When people use AI to build things (code, apps, documents), they put more effort into directing it up front but less effort into evaluating what comes back. They’re more specific about what they want, and less likely to question the reasoning, check facts or catch what’s missing.

It makes sense. If the code runs or the document looks clean, your brain moves on. You don’t dig into whether the logic holds up. But this is backwards. The stuff you’re building and shipping is exactly the stuff that deserves the most scrutiny.

A few things that have changed how I work with AI:

• I treat every first output as a rough draft, even when it looks finished. Especially when it looks finished.
• I front-load instructions like “flag what you’re uncertain about” and “push back if my assumptions are wrong.” This completely changes the dynamic. Only 30% of users do this, according to the research.

When I’m building something, I make myself ask “what’s wrong with this?” before I ask “how do I ship this?”

Source: Anthropic Education Report: The AI Fluency Index

There’s a name for it: distributional convergence. AI models predict the most statistically likely design choices. That means Inter font, purple gradients, white backgrounds and zero animation.

I’ve been deep in the weeds researching how to fix this with Claude Code, and the most counterintuitive finding was that more specific prompts often make it worse.

Telling Claude “use #2A3F5B for the background” triggers pattern matching. Saying “atmospheric backgrounds with layered depth” triggers creative reasoning. The difference in output quality is night and day.

A few things that actually move the needle:

• Ban the defaults explicitly. Telling it “never use Inter, Roboto, or purple gradients” eliminates the highest-probability generic outputs.
• Reference feelings, not specs. “Spotify met a disco ball at a Nordic design conference” communicates more than a hex palette ever will.
• Lean on typography. Extreme weight contrasts (100 vs 900) and 3x+ size jumps cause Claude to elevate all its other design decisions.
• Install Anthropic’s frontend-design skill. It’s one file that injects design-thinking context and steers toward bold aesthetic choices instead of the safe middle ground.

You can also use v0 or Lovable for initial visual exploration, then bring those components into Claude Code for production-quality code, accessibility and design system integration.

But some work has a ceiling. Formatting decks, internal status emails, cleaning data. You can spend 20 minutes or 2 hours, and the outcome has a capped potential.

Some work has no ceiling. Product positioning, key customer conversations, hiring. A small improvement there changes everything.

The smartest operators I know use AI only on the first category so they can obsess over the second. Try not to use AI to replace your thinking on the things that matter. Use it to eliminate the things that don’t.