates in Both Modes

Three capabilities hold across both worlds.

AI literacy. Enough understanding of model capabilities, limitations, and biases to make good design calls. Not engineering. Just enough fluency to know what's reasonable to ask of the system and what isn't.

Systems thinking. Moving beyond the screen to the ecosystem. Systems thinking is becoming the connective layer between traditional and AI-native UX. Designers are no longer only responsible for screens; they need to read operational dependencies, service ecosystems, business logic, data relationships, and how a decision in one surface ripples across every other. AI products especially demand this – they rarely live in a single interface, and the experience often spans channels, models, and human handoffs.

Behavioural fluency. The ability to read live context rather than rely on static personas. The designer who can describe what users actually do, in the moment, with the data in front of them, has an edge over the designer who can only describe who users supposedly are.

The future UX stack isn't replacing research, benchmarking, personas, journeys, or flows. It's expanding what designers can do with them. The craft doesn't disappear – it stops being held together by manual production.

Tomorrow's UX teams may not win by working harder. They'll win by thinking deeper, and moving faster. Because the future designer won't choose between working with AI and designing for AI. They'll learn to do both – fluently, in the same week, sometimes in the same project.

This two-mode shift is how I'm framing the next generation of my own UX practice.

Let's talk about your project.

LinkedInInstagramX