Agents as a team: brainstorm, expand, vet
⏱ 25 min5 steps
Remember the Thanksgiving kitchen from Agents, in plain English? This is the kitchen in action, on the hardest job there is: deciding what's worth your time.
I run my own ideas through three stages, with agents doing most of the legwork: , expand and .
It exists to stop two mistakes: thinking too narrow, and chasing an idea for months without ever deciding. If the first round looks thin, send a helper back for another go.
brainstorm Opening an idea up wide before judging it. Here, it starts with questions, not answers. All wordsvet Check an idea quickly and fairly, the same way every time, to decide whether it's worth more of your time. All wordsBefore this, you'll want:
Steps
Make a folder for the idea and go into it
Each idea gets its own in your . This one is called next-idea.
-
Open (Cmd+SpaceWindows key, type Terminal, press Enter).
-
Make the folder. Type this and press Enter:
mkdir ~/Backyard/next-ideamkdir ~\Backyard\next-idea -
Go into it. Type this and press Enter:
cd ~/Backyard/next-ideacd ~\Backyard\next-idea
You should see:
The now says next-idea, right before the %>.
prompt line The line in Terminal that ends with a % sign. The word right before the % is the folder you're in. All wordsThat's not what I see
- It says "cd: no such file or directoryCannot find path … because it does not exist".
- The folder wasn't made, or the name was spelled differently. Type ~/Backyard/next-idea~\Backyard\next-idea and press Enter, then ~/Backyard/next-idea~\Backyard\next-idea and press Enter.mkdir The command that makes a new folder. Type mkdir, a space and the new folder's name. It's short for make directory. All wordscd The command that moves Terminal into a folder. Type cd, a space and the folder's name. It's short for change directory, an old word for folder. All words
-
Start Claude Code and let it into the folder
-
Type this and press Enter:
claude -
If shows "Quick safety check", check that the folder under Accessing workspace ends in next-idea. It shows this the first time in a new folder.
-
Press the down arrow to Yes, I trust this folder, then press Enter.
You should see:
Claude Code's name and version at the top, a line starting with ❯ where you type, and ⏵⏵ on in the bar at the bottom.
auto mode Claude works without asking, and a separate safety checker blocks anything risky. All wordsThat's not what I see
- It says "zsh: command not found: claudeThe term 'claude' is not recognized".
- This Terminal window doesn't know about Claude Code. Close the window with Cmd+WCtrl+Shift+W, open a new one and try again. If it still says that, go back to Claude Code and .install Put a new program on your computer so you can use it. All wordssign in Tell Claude Code which Claude account is yours, so it knows you've paid for it. All words
- The folder it asked me to trust doesn't end in next-idea.
- Choose No, exit. Type cd ~/Backyard/next-idea~\Backyard\next-idea, press Enter, check the prompt line says next-idea, then type claude again.
-
Brainstorm: let Claude ask you questions first
Claude asks you questions until it understands what you're actually after. No ideas yet. This sets that up. Your idea goes at the end, as rough as it is.
- Copy the prompt below.
I have a seed of an idea, at the end of this message. Don't suggest anything yet. Ask me questions, one at a time, until you understand what I'm really after, who it's for and what I already have to work with. Then play back what you heard in a short paragraph and ask what's wrong with it. Here's the seed, as rough as it is:
- Paste it into Claude Code with Cmd+VCtrl+V.
- Type a space, then your idea, like something for people who look after an aging parent from far away.
- Press Enter.
- Answer the questions. Then correct the playback until it's right.
You should see:
Claude asks you one question about your idea, and waits for your answer.
That's not what I see
- Claude jumped straight to ideas.
- Type: Stop. Questions first, one at a time. It will.
Expand: four helpers, four different ways of looking
Four helpers each look at the idea a different way, and come back with concrete options. Each one names who pays, how much, and the one that would kill it.
- Copy the prompt below.
Now widen this into options. Start four helpers, each looking a different way: 1. boring but better: unglamorous work people already pay for, done cheaper or faster; 2. my edge: what I can do that a stranger couldn't, from what I've told you; 3. wild: a mechanism borrowed from a completely different industry; 4. coming soon: what's about to change in the next two years that opens a door. Each helper gives me five options. For each option: who pays, for what, roughly how much, and the one assumption that would kill it. Label anything you looked up with its source. Then put all twenty in one list. Don't judge them yet.
