šļø I walked past 10ish startup booths at Datadogās DASH 2026 last week. By the eighth one, I could basically predict the exact landing page before I got to it.
Same centered hero. Same two big gradient buttons. Same rounded cards with the same soft shadow. Same clean sans-serif font (iykyk). Same dark mode, same single accent color.
I could immediately tell that almost all of these products were built heavily with Claude Code or OpenAIās Codex. I use these tools every single day, so I can spot it the moment I see it. And donāt get me wrong, these are super talented people shipping insanely fast. The velocity is real.
One of my managers put me onto a line she heard on Lennyās Newsletter this week: āAI has no taste.ā Walking that floor at DASH, it clicked.
Velocity !== taste. AI is REALLY good at giving you the āaverageā answer. The safe layout. The default component. The thing 10,000 repos did. You ask it to build a landing page and it builds you THE landing page, basically the average of everything it ever scraped on the internet.
And thatās the trap. When everybody builds from the same model, everybody ships the same product. A product that looks like every other product is a product nobody remembers, I can guarantee it.
And to be clear, I do think itās totally fine for your actual codebase to lack taste. Code has a target: optimal complexity, the simplest thing that solves the problem, and once you hit both, youāre done. Design and product though? They have different targets. Thereās no āoptimalā accent color or āoptimalā feature. Thatās where the taste lives, the SPICE: knowing which idea is even worth building (most arenāt), killing that default gradient, making the one weird, opinionated call the model would never suggest on its own. That part you still canāt automate (at least for now š).
So maybe the new senior engineering skill isnāt just building fast. We can honestly all build fast now. Itās being the one with the eye, for product and design, to spot the median answer and say: nope, not that one.
Genuinely curious where you land: is ātasteā the real moat now? Or does good code have taste too, and Iām drawing the line in the wrong place? š§
P.S. Evidence this photo wasnāt AI-generated: Iām caught mid-sentence asking the photographer to take it in landscape instead. šø
Related: Building the eye to spot the default answer and steer against it is exactly what I coach, whether youāre an engineer or a product manager. More here: alextong.me/services