← All articles · Guides · about 5 min read
In the AI age, how to actually use AI well
Skip the hype for a second. The useful framing isn't "AI will replace us" or "AI will save us" — both are slogans someone's selling. The useful framing is simpler: you now have a very fast, very confident junior assistant who has read everything and understood none of it personally. Treat it like that person and you'll get a lot out of it. Treat it like a boss and you'll ship mistakes with a straight face.
Hand it the boring 80%
AI is genuinely good at the work that's tedious but not deep:
- Drafts. The blank page is the hardest part. Ask it for a first version of the email, the outline, the SQL query, and edit from there. You're not lazy; you've removed the friction that stops you starting.
- Summaries and rewrites. A long thread, a dense paper, a changelog — have it condense, then you decide what matters.
- Boilerplate code. The 40 lines of setup nobody wants to type. Generate, read, adapt. (You still read it. More on that below.)
- Translations and tone shifts. "Make this sound polite but firm" is a task it does in a breath.
- Brainstorming. Stuck? Ten bad ideas from a model often shake loose your one good one.
None of this needs the AI to be right about the world. It needs to be fluent, and it is.
Keep the 20% that involves judgment
Here's where people get burned: they hand over the part that requires knowing, not sounding. The model doesn't know your client, your bank balance, or whether the fix will take down production. It guesses plausibly. So:
- Any factual claim — a stat, a law, a date, a medicine, a library version — you verify. AI invents citations that look perfect. If it matters, you check the source yourself.
- Anything sent to a real person — a customer, a boss, a patient — you've read it end to end. "The AI wrote it" is not a defense when it's wrong or cold.
- Any code that runs — you understand it before it touches anything live. A confident one-liner that deletes the wrong table is still your fault.
- Any decision with money or safety — the model can lay out options; the call is yours. This is exactly why our password and WiFi pieces push you to think, not just click.
The clean rule: AI can propose, you dispose. The moment you stop disposing and start forwarding, you're the assistant and the model is the boss — and it's a careless one.
The thing nobody mentions: it makes you lazy at thinking
This is the real risk, and it's quiet. Every time you offload a small judgment to the model, you skip the rep that builds the skill. Do it enough and you can't tell good output from bad, because you stopped exercising the muscle. A developer who lets AI write every function stops noticing the bug; a writer who lets it draft every sentence stops hearing their own voice.
The fix isn't to quit AI. It's to keep a slice of the hard part for yourself. Write the outline before you ask it to flesh one out. Solve the bug once before you paste the error in. Use it to go faster, not to go on autopilot. The people who thrive with these tools treat them as a bike, not a chauffeur.
Why the small tools still matter
An AI model is bad at the concrete, fiddly, local things — and that's most of real work. It won't reliably resize your image to spec, split that bill, convert your CSV to JSON, or read the QR code on a screenshot. Those are our tools, and they do one thing exactly. The model talks; the tool acts.
A good workflow stacks them: let the model draft the newsletter, then read it aloud to catch the sentence that's wrong; let it write the code, then reshape the data it needs; let it summarize the docs, then shrink the screenshot you're attaching. AI does the thinking-shaped parts; small tools do the doing-shaped parts. Pretending one replaces the other is how you end up with a beautiful paragraph and a file you can't send.
And on privacy — be careful what you paste into a chatbot. A contract, a medical note, a customer list: if you wouldn't email it to a stranger, don't feed it to a model whose logs you can't see. For the fiddly local jobs, tools that run in your browser and never upload your file (like the ones here) are the safer pick. It's not paranoia; it's just knowing where the data goes.
A short list to work by
- Use it to start, not to finish. The last edit is always yours.
- Verify anything it claims is a fact. Confidence is not correctness.
- Keep one hard task per day that you do the old way, so the muscle doesn't atrophy.
- Pair it with small tools for the concrete work it fumbles.
- Don't paste private data into a box whose logs you can't audit.
That's the whole trick: a fast assistant in your corner, judgment kept in your hands, and a drawer of plain tools for the jobs it can't do. Use all three and the AI age feels less like a threat and more like a better Tuesday. Go see the tools →