Strategic Compaction: Direct AI Like a Pro
If you want Claude to produce work you'd actually send to a board or a client, stop treating it like Google and start treating it like a sharp associate. Strategic compaction is giving dense direction in few words, with your standards and constraints made explicit. The result isn't longer prompts. It's output that finally clears your bar.
Most accomplished people miss here in a predictable way. They tell Claude "write my Q3 strategy" or "make this email better," then judge the tool by the bland draft that comes back. Of course it's bland. You handed it almost no context, no stakes, and no standard. A smart human associate would have shrugged at that brief too. Claude is just being polite about it.
You already know how to do this. You've been doing it with people for decades. When you brief a good associate, you don't write three paragraphs of philosophy. You say: "Take the Henderson deal terms, compare them to the three comps in the shared folder, flag anything off-market, one page, I need it before the 2pm call." Dense context, clear standard, hard constraint, defined output. That's compaction. The same muscle that makes you good at delegating to people is exactly the one you need to direct Claude.
What does "compaction" actually mean in a prompt?
Compaction is packing maximum usable context into minimum words, then getting out of the way. It's the opposite of the bloated prompt templates floating around LinkedIn, the ones with eleven sections and a fake persona backstory. Claude doesn't need to cosplay as "a world-class strategy consultant with 30 years of experience." It needs to know what you know, what you want, and what's off-limits.
This matters more for a 50-something professional than it does for a new grad, and for a reason that surprises people. A 26-year-old often walks away satisfied, because anything semi-coherent looks impressive next to what they could write alone. You're asking Claude to do things you can already do at a high level. Your bar is higher and a lot more specific. Generic phrasing and safe recommendations fail you instantly. The only way to get output that genuinely helps is to load your experience into the brief and make your standard explicit.
Why do thin prompts always sound like everyone else?
Because the prompt is a shipping container and your context is the cargo. "Write my strategy" is an empty container. It travels fast and arrives carrying nothing. Claude fills that emptiness with the statistical average of everything ever written about strategy, which is exactly the mush you don't have time to sift through.
Your advantage is that you're sitting on a warehouse of cargo. Old memos. Client emails. Board decks. Deal notes. The awkward question a prospect raised last Tuesday. The way you phrase a tough "no" so the relationship survives. Thirty years of pattern recognition about what actually happens six months after the contract gets signed. None of that is in Claude's training data. All of it can go into your prompt or your Project.
So the move isn't to chase cleverer phrasing. It's to feed Claude your raw material and tell it precisely what to do with it. "Here are my last six discovery-call transcripts. Pull the five objections that come up most, and for each one draft a two-sentence response in my voice, which you can infer from how I answer in the transcripts." That request carries real cargo. What comes back is something a skeptical VP of Sales might actually use this week, not a warmed-over blog post.
The Brief, Bar, Bounds test
Before you hit Enter on anything that matters, run it through three checks. I call it the Brief, Bar, Bounds test. It takes about fifteen seconds and saves an hour of complaining about "AI fluff."
- Brief: Have you given Claude the actual material, not just a topic label? If your entire input is a noun or a vague verb, stop and add cargo. Paste the deck, the thread, the transcript, or at minimum a bullet outline of the facts.
- Bar: Have you named the standard the output must hit? "Board-ready." "Skeptical CFO is the reader." "This goes to outside counsel." "No adjectives I wouldn't say out loud." Claude can't guess your bar. You have to write it down.
- Bounds: Have you said what it must never do? "Don't invent figures." "If you're unsure, say so and ask." "Stick to issue-spotting, no legal conclusions." "One page." Sharp constraints are where most people leave value on the table.
When a prompt disappoints, it's almost always because one of those three is missing. Strong prompts often nail all three in under a hundred words. That ratio, dense direction in short form, is the whole game.
Vague prompt vs. compacted direction
The point isn't to write long prompts. It's to write loaded ones. The compacted versions below are barely longer than the vague ones. They just carry more of your judgment. Every clause is load-bearing.
| Vague prompt (empty container) | Compacted direction (loaded with cargo) |
|---|---|
| "Write a follow-up email to a prospect who went quiet." | "Prospect is a CFO at a mid-market manufacturer who went dark after our pricing call three weeks ago. Last objection was implementation risk. Draft a four-sentence re-engagement email in my voice (direct, no small talk, see the thread I'm pasting). Lead with one proof point, not an apology. Don't use 'just checking in.'" |
| "Summarize this contract." | "Turn this MSA into a one-page risk memo for a non-lawyer founder. Three buckets: deal-breakers, negotiate-these, standard-for-SaaS. Flag any clause that's off-market versus typical SaaS terms. Don't soften the deal-breakers." |
| "Help me prepare for my board meeting." | "Here's my board deck. Play the most skeptical director, a former operator who hates optimistic forecasts. Give me the eight hardest questions she'll ask and the slide numbers where my story is weakest. No reassurance. I want the uncomfortable version." |
Read the right column out loud. It sounds like an experienced partner handing work to someone they trust to execute. That tone alone lifts the output. The left column sounds like someone hoping a generic ask will produce specific work.
