Sascha Hjorth: how AI actually fits into a UX/UI engineer's daily work

Sascha Hjorth published this piece in Medium’s Bootcamp publication on May 26, 2026. The article takes a deliberately narrow position: instead of surveying what AI can theoretically do for designers, Hjorth describes three specific tasks where local large language models changed his process in concrete ways.

The tools Hjorth uses are not the cloud-hosted defaults. He runs LM Studio on his own machine to host local models, which means his experiments stay private and don’t require subscriptions. This constraint shapes the article’s tone—it reads as a practitioner’s report from controlled conditions rather than a promotional account of new platform features.

The three workflows Hjorth describes:

Design feedback and iteration. He feeds prototype screenshots and written descriptions to the local model, asking for critique of visual hierarchy, call-to-action placement, and interface grounding. The AI does not tell him what to do—it surfaces friction he then investigates himself. He treats the model as a provocateur rather than a decision-maker.

Research validation. After qualitative interviews, Hjorth loads the transcripts into LM Studio and asks the model to synthesize themes with supporting quotes. The output helps him spot patterns he might have anchored on too strongly during live sessions, then verify them against the raw data before writing findings.

Data interpretation. When faced with a set of user feedback metrics showing a spike in negative responses, he used an LLM to help put the numbers in context. The result—negative responses represented 0.04% of total page visits—prevented a reactive design change driven by a visible but statistically minor signal.

The central argument is that AI’s value in design work is not in generating outputs but in challenging existing assumptions. Hjorth explicitly writes that he refuses tools that don’t serve a specific, intentional purpose in his practice. The article is useful for designers who are skeptical of AI integration as a category but are open to targeted, verifiable uses in research-heavy work.