MiroFish: Using Claude as the Opposition I Don't Have
(The full prompt is at the bottom, if you just want to grab it.)
Lately I've seen a lot of people in the community using Claude to work through real-life problems, so I figured I'd share mine: #MiroFish, the thing I built to talk to whenever I have to make a big decision on my own.
I first heard the idea on Everyday with Captain's channel — the same video that got me building my KOS (Knowledge Operating System). I took it and ran.
The problem with being the thinker and the decider at the same time
I don't really have people around me who are free to sit and listen to me ramble all the time. Take wanting to try building a startup for once — the people close to me don't quite know how to advise on it either, because they haven't stood where I'm standing.
And I'm a nerd. When I'm unsure about something, I want every scenario on the table, not just the opinion of someone who happens to be biased in the same direction I already lean. It's not like I have so many friends I can call about everything, either.
Right now I'm running a startup by myself. In a single day I might switch between product, marketing, customer experience, the business model, and the technical parts I'm honestly not good at. I come up with an idea, research it, approve my own idea, done. Nobody pushes back, because I'm the only chairperson in the room for this session.
But the thing I think is good — is it actually good?
So I wanted an opposition. A loyal one.
When I came across this idea, I built MiroFish — my little fish — to play that person.
🐟🐟🐟
What MiroFish actually is
To me, it's a thinking partner. An advisor that won't make the decision for me.
It won't tell me my idea is "good" or "bad."
Its job is to:
Help me see the situation more clearly and open up more options.
ช่วยฟันธงว่าจริงๆ แล้วเรากำลังตัดสินใจกับปัญหาเรื่องอะไรกันแน่ บางทีเราคิดว่ามันเป็น A แต่จริงๆ ปัญหาคือ B อะไรแบบนี้ มันเอาความคลุมเครือออก
ขุดหาสมมติฐานที่เรายังไม่เคยพิสูจน์ บางทีตัวเราที่อยู่ในสถานการณ์นั้นๆ มี personal bias หรืออคติส่วนตัว และเราก็ตั้งสมมติฐานอะไรไว้ด้วยตัวเอง ทั้งที่ยังไม่มีหลักฐาน
ขยายทางเลือกที่เรามองไม่เห็น และเปิด third eye view ทำให้เรามองตัวเลือกได้เยอะขึ้น
ช่วยเราหาวิธีทดสอบที่ไม่เปลืองแรงมาก หาวิธีแก้ที่เล็กที่สุดที่พอจะทดสอบได้ก่อนทุ่มทรัพยากรลงไปเต็มๆ
Help me see the situation more clearly and open up more options.
Pin down what I'm actually deciding about. Sometimes I think the problem is A, when the real problem is B. It takes the fuzziness out.
Dig up the assumptions I've never tested. When I'm inside a situation, I carry my own biases, and I set up assumptions on my own without any evidence behind them.
Widen the options I can't see — a third-eye view that puts more choices in front of me.
Find the smallest way to test something before I pour resources into it.
I use MiroFish for decisions about business, money, and life direction, but the same principle works anywhere, as long as I'm still the one who owns the decision.
Use Case
I'd used plain chat for this before — on both Claude and GPT — but those tend to send me spiralling wider than I started. MiroFish answers in a template, broken into a comparison table, and holds focus much better. The spiralling it leaves to me.
A real example:
When I launched FKE (a French-teaching project I built myself), about 500-something people came through, but only 8 signed up. My first thought was:
“The conversion flow must be broken.”
I nearly rushed in to redo the sign-up page and change the whole flow. (I almost did, too.)
When I took it to the fish, it didn't rush to an answer. It asked back: this number only proves you have a conversionproblem — it doesn't prove the sign-up flow is the cause. Then it had me break down how many other possible causes there were, ranked by likelihood (with percentages next to each).
Just changing the question widened things from "the sign-up page and the steps" out to target audience, value proposition, motivation, timing, and friction somewhere else in the product entirely.
And when I actually went down the list, one by one, the answer was:
the CTA button wasn't visible enough, so people didn't click.
That was it. That was the whole thing. And it wasn't something I'd have thought of at the start.
Once I fixed it, more people clicked.
If I'd trusted my own bias from the start and torn the entire sign-up flow apart, I'd have burned time on a huge job — and probably on the wrong problem altogether.
The thinking behind MiroFish is turning into two more projects. What I care about isn't building a tool that gives people more information — it's designing a system that helps me see what matters and take the next step more easily. If this way of thinking clicks, it's the same one I'm turning into two projects right now. Take a look when you're done reading.
What to know before you use Claude as an advisor
Even when I tell Claude to look at every angle, a wrong question still gives a wrong answer.
Claude thinks from the information and the question you hand it. If your information is incomplete, or you start from a wrong understanding, the answer can come out wrong too — even when it sounds perfectly reasonable.
Before you act on any advice, check:
which parts are things you actually know
which parts are guesses
which parts have no evidence yet
what would change your mind
whether you can test this in a small way first
For high-stakes things — legal, medical, or financial decisions with real risk — talk to an expert as well.
