There’s a version of using AI that makes you sharper, and a version that makes you softer. Both look identical from the outside — same chat window, same box, same habit of opening a tab when you hit a problem.
The difference is which parts of the problem you’ve already worked through before you hit send.
What you’re actually outsourcing when you copy-paste too fast
A lot of productive AI use is essentially: take a thing I made, clean it up, expand it, reformat it. That’s fine. That’s editing work, and handing off editing work is how professionals have always operated — you write the brief, someone else polishes the copy.
But something different happens when you hand over the thinking work. Not the polish. The reasoning, the structure, the so-what. When you paste a document and ask “what should I take from this?” or paste a problem description and ask “what should I do?” — you’ve just skipped the most valuable part of the process.
Not because AI gives bad answers to those questions. Sometimes it gives genuinely useful ones. The problem is that you didn’t form your own view first, so you have no anchor for evaluating what comes back. You’ll tend to accept a reasonable-sounding answer not because you’ve weighed it but because it’s confident and complete and you have nothing to push against it with.
That’s the version that makes you softer over time.
Where the line actually sits
The most useful mental model I’ve landed on: use AI on outputs, not on inputs.
Outputs are things you’ve already produced — a draft, a list of options you’ve generated yourself, notes you’ve already processed. Hand those off freely. The AI is doing refinement work on top of something you made.
Inputs are raw material you haven’t processed yet — a document you just received, a question someone just asked you, a problem you’ve just encountered. If you hand those straight to AI without sitting with them first, you’re outsourcing the comprehension step. And comprehension is the thing that compounds. Every time you work through a hard document yourself, you build a small but real piece of domain knowledge. Skip that enough times and the gap shows.
This isn’t a moral argument. It’s a practical one. If you’re building expertise in an area — for freelance work, for a career, for genuine knowledge — the AI-as-first-pass habit quietly erodes the very thing that makes you valuable in that area.
What “thinking first” actually looks like in practice
It doesn’t mean spending an hour with every problem before touching AI. It means a much smaller thing: form a view before you ask for one.
Before you paste an article and ask for a summary, read it and write two sentences about what you think the key point is. Then paste it and compare. Before you describe a problem and ask what to do, jot down the options you’ve already considered. Then ask which you’re missing or whether you’ve framed it wrong.
This changes the interaction from “tell me what to think” to “here’s what I think — stress-test it.” The AI’s response is now something you can evaluate, not just absorb. You’ll catch errors much faster when you have a reference point to measure against.
I run into this directly with the automation pipelines I use Claude in: the human-in-the-loop value isn’t catching every error — it’s catching the errors that only matter if you already know what the output is supposed to achieve. An LLM running unattended will confidently produce something that looks fine and technically satisfies the prompt but drifts from the actual intent. You only catch that drift if you knew the intent going in.
The specific tasks where this matters most
Not all tasks are equal here. Some are genuinely fine to hand off. Others are exactly the kind where outsourcing the thinking creates real problems.
High-stakes thinking tasks — situations where your reasoning is the product: writing that represents your actual opinions, decisions where you’ll be accountable for the outcome, analysis you’ll have to defend. Use AI to pressure-test, not to originate.
Learning contexts — if you’re trying to get better at something, having AI do it for you is training yourself on the wrong thing. Use it to explain, to quiz you, to give feedback — not to produce the thing you should be struggling to produce.
Creative problems with no single right answer — AI is genuinely good at generating options you hadn’t considered, but it’s biased toward the statistically central answer. If you haven’t formed your own direction first, you’ll tend to accept “reasonable” rather than “right for this specific thing.” Your own instincts need to be engaged before its suggestions are actually useful to you.
Low-stakes formatting and production work — convert this to bullet points, write a subject line for this email, clean up these notes. Hand it off. Nothing is compounding at stake.
One caveat worth being honest about
This whole argument assumes you have time and motivation to engage with the thinking work yourself. Sometimes you don’t. Sometimes you’re under real time pressure and using AI as a first pass is just the right call. That’s fine. The issue isn’t any individual decision to take the shortcut — it’s when the shortcut becomes the default and you stop noticing you’re taking it.
The way I track it: occasionally do the task without AI first, then use AI to check. Not always. Just often enough to know your own baseline hasn’t moved.
If your unassisted attempt is noticeably weaker than it was six months ago on tasks you care about, that’s the signal worth paying attention to. If it’s held steady, you’re probably finding the right balance.
The real question isn’t whether to use the tool. It’s whether you’d still know what good looks like if it disappeared tomorrow.
