The Hidden Cost of Using AI for Everything

Everyone talks about the API bill. That is the boring cost. The interesting costs are the ones that do not show up on a monthly invoice but quietly eat your attention, your judgement and your output quality. Let me lay them out.

Cost 1: Decision fatigue

You ask the model for three variations of a tagline. You get three. Now you have to pick one. That sounds trivial — it is not. Each pick is a small decision, and decisions have a daily budget. Before AI, you made the choice once — you wrote the thing. Now you make it four times: write the prompt, read three outputs, pick one, tweak it.

Multiply this by every task in your day. By 4 PM your decision budget is blown, not on deciding strategy, but on choosing between AI-generated options. You are tired in a way that feels different from old-fashioned hard-work tired, and it is.

Cost 2: Verification tax

Reading to check is slower than writing. This is a deeply unintuitive fact, and it is why "AI first draft, human edit" is often slower than "human draft" for tasks under a certain length.

For an 80-word email, I can write it in 90 seconds. If I prompt, I need maybe 20 seconds to write the prompt, 10 seconds for the output, and 90 seconds to read it carefully enough to trust it. Net loss. For a 5,000 word document, the math flips — AI saves you hours. But most people automate the 80-word emails and not the 5,000-word documents, which is exactly backwards.

Cost 3: Death of serendipity

This is the quietest and most expensive one. When you prompt a model for "three product name ideas," you get three predictable product name ideas. You do not stumble into the weird association that would have been your best name if you had been walking and thinking for twenty minutes.

Creativity requires wandering. Wandering requires time. AI offers you a shortcut that skips the wandering entirely, and the shortcut is narrower than the wandering was. You get to the answer faster but the answer is worse because it is the average of every similar answer the model has seen.

Cost 4: Loss of detection instinct

Writers who prompt too much start recognising AI cadence less. Engineers who auto-complete too much stop recognising bad code patterns. Designers who AI-generate too much stop noticing subtle visual wrongness. Your detection instinct is calibrated by doing the work; skipping the work decalibrates it.

When your own instincts are sharp, AI is a force multiplier. When they are dull, AI is a mask over mediocrity and you do not know which is which.

Cost 5: Identity drift

This one sounds soft but I mean it. Craft is part of identity for a lot of people. If you used to feel like "a writer," and now a model does most of your drafts, the feeling goes away slowly. You do not become "a writer who uses AI." You become "a person who edits AI outputs." Some people are fine with this. Many are not, and they find out six months later when they notice they have stopped identifying with the craft they used to love.

The rule I actually follow

One line: if the task is the thing I most want to get good at, I do it myself. If it is not, I automate it without guilt.

• Writing long-form thinking (this blog) → I do it myself, AI only for copy-edits
• Scheduling, routine emails, admin → AI, automated, no guilt
• Hard engineering problems for my product → I solve, AI helps with boilerplate
• Code that I do not want to learn (regex, bash scripts) → AI all day
• Product strategy and positioning → I do it, AI is a sparring partner not a writer

Autopilot is great until you realise you forgot how to fly the plane.

The goal is not to use less AI. The goal is to use AI in the places where automation makes you stronger, and do the work yourself in the places where automation makes you weaker. That line is different for every person. Figuring out your own version is a skill.