I Built a 9-Person Panel of TV Characters to Kill My Bad Ideas

I've built a lot of things nobody asked for.

Not because I'm reckless. I do market research. I talk to people. I read the subreddits. And then I build anyway, and six months later I'm staring at a product with zero paying customers wondering what I missed.

The problem wasn't research. It was that I was evaluating my own ideas, and I'm terrible at it. I'm too optimistic. I skip past the hard questions. I talk myself into demand that doesn't exist.

So I built a council.

What the Fellowship Is

The Fellowship is nine people who evaluate every idea I have before I touch a line of code or spend a dollar. They're fictional characters from shows I actually watch, each playing a role that matches their expertise on screen.

I run it as a Claude Code skill. When I have an idea, I type /council <idea> and they tear it apart.

Here's the roster:

Member	Show	Role	What they ask
Jack Donaghy	30 Rock	Marketing	Is anyone searching for this right now? What's the keyword volume?
Saul Goodman	Better Call Saul	Legal	Who sues you, and why? What's the liability angle nobody else sees?
Dwight Schrute	The Office	Sales	Who writes the check, how big, how fast? Name the buyer.
Gilfoyle	Silicon Valley	Engineering	Can you actually build this? What breaks at scale? What's the moat?
Stringer Bell	The Wire	Ops	What does Monday morning look like? How many hours a week does this take?
Marty Byrde	Ozark	Accounting	Where does the money actually go? What are the hidden costs?
Laurie Bream	Silicon Valley	Finance	What's the CAC/LTV? Path to break-even? Does this compound?
Dennis Reynolds	Always Sunny	Product	What do people actually want versus what they say they want?
Ben Wyatt	Parks and Rec	Growth	Show me the numbers. Retention curve? Referral loop?

The Rules

Each council member scores the idea from 1 to 10 on a single question: are people actively looking for this right now? Not in theory. Not someday. Right now.

They also have to name a specific channel where buyers exist. Not "LinkedIn." A specific LinkedIn group. Not "search." A specific search term with real volume. If they can't name it, that's a signal.

Average below 6: the idea is killed. Average 6 or above: it survives, and the council produces a three-step action plan for that Monday.

Members argue. They interrupt each other. Dwight doesn't care what Laurie thinks about the financial model until he knows someone will write a check. Gilfoyle will tell you your architecture is a catastrophe while Marty is still asking where the margin actually comes from.

No softening. No "it depends." Pick a position.

Why Fictional Characters

A few reasons this works better than I expected.

First, the characters have defined viewpoints I can't easily override. When I ask myself "what would a CFO say about this?" I unconsciously select the answer I want to hear. When the character is Laurie Bream, she answers in a specific way that I can't manipulate because I know exactly how she thinks. She models everything. She kills ideas that don't compound. That's fixed.

Second, they argue differently than I expect, which is the point. Saul's legal concern about a simple SaaS idea is almost never the thing I was worried about. Dennis's product insight about what users actually want versus what they say they want consistently surfaces something I missed. The friction is real.

Third, it's fast. A full council evaluation takes about two minutes to read. A traditional market research exercise takes days and produces a document that still tells me what I want to hear.

What It's Killed

Ideas I was excited about that didn't survive:

Listing on the QBO App Marketplace. $300/month to be listed. Sounds like a distribution channel until Dwight asks who writes the check and you realize the answer is "nobody, because you don't have paying customers yet." Killed before I paid the first month.

Cross-referencing SAM.gov fraud data for whistleblower opportunities. Looked promising on paper. Stringer asked what Monday morning looked like operationally. The answer was manual batch processing with no real signal at the end. Killed.

AI red-teaming bounties. I spent time on this. Six attempts, zero payouts. Ben asked for the conversion rate before I went further. The math didn't work. Closed.

Ideas that survived and I actually built: the compliance screening API now live on RapidAPI, the data products on the fraud analysis side, a few consulting packages that matched real search demand Jack identified.

The Underlying Pattern

Every idea I've killed would have taken 40 to 200 hours to build. At a minimum. The council takes two minutes.

That math is the whole argument. If the council is right even half the time, it's saved hundreds of hours this year.

The deeper thing is: I stopped trusting my own enthusiasm as a signal. Enthusiasm is not demand. The council is a filter that separates the two. When an idea survives the council, I feel differently about building it because nine distinct lenses looked at it and found real buyers, real channels, real unit economics that make sense.

When an idea doesn't survive, I feel relieved. Better to know now.

If You Want to Build One

The council is a Claude Code skill, which means it lives in a file at ~/.claude/skills/council/ and activates when I type /council from any project directory.

The mechanics are simple: YAML frontmatter that describes when to trigger the skill, a table defining each council member's role and lens, scoring rules, and an output format. The whole thing is under 60 lines.

The harder part is picking the right members. Each role needs to ask a genuinely different question that you personally tend to skip. Marketing and sales sound similar but they're not -- Jack wants search volume and buyer intent signals, Dwight wants to know who signs the check. Finance and accounting sound similar but Marty wants to understand cost structure and Laurie wants to know if the business compounds.

If two members would give you the same answer, replace one of them.

The fictional character framing matters more than I expected. Pick people from things you've actually watched. The more specific your mental model of how they think, the harder it is to put your own answer in their mouth.

The council is not a methodology. It's a habit. Every idea, every time, before a single hour of work. The cost of the habit is two minutes. The benefit is not building another thing nobody asked for.