At our latest Have It For Breakfast podcast we ran a live experiment; redesigning the 1941 Willys MB Jeep for 2030; matching the estimate price point of $20,000 (adjusted for inflation), flat-pack, repairable anywhere, one hour on the clock and same AI tools for everyone (Vizcom).
Henrikโs Willys Jeep 2030 Concept
๐๐ฒ๐ป๐ฟ๐ถ๐ธ ๐ฆ๐๐ฎ๐ป๐๐ถ๐ด, interning in exterior design at Zeekr in Gothenburg, worked backwards from manufacturing. Sheet metal, CNC water-jet cuts, brake-press bends. Panels that could be swapped between front and rear so damage became cheap to fix. A roof that detaches into a boat. He was refreshingly blunt about the tool itself: โItโs a lot of hit and miss. When I sketch something out, Iโm in a lot more control.โ His craft led. AI followed.
Ericโs Concept
Willys Jeep for 2030Willys Jeep for 2030๐๐ฟ๐ถ๐ฐ๐ธ ๐ ๐ฎ๐ฟ๐๐ถ๐ป๐ฒ๐, an industrial designer in Munich, refused to design a car at all for the first thirty minutes. He generated abstract modular sculptures from blocky sketches, iterating at different prompt strengths until the form language felt right. Only then did a chassis appear underneath. What emerged was a narrow quadricycle with swappable rear modules: pickup, cargo box, food truck, tow crane. Less a vehicle, more a platform.
Rahanโs Concept
๐ฅ๐ฎ๐ต๐ฎ๐ป ๐๐ผ๐ ๐น๐ฒ๐, a creative director in Singapore, started with people. Africa in 2031. Merchants, nurses, soldiers. He built the world before he built the product. His output looked like something Charles Eames and Buckminster Fuller might have argued about: balsa frame, polycarbonate panels, an E Ink dashboard, the whole thing clearly inviting you to tinker with it.
Three methodologies. Three outputs that donโt look like they came from the same tool, because in a meaningful sense they didnโt..
๐ง๐ต๐ฒ ๐ฐ๐ผ๐๐ป๐๐ฒ๐ฟ๐ถ๐ป๐๐๐ถ๐๐ถ๐๐ฒ ๐ฝ๐ฎ๐ฟ๐
๐๐ฟ๐ถ๐ฐ ๐ฆ๐๐ผ๐ฑ๐ฑ๐ฎ๐ฟ๐ฑ, who runs Car Design Academy with ๐๐ฎ๐๐ถ๐ฑ โ๐ก๐ฒ๐บ๐ผโ ๐ก๐ฒ๐ฎ๐น, taught the framework behind the session, put the problem simply: if you ask AI for a car, you get a car. A derivative one, averaged from every car image in the training data.
So the designers didnโt ask for cars. They asked for sculptures, for worlds, for sheet-metal furniture. Brand cues went in at the end of the prompt chain, not the beginning. Eric demoed this beautifully earlier in the session with a sports car concept that started as two lines and a fluid curve, and only earned the word โPorscheโ in its final generation. The iconic elliptical headlamps made the cut. Almost nothing else did.
๐ช๐ต๐ฎ๐ ๐ฎ๐ฐ๐๐๐ฎ๐น๐น๐ ๐ฐ๐ต๐ฎ๐ป๐ด๐ฒ๐
Speed is the obvious headline, and itโs the least interesting thing about this. The interesting shift is that Stoddard and Neal are now running AI-generated customer focus groups on AI-generated designs, resetting the same virtual panel repeatedly without the contamination a real panel picks up after its first exposure. Theyโve also put non-designers through the same framework, marketers, engineers, one former Wall Street trader, and watched them produce work indistinguishable from trained designers. The work in this session was distinctive because three designers with sharply different points of view refused most of what the tools offered them. Taste was the limiting reagent, not compute.
If you want to see the next one, weโre planning another 60-minute challenge with a different product. Tell us in the comments what youโd want redesigned.
Originally published in the For Breakfast newsletter on LinkedIn, April 22, 2026. Full video episodes are on YouTube at @HaveItForBreakFast.