- The Last Two Weeks
- Posts
- Stars, Bubbles, and the Best AI Model
Stars, Bubbles, and the Best AI Model
Plus the wrong way to design with code.
Ted Goas
December 19, 2025
đ What Iâve been reading
Yes.
Well, probably. But thatâs not always a bad thing. In The Benefits of Bubbles, Ben Thompson notes the infrastructure and knowâhow left behind become the runway for the next decade.
A host of companies went bankrupt after a frenzied period of building far more fiber than could ever be justified by current usage. [âŚ] That fiber, however, became the background of todayâs Internet; it basically existed for free.
The 2001 Dot Com bubble wasn't all bad. With so many businesses getting online, it forced us to build the fiber infrastructure that we're using today. It also got brick-and-mortar businesses to realize the internet was the future and everyone started rowing in the same direction. The same can be said of the AI era we're in.
The same can be said of the AI era we're in.
The real bottleneck for AI is power as in energy. We just donât have enough, and thatâs putting pressure on governments and businesses to speed up energy innovation. Benedict Evans also calls this out in his most recent presentation.
Power generation is exactly the sort of long-term payoff that might only be achievable through the mania and eventual pain of a bubble
OpenAI and Anthropic could go out of business, but they're forcing us to upgrade our energy infrastructure, which is something we'll always need.
Something to temper my cynicism đ
You ever notice how, when an important project comes up at your company, the same (often senior) people get tapped nearly every time? Well⌠its not just your company. This sorta thing happens everywhere.
This is what Hardik Pandya refers to as star performers. While these stars handle highâstakes problems, strong systems deliver consistent results at scale. You need both.
Systems raise the floor. They ensure consistent quality on routine work. [âŚ] Star performers raise the ceiling. They handle the novel stuff, the ambiguous stuff, the stuff you genuinely havenât seen before.
My takeaway: When building teams, create strong systems for reliable execution, and deploy your best people for the rare, highâstakes moments.
The goal isnât to make your entire team interchangeable at the highest level. Itâs to accept that star talent will always be non-fungible, and build accordingly.
Last week AI code editor Cursor announced a visual editor in their product and the community has opinions. My favorite is from Linearâs Karri Saarinen:
Once you become the builder, you start making more conservative bets. You gravitate to what you already know is feasible or supported. You make smaller iterations. You stop dreaming something big. This is not design.
This has been in the back of my mind for a while now.
I get that using AI to boil down requirements docs or brainstorm ideas, but it's not a great for designing. Dropping a PRD into Figma Make or Windsurf and iterating from the output is not going to produce anything great. It is not a substitute for using your brain.
Remember AI is epically bad at creativity.
Design to code tools are useful. They are useful for coding something when you donât feel like designing.
Take this Design Prompts site I found last week: I'm currently redesigning my site (aren't we all) and already had a good idea of what I want because I've done that work the hard, sometimes slow, boring way. This site helped me fine-tune small bits, but out of the box all of the designs on this website suck. They all scream âGenerated by AI!â This is not where we want to be.
My message to follow designers: Before going all-in on AI'ing your entire design process, remind yourself what got you here. And this is coming from someone who can design in code without AI.
(If you canât tell, I was hot when I wrote this.)
𧜠Stray Links
Iâm usually not into daily games like Wordle or Connections, but Iâve been hooked on Clues by Sam for a month now. It tickles my brain in just the right way.
I posted on LinkedIn about what makes someone a senior designer, which stemmed from an article about what makes someone a senior engineer.
Good read for anyone working on AI products: UX Is Your Moat (And Youâre Ignoring It). The best model doesnât win, itâs the product with the stickiest UX around the model that wins.
I use AI everyday, but I donât use it for design that much. I mostly use it to speed up all the mundane tasks around design. To that end, I love how Googleâs Marily Nika shows how she uses AI to do market research and decide what to build. Even watching just the first 9 minutes is mind blowing! Another tactical episode of How I AI.
Hyvor Post is a privacy-first newsletter platform that reminds me of Substack without the social networking effect or, ahem⌠baggage.
Someoneâs trying to reclaim Twitterâs âabandonedâ trademarks (Twitter, tweets, the blue bird) since X has no pans for them. Sounds like it wonât work, but one can hope!
TIL Finland has income-adjusted fines. For example, a Finnish millionaire was hit with a âŹ121,000 speeding fine. More of this please!
đ Some nice eye candy: Shopifyâs Winter Release edition. In fact, all of their release pages are nice!
â One Good Quote
AI canât do your job, but an AI salesman can convince your boss to fire you and replace you with an AI that canât do your job.
âł A classic I periodically re-read
Leaders arenât born, theyâre made. Back when I was trying to grow past the âSenior Designerâ role, Mazz Mosley article on 24 Ways helped me understand leadership as a set of skills I could learn, not a title I had to earn. Still a favorite.
Thanks for reading âď¸
- Ted (@tedgoas)