When not to use AI.
There's a quiet damage being done by people who reach for AI on every task. Their judgment atrophies. Their voice disappears. Their colleagues stop trusting their work because it sounds like everyone else's AI-generated output. Knowing when not to use AI is part of using AI well. This lesson covers six situations where reaching for an AI costs you more than just doing the work yourself.
The mental model
AI is a force multiplier. Force multipliers applied to the wrong inputs multiply the wrong outputs.
If your draft email is good and you ask AI to polish it, you get a slightly better email. If your draft email is bad and you ask AI to polish it, you get a polished bad email. Plus you've outsourced the thinking that would have made your judgment better next time. That's the quiet cost.
The question isn't "can AI do this?" The question is "does outsourcing this make me better or worse over time?" Some tasks are pure cost reduction (great for AI). Others are how you develop your judgment, voice, and reputation (terrible to outsource).
The 6 cases to recognize
When your voice is the value
Personal newsletters. Wedding toasts. Sympathy notes. Performance reviews of someone you've worked with for 5 years. These work because they sound like you. AI flattens you into the most-average possible voice. The reader can usually tell, even when they can't articulate why.
When you'd learn something by struggling
The first time you try to write a SQL query, structure a financial model, or draft a contract — that's how you develop competence. Asking AI to do it for you produces a working result and zero learning. Six months later, you still can't do it. Twelve months later, you're dependent on AI for things you should know cold.
When the work needs to be legally defensible
Contracts, regulatory filings, medical advice, official disclosures, anything where being wrong has serious legal consequences. AI hallucinates. It cites statutes that don't exist. It misstates case law with confident specificity. Putting AI-generated text into anything that could end up in court is a quiet career-ender waiting to happen.
When the situation needs human emotional intelligence
Firing someone. Apologizing to a customer after a real screwup. Telling a teammate their work isn't ready. Asking your manager for a raise. AI will produce text that says the right words but smells like AI — and the person on the other end will feel it. These moments are where trust is built or lost; AI can't engineer trust.
When you don't know enough to evaluate the output
Asking AI for legal advice when you can't tell if it's right. Asking AI for medical guidance when you can't recognize the danger signs. Asking AI to write code in a language you don't know — and then deploying it. AI output looks plausible by default. If you can't tell the difference between right and plausibly wrong, you shouldn't trust the output.
When the task takes 30 seconds to just do
Writing "thanks!" Replying to a meeting confirmation. Composing a Slack message that's 20 words. The overhead of opening AI, prompting it, evaluating, and pasting back is often longer than just typing the thing. AI for small tasks is performative — it makes you feel productive without saving time.
The compound costs of AI-everywhere
Beyond the individual cases above, there's a longer-term pattern worth being honest about: people who rely on AI for everything become noticeably worse at the things they outsource.
Writing is the clearest example. People who let AI write all their emails for two years often struggle to write a real one when it matters. Their muscle for finding the right phrasing, judging tone, structuring a paragraph — it weakens like any unused muscle. Same with coding. Same with critical thinking on data.
The solution isn't to avoid AI. The solution is to use AI like a power tool: deliberately, with awareness of what you're outsourcing. Do enough of the underlying skill manually that you stay sharp. Let AI accelerate what you already know how to do.
Final challenge: audit your last week
Look back at the AI prompts you've sent in the past 7 days. For each one, ask:
- Did using AI here actually save me meaningful time?
- Was this a task where my voice or judgment mattered?
- Did I learn anything from outsourcing this, or am I quietly getting worse at it?
- If a colleague had asked me to do this, would I have written it differently than AI did?
For the ones where the answer is uncomfortable — flag those tasks. Stop using AI on them for two weeks. Notice how your work changes.
What you can do now
- Recognize the 6 situations where reaching for AI costs more than it saves
- Spot when your voice is the actual value (and protect it from AI flattening)
- Use struggle deliberately to build skills you'll need long-term
- Keep AI away from legally consequential text and high-stakes emotional moments
- Notice the compound cost of AI-everywhere on your judgment over time