The Five Rules
This is the foundation. Five short rules that change how AI behaves for you — and a sixth that explains how you actually get good. You can skim them in eight minutes today and come back. Don’t skip Rule 6.
Stop asking it questions. Start giving it jobs.
When most people open their AI, they type a question — “What are the key risks in a leveraged buyout?” — and get back a Wikipedia article. Then they decide AI is useless.
Instead of asking a question, give it a job:
“I’m reviewing a sponsor-backed LBO with 5.5x total leverage. The add-backs represent 30% of adjusted EBITDA. Write me the three strongest pushback arguments I should bring to the credit committee about why these add-backs are aggressive.”
Same topic. Completely different output. The first gives you a textbook answer. The second gives you something you can use in a meeting.
| Stop doing this | Start doing this |
|---|---|
| “What is covenant analysis?” | “Here’s the covenant package. Tell me which terms are borrower-friendly and why I should push back.” |
| “How do I analyze a CIM?” | “I’m uploading this CIM. Extract the top 5 risks to credit quality based on my firm’s criteria.” |
| “Explain EBITDA adjustments.” | “Here are the proposed adjustments. Tell me which ones won’t survive diligence and why.” |
| “What’s a good client email?” | “Draft a follow-up to a CFO who went silent after our second meeting. Direct but not aggressive. Goal: a meeting this week.” |
Google answers questions. AI does work. Treat it like Google and you get Google-quality answers — generic and surface-level. Treat it like a junior analyst sitting next to you and you get drafts, analysis, pushback, and output you can build on.
Right now, your AI knows nothing about you.
Open your AI right now. It has no idea what your job is, what industry you’re in, how senior you are, what your firm’s standards look like, or what kind of work you produce. It’s like hiring someone and never telling them what company they work for. Of course the output is generic.
The fix: every AI tool lets you store background about yourself that runs silently behind every conversation. Set it up once, and every output improves automatically. The setup guide walks you through it for Copilot, ChatGPT, and Claude with ready-to-paste text.
Before vs. after
“Help me with this deal memo.”
“Sure! Here’s a general template for a deal memorandum. A deal memo typically includes an executive summary, company overview, financial analysis…”
“Help me with this deal memo.”
“Based on your firm’s IC format, I’ll structure this as: (1) Investment Thesis with the key return drivers, (2) Business Quality, (3) Key Risks with mitigants, (4) Preliminary Terms. Want me to start with the thesis, or do you want to upload the CIM first?”
Same question. Wildly different output. The only thing that changed is that the AI knows who it’s talking to.
Talk to it like you’re briefing a new analyst.
People Google “best AI prompts” and copy-paste them like magic spells: “Act as a world-class financial analyst with 20 years at Goldman Sachs…” This is useless. The AI doesn’t get smarter because you told it to pretend to be someone impressive.
Talk to it the way you’d brief a sharp new analyst on day one. You wouldn’t hand a new hire a magic phrase and walk away. You’d say: “Here’s what we’re working on. Here’s what I need. Here’s what matters. Here’s what to watch out for. Go.” That’s the entire skill.
| Step | What it means | Example |
|---|---|---|
| What I have | Give it the raw material | “B2B software co. $45M revenue, 85% gross margins, growing 12% YoY.” |
| What I need | Tell it exactly what to produce | “A one-page risk summary for my credit committee.” |
| What to avoid | Set guardrails | “No generic risks. Focus on what threatens debt service coverage.” |
| What to fix | After the draft, redirect | “Too academic. Make it presentable. Add customer concentration.” |
“Act as a senior credit analyst and analyze this company for me.”
“I’m reviewing a mid-market healthcare services company, PE-backed, 4x levered. I’m uploading the CIM. Pull out the three biggest risks to the thesis and explain why the sponsor’s EBITDA adjustments might not hold up in diligence. Write it the way I’d present to my IC — direct, no hedging, specific to this deal.”
The second one works because you told it what you have, what you need, what to watch for, and how to write it. That’s not prompt engineering. That’s clear communication.
Your first answer is a first draft, not a final product.
Most people type something in, get a response, think “okay but not great,” close the window, and do it themselves. That’s like asking an analyst to write something, reading their first draft, and firing them instead of giving feedback.
