AI can help you listen to users. It cannot listen for you.

UX research is being promised a productivity era of its own. Transcripts summarise themselves. Tickets cluster into themes. A model can produce a persona complete with a name, a quote, and a favourite app. Some of that is genuinely helpful. A long interview is easier to share if someone has pulled out the moments that matter. At Always 49 we still draw a hard line. AI can help you listen. It cannot listen for you. The moment you skip the human, you start designing for a plausible average of what users usually say, which is not the same as what your users need.

Personas generated from a prompt are the clearest example. They feel like research because they are specific. Specific is not the same as true. A made-up operations manager called Priya who “values efficiency and is frustrated by legacy tools” can survive a whole workshop without anyone noticing that nobody spoke to an operations manager. We would rather a messy note from a real forty-minute call than a beautiful fiction. Fiction is how teams fall in love with a solution that was never asked for.

Where the tools are good is in the grind after you have done the real work. If we have talked to eight people, an assistant can help us find repeated phrases, or remind us of an offhand comment in interview six that matches a support ticket from last month. That is pattern spotting. Pattern spotting still needs a researcher who was in the room, who noticed the pause before someone said “it’s fine”, who knows that “it’s fine” often means “I have given up”. Larissa’s job as Head of Research is not to produce artefacts. It is to protect that kind of noticing.

There is also a consent and care problem that the industry is sliding past. Users who give you their time are not raw material for a model. Recordings, tickets, and chat logs can contain distress, health information, or the inner workings of a charity. Feeding that into a generic tool because it is convenient is not user-first. It is extraction. Our view is that research ethics did not become optional when summarisation became cheap. If we cannot explain where the words went, we should not have collected them in that form.

Project process has to hold this too. Discover is a workshop and a set of conversations, not a prompt. We map current processes with the people who live them. We watch the workarounds. We ask what they tried before they called us. That work supports clients because it stops the project being a guess dressed as a specification. If a studio now offers “AI discovery” as a way to skip talking to staff, they are selling you a cheaper way to misunderstand your own organisation. Misunderstanding is expensive. It just invoices later.

We use the same scepticism on supposedly research-driven UI generators. A layout that “matches best practice” is often a layout that matches last year’s SaaS average. Best practice is a useful prior. It is not evidence that this nurse, this volunteer, this warehouse lead, this donor can do the task. Evidence still looks like a person attempting the task while you try not to help. That scene has not been automated. If it ever is, we will still want to be in the room, because the room is where respect lives.

There is a healthier future, and it looks like amplification rather than replacement. Researchers spend less time formatting notes and more time with people. Designers get faster first drafts and then test them harder. Product managers hear clusters of complaints without drowning in a spreadsheet. All of that only works if the organisation still values sitting with discomfort: the user who is confused, the client who is wrong about their own process, the idea that does not survive contact with a prototype. AI is very good at removing discomfort. Discomfort is often the information.

Always 49 will keep using assistants to tidy, cluster, and draft. We will keep putting humans in front of humans. Software that is humane, empowering, and accessible by default does not come from a summary of a summary. It comes from attention. Attention does not scale by ignoring the people you are meant to be paying attention to.