MiroFish is a thinking aid, not the one who's accountable for how it turns out.
The starter prompt
The prompt below is a starter version you can use with Claude directly.
Your job is to fill in 5 things:
the decision you're facing
the evidence you have
the real constraints (time, budget, people — put it all in)
the assumptions you're holding
your deadline
Then it'll look at the decision through reversibility (can you walk it back if you're wrong), downside risk, how hard it actually is to execute, and what information is still missing.
This isn't my whole internal system. But it's the core — enough for a solo operator to start testing the same way of thinking, without writing any code or installing anything extra.
Copy the prompt and swap the bracketed parts for your own situation.
MiroFish Starter Prompt
You are my decision-review partner.
Your job is not to make the decision for me, agree with my current opinion, or create artificial disagreement. Your job is to help me examine the decision more clearly, uncover hidden assumptions, expand the available options, and identify the smallest useful next step.
My situation
Decision I am facing:
[Describe the decision in one sentence.]
Why this decision matters now:
[Explain the immediate context.]
My current options:
[List the options you are already considering.]
My current lean:
[State which option you currently prefer and why.]
Constraints:
[Time, money, energy, skills, commitments, or other limits.]
Evidence I currently have:
[List facts, observations, user feedback, or numbers.]
What I may be assuming:
[List anything you suspect might be an untested assumption.]
Decision deadline:
[Add a date or write “no fixed deadline.”]
Reversibility:
[Explain what can and cannot be undone.]
How to analyze the decision
Restate the actual decision in neutral language. Separate the underlying problem from the most visible symptom.
Before analyzing, identify important missing information. If necessary, ask me no more than five targeted questions.
Do not accept my current framing automatically. Assume that my preferred explanation may be wrong and present the strongest reasonable alternatives.
Generate materially different options. Do not present minor variations of the same idea as separate choices. Include delaying the decision or taking no action when those are genuinely viable options.
Examine each option through these lenses:
available evidence
execution difficulty
downside risk
opportunity cost
reversibility
effect on other people
what must be true for the option to work
Make the strongest case against my current lean. Do not disagree merely to appear critical; explain what evidence would support or weaken the objection.
Clearly label:
Known facts
Reasonable inferences
Important unknowns
Assumptions requiring validation
Do not invent certainty scores or pretend that incomplete information is precise.
Recommend either:
a decision, when the evidence is sufficient, or
the smallest experiment that would reduce the most important uncertainty.
State the flip conditions: what new evidence would cause the recommendation to change?
Output format
The real decision
[Neutral restatement]
What may be framed incorrectly
[Hidden assumptions or symptom-versus-problem issue]
Options worth considering
[Materially different options]
Strongest argument against my current lean
[Best counterargument]
Facts, inferences, and unknowns
[Clearly separated]
Recommendation
[Decision or smallest useful experiment]
Flip conditions
[Evidence that would change the recommendation]
Next action
[One concrete action I can take now]
Do not flatter me or reassure me automatically. Be direct when my reasoning is weak, but do not perform contrarianism for its own sake. The goal is a better decision, not a more impressive analysis.
How to get more useful answers out of it
1. Don't start from what you want Claude to do.
"Help me design a new page" or "help me lay out a marketing plan" is giving an order — it's not reviewing a decision. Start from the decision instead:
Should I rush to fix the sign-up flow now, or check other causes first?
That gives Claude room to think with me, instead of being told to execute a solution I've already picked.
2. Give real evidence, not just feelings.
Instead of "the people coming in probably don't understand how to use it," put in what you actually observed:
When FKE launched, about 500-something people came through, but only 8 signed up — and I haven't checked yet whether the people coming in are actually the target user.
That detail lets Claude separate what's evidence from what's still an assumption.
3. State your constraints honestly.
If you've only got two weeks to test, a tight budget, or you're working alone, say so plainly. Advice that's good but impossible to actually do isn't worth much.
4. Don't put in personal or confidential information you don't need to.
Strip out real names, customer data, passwords, anything financial that identifies a person, and internal details that don't matter to the decision — every time.
5. Use the answer as a starting point, not a final verdict.
Claude can widen the frame and point at things you'd have missed. But the person who has to check the evidence, accept the risk, and live with the outcome is still you.
After building MiroFish, I don't come away feeling like I have more answers every time. Plenty of times I end up with more questions instead. But it opens up angles I genuinely wouldn't have reached, in the moments when I have to think alone.
For a solo operator, maybe what's missing isn't one more answer.
Maybe it's someone — or something — that keeps the first idea in your head from hardening into biased confidence too easily.
If you try this prompt and get an angle you'd never thought of, come tell me. I'd love to know what everyone's own version of MiroFish ends up helping them decide.
May you all have a good opposition of your own. And once you've tried it, don't forget to come swap notes.
🐟🐟🐟
If MiroFish helped you make a hard call more calmly, that's the whole idea behind what I'm building next. Wealth Tracker brings it to money decisions; FKE brings it to learning French. Pick the one you'd like to follow.
Comments