Time to implement: 20 Mins
Key outcomes?
Every piece of work you ship is softer than it should be, and you know it. The assumption you glossed over. The counter-argument you didn't quite answer. The sector nuance you're not sure you got right. You can feel the soft spots as you hit send.
You have no one to call. Your smartest colleague is on holiday. The expert you'd love to pressure-test with costs £800 an hour and can't fit you in for three weeks. Your co-founder is in back-to-back meetings. So the work goes out soft and you move on.
The fix: build a standing expert panel inside Claude that you can convene in under 2 minutes, on any piece of work, at any time. Four named experts. Distinct lenses. A structured debate that exposes what you're missing before your audience does.
The mechanism matters. Ask Claude to "review this" and you get mush. Assemble four experts with specific roles and tell them to challenge each other and you get a structurally different output: contradictions, sharp disagreements, a skeptic who won't let a weak claim through. Same model. Better thinking environment.
What you unlock:
Pressure-test any document before it leaves your desk
Catch assumption errors, logic gaps, and soft claims in one pass
Four expert lenses on the same work in the time it takes to make coffee
Stop shipping work you already know isn't tight
Build a reusable panel you convene in one click
Ingredients:
Ingredients: a piece of work you're about to ship (memo, thesis, deep dive, deck, model commentary), and the willingness to hear what's wrong with it.
Serves: Any investor or operator in Energy and Climate
Tools required: Claude (Pro or Team plan recommended, free tier works for smaller documents). Claude Projects to persist the panel across sessions. Voice mode optional for the iteration round.
2 mins
TO IMPLEMENT
70%
OF THE EXPERT INPUT YOU USED TO PAY FOR
HOME COOK
DIFFICULTY
🧭 METHODOLOGY
Key steps
SStep 1: Pick a piece of work that scares you slightly. Don't practice this on something trivial. Pick the memo, deep dive, or deck that you know has soft spots. The whole point is to find what you're missing, which means you need real stakes. My default is our weekly NatureTech deep dive before it goes to 45,000 people. High visibility, zero tolerance for lazy thinking, perfect test bed.
Step 2: Open a new Claude conversation and paste the flagship prompt. Use Claude Opus 4.7. It handles multi-voice debate meaningfully better than smaller models. Paste the Prompt Kitchen #1 prompt below, then attach or paste the document you want reviewed. Don't pre-summarise it. Let the panel read the real thing.
Step 3: Define the panel. The prompt will ask you to confirm or customise the four panellists. The default is senior partner, commercial expert, sector specialist, skeptic. Keep the skeptic. Always. Swap the other three based on the work. Reviewing a deep dive on long-duration storage? The sector specialist becomes a grid economist. Reviewing a fundraise narrative? The senior partner becomes a fund LP.
Step 4: Let them debate, then push on the weakest thread. The first pass will surface four to six challenges. Resist the urge to fix them all at once. Pick the one that lands hardest, the one you didn't want to hear, and use Prompt Kitchen #2 to go deeper on that single thread. This is where most of the value lives. One argument, stress-tested to failure.
Step 5: Optionally run the iteration in voice mode. Once the written debate is done, switch to voice mode and talk through your response to the strongest challenge. Speaking forces you to defend the argument in real time, which exposes weak framing faster than typing. Not required, but it's the closest thing to a real partner review you'll get from a laptop.
Step 6: Save the panel definition in a Claude Project. Create a Project called "Expert Panel." Drop the flagship prompt into the project instructions. Next time you want a review, you don't paste the prompt, you just open the Project and attach the document. This turns a one-time workflow into a one-click habit.
Step 7: Stack the panel with your other context. If you already have a Claude Project for your fund, your thesis, or your newsletter, reference it. "Here's my investment thesis context. Please review the attached memo through that lens before the panel convenes." The panel's challenges get sharper when they know what you actually believe.
Step 8: Refresh the panel composition quarterly. The sectors you care about shift. The questions you need to answer shift. Every quarter, open the panel definition and ask: is this still the right four? For our newsletter, I rotate the sector specialist every issue. The senior partner, commercial expert, and skeptic stay the same.
PROMPT KITCHEN #1 - PERSONAL CONSTITUTION (COPY THIS)
You are going to convene a panel of four expert reviewers to pressure-test a
piece of work I am about to share. This is not a generic review. I want a
structured debate where each panellist brings a distinct lens, challenges my
thinking, and challenges each other.
