The Difference Between Using AI and Working With AI
Using an AI is asking a question, taking the answer, and moving on, treating Claude like a search engine that writes in sentences. Working with an AI is briefing it on your real situation, then building across a conversation until you have something you couldn't have produced this fast any other way. The gap between the two outputs is enormous, and it has nothing to do with how smart the model is.
Why am I getting such generic answers from AI?
Ask a question with no context, get an answer with no context. The response is calibrated to a generic version of whoever might type that question, not to your situation, your expertise, or the constraints you're actually working inside.
Take a commercial litigator who types "how should I structure a deposition for a hostile witness." She'll get a competent, textbook answer, and it'll be less useful than what she'd build herself, because the model knows nothing about her witness, her case theory, her jurisdiction, or the facts she has to establish. The question isn't wrong. The framing is. Asking AI questions is a thin slice of what the tool can do. Briefing AI on a situation and working through it with the model is a different activity entirely.
What does "briefing" AI actually mean?
Briefing is what you already do when you bring a sharp new colleague up to speed on a matter. You tell them the situation, what's happened so far, what you're trying to accomplish, the constraints, who the parties are and what they want, what you've already tried, and where you're unsure.
Do that with Claude before you ask, and the output shifts noticeably. The answer fits your real situation instead of a generic one. The follow-up questions sharpen. You can direct the work rather than accept or trash it. A corporate strategist who opens with "I've got a portfolio company in industrials deciding between an organic growth path and an acquisitive one, and I need help structuring the analysis" gets something far more usable than whoever asks "how do you evaluate organic versus acquisitive growth." Every detail in that brief, the sector, the portfolio context, the two defined paths, the strategic framing, shapes the response.
A simple rule for when to brief: the 30-Second Test
There's a simple test for this. Before asking Claude for anything substantive, ask yourself: "Could a sharp junior colleague get this right without knowing the client, the context, and the goal?" If the answer is no, you need to write a brief. That thirty-second filter separates quick queries from serious work.
The ongoing session is where the real work happens
The second difference is what you do across a conversation, not within a single exchange. Most people treat each AI answer as finished. They get a response, leave, and come back later with a fresh question in a brand-new chat. Every conversation restarts from zero.
Working with AI means building inside the conversation. You ask, you get an answer, you push back on the parts that miss, you demand more specificity, you add context, you redirect, you iterate. The session becomes an actual work session, and what comes out at the end looks nothing like the first reply. A tax attorney structuring a complex estate plan doesn't ask one question and stop. She lays out the situation, reads the first analysis, challenges the parts that miss nuances she knows from the client relationship, adds constraints, and keeps going until it's genuinely useful. That's not a technique you memorize. It's a shift in posture, from consuming output to producing work with a collaborator.
Your expertise is the input, not the AI's
Here's where the 45-to-62 reader has a real edge, and it's worth saying plainly: your decades of judgment are the moat, and they show up in the briefing. Generic requests get generic answers. Specific requests, the kind only a domain expert could even formulate, get responses calibrated to that expertise.
A wealth advisor who asks "how do I handle a client taking on too much risk" gets platitudes. The same advisor who writes "I've got a 72-year-old client, roughly 60% of his liquid net worth in stock from a company he sold fifteen years ago, resistant to diversification because the stock keeps winning, and I need to walk him through concentration risk without triggering a defensive reaction" gets something he can use Monday morning. The specificity in that second prompt isn't AI skill. It's wealth-management skill. Claude can't manufacture that brief. The professional brings it. That's the whole leverage point, and it's exactly the thing a 30-year-old with the same tool can't replicate.
I was wrong about prompt engineering
For a while I told people the answer was "prompt engineering": learn the syntax, the formatting tricks, the magic phrasings. After watching a lot of professionals use these tools for real client work, I've decided that's mostly a distraction, and I'll own that I sent a few people down the wrong path. The syntax barely matters. What matters is content: knowing what your situation actually contains that the model needs. You already have that. You don't need to learn to prompt. You need to stop treating Claude like a search bar.
Querying vs. working with AI, side by side
| Using AI (querying) | Working with AI (briefing) | |
|---|---|---|
| Input | One-line question, no context | Situation, constraints, goal, parties, what you've tried |
| Session | Fresh chat each time, starts from zero | One running session you build on and redirect |
| Your role | Consumer of output | Director of the work |
| What carries it | The model's general training | Your domain expertise, supplied up front |
| Output | Generic, needs heavy rework | Calibrated, close to client-ready |
| Best for | Quick facts, self-contained tasks | Analysis, judgment calls, high-stakes drafts |
How this looks in practice: three common patterns
An architect, seventeen years in. Used to ask Claude to "write a client proposal for a residential renovation." Now she briefs it on the client relationship, the scope, the client's communication style, the budget sensitivity, and what the proposal has to accomplish beyond describing the work. A proposal she used to wrestle with for two hours is now done in 45 minutes, including her final edits and polish.
A senior HR director, twenty-two years. Stopped asking "what are best practices for a performance conversation." Started writing "I have a direct report, eight years here, strong technical record, increasingly disengaged since a reorg three months ago, and I need a conversation that surfaces what's really going on without putting him on the defensive or triggering a formal HR process." What comes back is built for that situation, not a generic script.
An independent supply-chain consultant, nineteen years. Briefs Claude at the start of every engagement, client background, project context, what's been agreed, where the tensions sit, and uses the session as a thinking partner throughout. Her biggest surprise: she articulates her own thinking more clearly because she has to explain it to the model first.
Common questions on briefing vs. querying
Must every prompt be a long, detailed brief?
No. For self-contained tasks ("draft an email confirming our call" or "summarize this document"), short prompts work fine once you've established context in the chat. The detailed brief pays off for complex analysis where your judgment is required.
Isn't this just 'prompt engineering'?
No. Prompt engineering focuses on syntax and formatting tricks. Briefing is about content. You aren't learning new computer syntax; you're supplying the business context you already know inside and out.
Does this method only work with Claude?
The principle of supplying context improves results on any capable model. In practice this workflow is particularly suited to Claude, which excels at holding context over long, nuanced conversations.
What's the fastest way to get better results today?
Before your next interaction, write two or three sentences of context first. Who you are, what you're working on, what the output has to accomplish. That one habit changes most conversations on its own.
Use this on your next real task, not a toy one. Open one Claude chat, spend three minutes laying out the whole situation before you ask for anything, then stay in that chat and keep pushing until the output is good enough that you'd be comfortable sending it to a client after your edit. Do that once with a piece of work that matters, and the old search-bar habit will feel painfully thin.