- Paste it into Claude Code with Cmd+VCtrl+V and press Enter.
You should see:
Lines with while the helpers work, often at the same time. Then one list of twenty options, each with who pays, how much and what would kill it.
agent Claude given a goal and some tools. It plans, does a step, checks the result and keeps going until the job is done. All wordsThat's not what I see
- The four lists look the same.
- Tell Claude which two feel too alike, and ask one of those helpers to try again, further from the others. The point of four lenses is four different directions.
Vet: test the cheapest few, the same way every time
The cheapest few get tested the same way every time, quickly. Each comes back as STOP, with the evidence, or as the one next check worth doing.
- Copy the prompt below.
Pick the three options that are cheapest to test. For each one, in under 10 minutes of looking: state the assumption that would kill it; find the best evidence for and against it, with sources; and give me one of two answers. STOP, with the evidence that killed it. Or NEXT CHECK, with the single cheapest thing I could do this week to find out more, and what result would make me stop. Don't recommend building anything.
- Paste it into Claude Code with Cmd+VCtrl+V and press Enter.
A STOP is a win: it's months you didn't spend. The agents never decide what you pursue. You do.
Then go back to Check the idea first and do the next check with real people. Agents can find the evidence. Only people can tell you whether they'd pay.
You should see:
Three options, each ending in STOP with the evidence, or NEXT CHECK with one thing to do this week and the result that would make you stop.
That's not what I see
- Every option came back as NEXT CHECK.
- Ask: If you had to stop one of these today, which, and why? A vet that never says stop isn't doing its job.
Your challenge
Run one idea all the way through: brainstorm, expand, vet. Then do the one next check the vet recommends, or stop if it says stop.
Last checked 2026-10-20
Connect and automate › 3. Agents
Agents as a team: brainstorm, expand, vet
One-page card · print it or keep it open
What you'll do
Remember the Thanksgiving kitchen from Agents, in plain English? This is the kitchen in action, on the hardest job there is: deciding what's worth your time.
The steps
- Make a folder for the idea and go into it
- Start Claude Code and let it into the folder
- Brainstorm: let Claude ask you questions first
I have a seed of an idea, at the end of this message. Don't suggest anything yet. Ask me questions, one at a time, until you understand what I'm really after, who it's for and what I already have to work with. Then play back what you heard in a short paragraph and ask what's wrong with it. Here's the seed, as rough as it is:
- Expand: four helpers, four different ways of looking
Now widen this into options. Start four helpers, each looking a different way: 1. boring but better: unglamorous work people already pay for, done cheaper or faster; 2. my edge: what I can do that a stranger couldn't, from what I've told you; 3. wild: a mechanism borrowed from a completely different industry; 4. coming soon: what's about to change in the next two years that opens a door. Each helper gives me five options. For each option: who pays, for what, roughly how much, and the one assumption that would kill it. Label anything you looked up with its source. Then put all twenty in one list. Don't judge them yet.
- Vet: test the cheapest few, the same way every time
Pick the three options that are cheapest to test. For each one, in under 10 minutes of looking: state the assumption that would kill it; find the best evidence for and against it, with sources; and give me one of two answers. STOP, with the evidence that killed it. Or NEXT CHECK, with the single cheapest thing I could do this week to find out more, and what result would make me stop. Don't recommend building anything.
If it goes wrong
- It says "cd: no such file or directoryCannot find path … because it does not exist". The folder wasn't made, or the name was spelled differently. Type mkdir ~/Backyard/next-idea~\Backyard\next-idea and press Enter, then cd ~/Backyard/next-idea~\Backyard\next-idea and press Enter.
- It says "zsh: command not found: claudeThe term 'claude' is not recognized". This Terminal window doesn't know about Claude Code. Close the window with Cmd+WCtrl+Shift+W, open a new one and try again. If it still says that, go back to Install Claude Code and sign in.
- Claude jumped straight to ideas. Type: Stop. Questions first, one at a time. It will.
- The four lists look the same. Tell Claude which two feel too alike, and ask one of those helpers to try again, further from the others. The point of four lenses is four different directions.
Words
- brainstorm: Opening an idea up wide before judging it. Here, it starts with questions, not answers.
- lens: A way of looking at something. Four lenses means four different starting points for ideas.
- vet: Check an idea quickly and fairly, the same way every time, to decide whether it's worth more of your time.