Build a standing Project, not a pile of one-off chats
The single highest-return setup is also the most boring. In Claude, create one Project for one domain you work in repeatedly: client reporting, hiring, sales emails, internal memos. Upload the reference material you're comfortable using. Then write a short standing instruction that defines the relationship once, so you never have to re-brief from scratch.
Something like: "You're my analyst for client deliverables. Turn rough inputs into clear, adult, non-hype memos. Preserve nuance. Strip out buzzwords. When a claim needs a source, flag it instead of inventing one. Default to one page unless I say otherwise." That paragraph does more work than fifty clever prompts, because every future request in that Project inherits it. You've hired the chief of staff and explained the job. Now you just hand over files and say "you know what to do."
I'll admit I resisted this for a while. I figured standing instructions were a gimmick and that I'd get better results steering each conversation by hand. I was wrong. The hand-steering tax adds up fast across a dozen sessions a week. A well-written Project instruction is the difference between managing an employee and re-interviewing them every single morning.
How is directing Claude different from directing a junior analyst?
Mostly it isn't, which is the point. But the few differences matter, and knowing them keeps you from over-trusting or under-using the tool.
| Junior analyst | Claude |
|---|---|
| Learns your standards slowly, over months, through feedback. | Learns them instantly if you write them down in a Project instruction; forgets nothing inside it. |
| Will tell you when a request is impossible or the data's missing. | Will cheerfully invent something plausible unless you put "flag it, don't guess" in the bounds. |
| Owns a slice of the judgment with you. | Owns none of it. The decision and the liability stay entirely yours. |
| Costs a salary; scales to one of them. | Costs about twenty dollars a month; scales to every task at once. |
That second row is the one that bites people. Claude won't push back the way a good associate does. So you build the pushback into the brief: tell it to surface uncertainty, name its assumptions, and refuse to fabricate. Direct the machine to do the thing a sharp human would do unprompted.
Is it too late to learn this if I'm in my fifties?
It's the opposite of too late, and it has nothing to do with being technical. Strategic compaction rewards the one thing that compounds with age: judgment. A 28-year-old can out-type you. They can't out-judge you. They don't know which objection actually kills a deal, what a board really fears, or which clause bites you in year three. That knowledge is the cargo, and you have warehouses of it. AI doesn't replace the moat. It widens it, because now your judgment moves at the speed of the machine.
Here's a pattern I see constantly. A 54-year-old fractional CFO I know was burning six or seven hours a week turning messy client financials into plain-English board summaries, work she found tedious and slightly beneath her. She built one Claude Project, uploaded a dozen of her best past summaries as the standard, and wrote a tight standing brief. Now she pastes the raw numbers, gets a draft in her own format, and spends forty minutes editing instead of six hours building. She didn't learn to code. She learned to delegate to a machine the same way she delegates to people. The skill was already in her hands.
What can't you compact away?
The judgment itself. Compaction moves the low-value cognitive labor off your desk, the formatting, the first-drafting, the synthesis, the "turn this pile into that shape" grunt work. It doesn't move the decision. You still decide whether the memo's "deal-breaker" really is one. You still own the relationship, the read on the room, the call.
That's why your edge gets sharper, not duller, as the tools improve. When AI can produce a competent first draft of almost anything in seconds, the scarce skill stops being production. It becomes direction and discernment: knowing what to ask for, and recognizing when the answer is quietly wrong. That's a fifty-year-old's home court.
One caution worth stating plainly. Don't load confidential client files, sensitive financials, health or legal material, or proprietary company data into any AI tool unless you understand its data settings and your own obligations. Direct sharply, but direct responsibly.
Start tomorrow with one thing. Pick the most tedious recurring task on your desk, the report you dread or the email you keep rewriting. Build one Project, upload three examples of your good past work as the bar, and write a six-line standing brief. Then hand over the next real one and edit, don't create. You'll feel the difference on the first pass. You were never supposed to type faster. You were supposed to direct better, and you've been training for that your whole career.