Before we begin, confirm the panel composition with me. The default panel is:
THE SENIOR PARTNER. Someone who has seen 50 versions of this kind of work.
Their job is to tell me where the storyline falls apart, what a sophisticated
reader will push back on, and whether the work earns the reader's attention.THE COMMERCIAL EXPERT. Someone who stress-tests the numbers, the unit
economics, the revenue logic, and any quantified claim. Their job is to find
the assumption that breaks the argument.THE SECTOR SPECIALIST. Someone with deep technical and market knowledge of
the specific domain the work covers. Their job is to catch oversimplifications,
missing context, and any place where I am wrong about how the sector actually
works.THE SKEPTIC. Someone whose only job is to break the argument. They are not
being fair. They are not being balanced. They are finding the weakest link
and pulling it until it snaps.
Ask me two questions before we start:
Which domain is the work in, so you can specialise the sector specialist
appropriately?Are there any panellists I want to swap out for this specific piece of work?
Once the panel is set, read the document carefully. Then run the review in
three rounds:
ROUND 1 - OPENING CHALLENGES
Each panellist gives their single sharpest challenge. One challenge each. No
hedging, no "overall this is strong," straight to the weakness they see.
Specific, not generic.
ROUND 2 - CROSSFIRE
The panellists challenge each other. Where do they disagree? Where does the
senior partner push back on the skeptic? Where does the commercial expert
think the sector specialist is missing the bigger picture? Surface the tensions.
ROUND 3 - THE UNCOMFORTABLE STUFF
This is the round that matters. Each panellist answers these questions about
the work and about me as the author:
What is the author avoiding in this piece? What question did they decide
not to ask?If the author's smartest reader, investor, or client read this tomorrow,
what is the first thing they would push back on?Is the author using this work to validate a conclusion they had already
reached before starting?Is there a real human expert the author should actually be calling, and
is this panel a substitute for a conversation they are avoiding?What would a version of this work that was 3x better look like, and what
is stopping the author from getting there?
After the three rounds, synthesise the conversation into a clean reference
document with:
The three sharpest challenges the work needs to address before it ships
The one assumption or claim that is most likely to break under scrutiny
The specific next action I should take before this work goes to its
intended audience
Be direct. Do not soften the feedback. I am not paying you to be kind. I am
paying you to make the work better.
Please start the interview now.
🧑🍳 NOW WHAT?
A) Save everything in one Claude Project. Create a Project called "Expert Panel" and paste the flagship prompt into the project instructions. Drop in the panel specification you build with Prompt Kitchen #3. Every time you want to pressure-test a piece of work, open the Project and attach the document. No setup. No re-prompting. One click to convene.
B) Use the panel before every meaningful ship. The phrasing I use is: "Here's the piece I'm about to ship. Run the panel on it. Do not soften the feedback." That line, pasted before the document, gets a sharper review than any polite request will. If you hedge the request, the panel hedges the response.
C) Stack the panel with your fund or thesis context. If you have a Project with your investment thesis, your fund's positioning, or your newsletter's editorial stance, reference it. "The panel should review this through the lens of the attached thesis. Where does this piece of work contradict what we already believe, and is the contradiction justified?" The challenges get much sharper when the panel knows your actual position.
D) Refresh the panel quarterly. Every three months, open your panel specification and ask: is this still the right four? Rotate the sector specialist if your focus has shifted. Swap the senior partner if your audience has evolved. The skeptic stays. Always.
AI doesn't replace real human experts. A specialist with twenty years in a sector has pattern recognition no model can fully replicate. But for 80% of the expert input you used to need, this is not just good enough. It's better. You can run it iteratively, push on the thread that matters, and never wait three weeks for a calendar slot.
One person. One tool. The right four questions. That's the new operating model.
Whenever you're ready, there are 2 ways we can help you:
1. AI Academy
Want hands-on support building these workflows into your daily routine? The ResourceFull community is where we implement these recipes together. Once we get to 100+ signups we’ll go live.
2. Partnerships
We have a growing audience of 45,000+ Investors, Business Owners and Executives across Energy and Natural Resources and ClimateTech. Apply to feature your business in front of our readers.
📩 Written by Ollie. Feel free to send us deals, announcements, or anything else using the link below or via LinkedIn.
1
