AI Makes Average Cheap. Your Job Is to Buy Something Better.

Your tools can now spit out a thousand plausible screens before lunch. That sounds like a win until you look at what comes out. It works. It's fine. It's forgettable. Let me catch you up on what a stack of essays this month says about where your team's real value moved, and what you should protect on the roadmap because of it.

Average is the default setting

Generative AI pulls toward the middle on purpose. It learns from patterns that repeat, so it hands you the safest, most common answer. Thomas Sokolowski tried Claude's new design feature and called the results "algorithmic mediocrity". Not terrible. Not distinctive. Just the average of the average.

He makes a point worth sitting with. Roughly 5 percent of tech websites are genuinely bad, 5 percent are excellent, and 90 percent are plain average. AI is a channel that takes average input and makes more of it, faster. He calls it a mediocrity amplifier. If your team leans on it to produce, you get the same output, just sooner and sloppier.

The scarce skill is knowing which one deserves to exist

When making one screen was hard, making it was the job. Now that making is cheap, the value shifts to choosing. Patrick Neeman put it plainly: when a tool can generate a thousand screens before lunch, the scarce skill is knowing which one deserves to exist. That is judgment, and it does not come free with the subscription.

Addy Osmani draws the same line. The durable thing is not getting better at problems that already have a known answer. It is picking what to build and judging whether it is any good. Phil Morton makes the mechanical case: you cannot build an automatic grader for taste, because there is no correct answer to grade against.

Taste is a library you have to build

Taste is not a vote. Fabricio Teixeira argues we keep treating it like something you can collect, measure, or encode, as if enough data settles it. It doesn't. Taste starts with knowing what good looks like, built from exposure to strong work, from studying proportion, type, and composition until you recognize quality on sight.

That has a real cost for how your people spend time. Sokolowski notes many so-called AI experts now run days that are half maintenance, a third admin, and a thin slice of actual output. Chasing every new model eats the hours you need to build the reference library in the first place. Learning new tools is not the same as building judgment.

Solutions don't save a project. Strategy does.

Good craft with no strategy burns money. Zeeshan Khalid watched a team log over 2,000 hours on a six-month project that ran eight. Interviews, journey maps, high-fidelity mockups, all of it beautiful. Not one artifact got approved. The veteran lead kept saying they needed to refine the solution a little more. The problem was never the solution.

This is the trap AI makes worse. Faster production without strategic alignment just gets you to "nothing to show for it" quicker. Tanner Christensen adds the part that stays yours no matter how cheap making gets: accountability does not go down with the cost of production.

The deep cut

The risk is not that AI produces bad work. It produces fine work, and fine passes review. Your team will ship more, faster, and feel productive while the output flattens into the same 90 percent everyone else ships. The fix is a budget decision. Protect the hours your senior people spend judging and choosing, and stop measuring them by volume of screens. Roger Wong noted that experienced people get more useful work from each prompt, so the payoff from their judgment compounds. Spend their time there, not on operating the machine.

Three questions for your team

• Look at your last three shipped features. Which ones were chosen on purpose, and which ones were just the first plausible thing the tool handed us?
• How much of our senior designers' week goes to judgment and strategy versus producing screens and testing new tools? Are we rewarding the right one?
• Before we generate a single option, can we say in one sentence why this thing should exist? If not, why are we in the tool yet?