I use AI assistants every working day, and I watch a lot of people use them too. The gap between the ones who get enormous value and the ones who get a clever toy almost never comes down to the model. It comes down to one habit. The people who get the most treat the assistant's attention as a budget they're spending, and the people who get the least pour that budget away on noise without ever noticing.
Let me explain what I mean, because once you see it you can't unsee it, and it's the single most useful thing I can tell a business owner about working with this technology.
The thing nobody tells you: AI has a working memory, and it's finite
An AI assistant holds your conversation, the documents you've shown it, and everything it's said back, all in one working memory. That memory is large, but it is not unlimited, and here's the part that matters: everything you put in it competes for the same space. Fill it with the wrong material and there's less room for the actual thinking, the work costs more, and on a long task the assistant gets vaguer exactly when you need it sharpest, because it has to start compressing what it's holding to make room.
So the goal isn't to feed it as much as possible. The instinct to dump everything in "just in case" is precisely the thing that makes it worse. The goal is to spend that working memory deliberately. The cheapest piece of information is the one you never had to load in the first place.
The habit, in five moves
None of this is technical. It's the same discipline as briefing a sharp new hire well: give them what they need, not everything you own.
1. Ask for the part, not the whole.
If you want help with one clause of a contract, point at the clause, not the whole contract. If it's one section of a report, paste the section. People routinely hand over a forty-page document to ask about a single paragraph, and then wonder why the answer is unfocused and the bill is higher. The assistant just spent most of its attention reading thirty-nine pages it didn't need. Narrow the ask and the quality goes up, not down.
2. Let it find before it fetches.
If you're working across a lot of material, ask the assistant to locate the relevant bit first, then look at just that bit, rather than swallowing the whole pile to go hunting. Search, then read the slice. It's the difference between asking someone to read an entire filing cabinet versus asking them to find the one folder and open it. Same answer, a fraction of the effort.
3. Park the bulk outside the conversation.
Big reference material, a long dataset, a sprawling export, doesn't need to live inside the chat for the whole session. Put it in a document and point the assistant at it when, and only when, it's needed. Carrying a large block of text through every single turn of a conversation is like rereading the same fat appendix before every sentence you write. Set it aside. Bring it back deliberately.
4. Ask for the answer, not the dump.
When you delegate a chunk of work, ask for the conclusion, the recommendation, the three numbers that matter, not a regurgitation of everything it looked at. This keeps the working memory clean for the next step and, frankly, makes the assistant more useful to a busy person. You wanted a decision, not a transcript.
5. Don't make it re-confirm what it already knows.
If the assistant already has a fact, or already worked something out earlier in the conversation, you don't need to send it back to check. Re-fetching what's already in front of you is pure waste, time, money and attention spent to arrive back where you started. Trust the work it's already done and build on it.
The counter-move: don't over-engineer the economy either
Here's the balancing truth, because I'd be giving you half the picture without it. You can take this too far. If you genuinely need the whole document, give it the whole document. Don't tie yourself in knots rationing a few sentences, and don't build an elaborate process to save a trivial amount of attention, that ceremony costs more than it saves. The discipline is for the cases that actually matter: the long tasks, the big files, the all-day working sessions. For a quick question, just ask the quick question. Correctness first, always. Economy serves the work; it doesn't replace good judgement about when it's worth the bother.
Why this is a leadership point, not a tech tip
I keep coming back to this with our own team and with clients, because the businesses getting real leverage from AI aren't the ones with secret access to a better model. Everyone has roughly the same tools now. The advantage is entirely in the habits, in knowing how to brief the thing, what to hand it, what to hold back, and when to stop. That's a skill, it's learnable, and it's exactly the kind of practical capability that separates a business quietly compounding an advantage from one that bought some licences and hoped.
The people who get the most out of AI aren't using a better tool. They're spending its attention like it's their own money, because in the end, it is.
Less is more, it turns out, even here. Especially here. If your team is using these tools and you suspect they're getting a fraction of what's possible, that's almost always true, and it's almost always fixable in an afternoon. Usually it's not a bigger model you need. It's the habit above, made deliberate.