The first response is a starting point. The real skill is knowing what to say next. After every answer you have three moves:
| Move | When | What to say |
|---|---|---|
| Sharpen | Right direction, too vague | “Too high-level. Give me specific numbers and examples.” |
| Redirect | Wrong direction | “You’re focused on revenue growth. I care about cash-flow sustainability. Start over with that lens.” |
| Push deeper | Surface is fine, need more | “Good start. Now stress-test point three — what if their largest customer is 40% of revenue and doesn’t renew?” |
Three rounds, four minutes
“Review this covenant package and tell me what’s weak.”
Gives you a general list of covenant concerns.
“The leverage covenant looks loose. Walk me through exactly what happens if EBITDA drops 15% — do they trip it, or does the calculation method let them hide it?”
Now gives you a specific scenario analysis with the actual math.
“Good. Now write that as a two-paragraph risk flag I can drop into my IC memo.”
Gives you a polished, copy-paste-ready paragraph.
The people who get the most out of AI aren’t better at the first message. They’re better at the second and third. They know how to steer. Most people quit after Round 1.
Make it fight you.
By default, AI agrees with everything you say. You say “I think this is a strong credit,” it says “Great analysis! Here are more reasons it’s strong.” That’s not helpful. That’s dangerous — you just got your own thesis repeated back with a bow on it.
In private credit, the whole game is downside protection. You already know the bull case; the sponsor told you the bull case. What you need is someone to find the holes before your IC does. AI does this extremely well — but only if you tell it to.
“When I share analysis or a thesis, your job is to find the weaknesses — not to validate me. If my logic has gaps, say so directly. If my assumptions are aggressive, flag them. I’d rather hear the problems from you now than from my committee later.” Once it’s in your settings, it runs automatically.
“I think this company’s 20% EBITDA margins are sustainable.”
“That’s a reasonable view. The company has demonstrated consistent margins over the past three years.”
“I think this company’s 20% EBITDA margins are sustainable.”
“Those margins are supported by a single long-term contract that reprices in 18 months. If you lose even 200bps on renewal, debt service coverage drops below 1.2x. Are you sizing the facility to the current margin or to a downside scenario?”
How you actually get good at this.
There’s no course, no certification, no YouTube playlist that teaches this. The people who are best at using AI all have the same story: they just started using it. Not by studying it — by doing things with it.
The gap between “I have Copilot” and “I’m producing better work twice as fast” isn’t knowledge. It’s reps. You need the instinct for how to give context, how to steer, and when to push back. The fastest way to build it is to use AI as a learning accelerator in your day-to-day life — not to replace your thinking, but to expand it.
Get in the habit
- Planning a trip? Plan it together — better itinerary, less browsing.
- Comparing two products? Have it break down the tradeoffs.
- Something wrong with your car? Describe the symptoms and get a real explanation before the shop.
- Trying to understand a financial concept, a tax rule, a negotiation tactic? Go deeper than a search ever could.
Why it works
Every one of these builds the exact muscles you’ll use at work: giving context, being specific, iterating, pushing back. And you learn something real in the process. How many times a day do you hit something you’re curious about and just move on? Pull out AI in those moments instead. That’s the practice.
The voice trick
This is the single most underrated way to use AI, and almost nobody talks about it. Put your headphones in. Go for a walk. Talk.
Most AI tools have a voice mode — or just use your phone’s voice-to-text and paste it in. When you talk out loud, you think differently than when you type. You’re less filtered. You say what you actually mean instead of trying to write the “perfect prompt.”
- Stuck on a problem at work? Walk fifteen minutes and explain it out loud. Half the time you’ll solve it yourself while talking. The other half, the AI catches what you missed.
- Have a vague idea you can’t articulate? Talk it through, then say: “I just rambled for five minutes. Organize what I said into a clear structure.”
- Staring at a blank screen? Talk through what you want to say, then ask it to clean it up.
Voice-to-text doesn’t produce clean input. That’s the point. Raw thinking goes in, structured output comes out. The AI is very good at that translation.
What getting good actually looks like
| Week 1 | Basic questions, generic answers. You think it’s overhyped. |
| Week 3 | You give it more context, answers get noticeably better. You catch yourself thinking “I should ask the AI about this” all day. |
| Month 2 | You stop thinking about “prompts.” You just talk to it, and you know exactly what to say to fix a weak answer. |
| Month 4 | Colleagues notice your work is faster and better. They ask what you’re doing differently. You can’t fully explain it — it’s become instinct. |
That progression doesn’t come from reading about AI. It comes from making it part of how you think and learn every day.
Pick one thing you’re genuinely curious about — work or not — and have a real back-and-forth with AI about it right now. Push back on its first answer. Ask it to go deeper. That five-minute habit, repeated, is the whole game.