Hatch: Design Talks

There are a handful of non-obvious ideas - prompt engineering as a UX/writer responsibility, designing for conversation rather than interfaces, and LLM unpredictability as a novel design constraint - but roughly half the episode is generic career-transition advice and conference promotion that produces no learnable insight.

“it is something where it is essential, it is crucial for UX roles and not just designers, also researchers and also writers to be involved and have a say in how those prompts are engineered”

“you really have to design conversations more than you have to design interfaces”

Reframing prompt engineering as a UX craft and arguing for designing failovers and conversation flows rather than static interfaces are somewhat fresh angles, but the broader 'AI changes everything for designers' narrative is extremely well-trodden by mid-2023 and the career-management section is entirely recycled.

“prompt engineering that sounds that would be an engineer job right...but it's the tool, the biggest lever you have to really impact the experience”

“you can design failovers. You can design how to steer it in the direction that you want”

Goltermann is a practitioner who genuinely shipped an LLM-powered product at a large, ML-mature organisation, which is relevant and credible; however he is a mid-level design manager, not a senior product or technical leader, and the transcript reveals limited depth beyond his immediate team experience.

“I'm now a design manager at Booking.com...My focus is on the way how people search and find accommodation in Booking.com”

“Booking is quite a mature organization set up around machine learning. So there's a machine learning hub in Tel Aviv and there's machine learning scientists and principal engineers”

Almost entirely devoid of concrete numbers, timelines, or measurable outcomes - no conversion data, no accuracy metrics, no user research findings from the trip planner, and only a single named example (the Tel Aviv ML hub) to ground the claims.

“there's a machine learning hub in Tel Aviv and there's machine learning scientists and principal engineers and all these kinds of things”

“even if you go to chat GPT and you put in the same prompt twice, you will get a different answer the second time than you got the first time”

The host's questions are almost entirely open and soft ('how did you handle that personally?', 'was there anything specific that you found surprising?'), with no follow-up challenges, no requests for data, and frequent pivots back to conference promotion rather than productive probing.

“And how was actually to make that transition between like individual contributor into a manager? What do you think were the challenges that you had doing that?”

“Has there been any moments building these new features powered by AI where you were like, okay, this really changes everything”