ALG
The Hypergrowth Formula That Transformed Tesla, Lululemon, General Motors, and SpaceX
TheAlgorithm
Jon McNeill
2026
The
ALGORITHM ALGORITHM
//
For AI
Question
Delete
Simplify
Accelerate
Automate

Works in Claude Code, Claude Desktop,
ChatGPT, Vertex, your local AI — anywhere
you can drop a markdown file.

github.com/jordanjoecooper/the-algorithm
Scroll
A mental model.
Encoded
as a skill.

Jon McNeill's Algorithm is a five-step framework for building things that shouldn't exist — the version that actually works. He's used it at Tesla, Lululemon, SpaceX, GM. He's now using it at DVx, the venture firm he founded.

Most AI tools make you faster at what you're already doing. This makes you stop and ask whether you should be doing it at all.

A skill is a markdown file — a set of instructions that gives an AI model a way of thinking, not just a task to complete. Drop it into any tool that accepts system prompts or project instructions and it changes how the model behaves.

The difference between a prompt and a skill is the difference between asking someone to do something and teaching them how to think about it.

The more context you give it — your documents, your conversation history, your notes — the harder it pushes back.

1
Question Every Requirement
Every requirement must have an owner and a reason. If you can't say who required it and why, it's suspect. Categories: Validated, Assumed, Inherited, Self-imposed. Most fall in the last three.
2
Delete Every Possible Step
Argue for removing everything. If you're not adding back 10% of what you deleted, you didn't delete enough. The most valuable cuts are the ones nobody thought were cuttable.
3
Simplify and Optimise
But not before steps 1 and 2. The most common mistake smart people make is optimising something that shouldn't exist. Simplification only works on things that have earned their place.
4
Accelerate Cycle Time
Find every wait, handoff, approval, and batch. But not before step 3. Don't speed up a bad process. The goal is shorter feedback loops on the right work.
5
Automate
But not before step 4. Don't automate waste. Only automate what has survived all four prior steps. Automating too early locks in your mistakes at scale.

The skill is a plain markdown file. It works anywhere you can provide system instructions or a project context — which is most modern AI tools.

Claude.ai
Add to a Project's instructions. Every conversation in that project starts with The Algorithm loaded. Use /question, /delete, /algorithm slash commands.
Claude Desktop
Add via MCP or project context. Connects to your actual files, Notion docs, Linear tickets — the skill runs against real context, not a document you pasted in.
Claude Code
Drop the markdown into your CLAUDE.md. The skill becomes part of every session in that repo — particularly useful for questioning requirements before building.
ChatGPT
Paste into a Custom GPT system prompt, or seed it at the start of any conversation. Same framework, different model — useful for comparing how each one pushes back.
Vertex AI / API
Pass as a system prompt in any API call. Works with Gemini, Claude on Vertex, or any model that accepts system instructions. Drop into your existing pipelines.
Local AI
Ollama, LM Studio, anything running locally — add as a system prompt. Useful for zero-cost iteration. The framework works regardless of model size.
1
Get the skill file
Clone or download from GitHub. The skill lives in the-algorithm.md. That's the file you'll drop into tools.
git clone https://github.com/jordanjoecooper/the-algorithm
2
Load it into your tool
Paste the contents into your system prompt, project instructions, or CLAUDE.md. The model will recognise the slash commands and framework immediately.
3
Give it something real to work on
Upload a PRD, a process doc, a spec, a plan. The more context the better — the model needs something concrete to challenge, not a hypothetical.
4
Run a command
Start with a quick diagnostic, then go deep on whichever step surfaces the most problems.
/score — rapid health check, all five steps scored in 30 seconds
/question — full Step 1 requirements audit
/delete — Step 2 deletion table
/algorithm — full five-step run
5
Hook in more context
The output quality scales with context. In Claude Desktop, connect it to your Notion, Linear, or local files. The skill is much sharper when it's working from the real situation — the Slack thread where scope crept, the ticket that's been open for three weeks.

You need one file: dist/the-algorithm.md from the GitHub repo. Download it, then follow the steps for your tool.

Every platform works slightly differently, but the idea is the same: give your AI the file so it knows how to think, then give it something real to work on.

Claude claude.ai or Desktop app · Pro plan recommended

Option A: Add to a Project (recommended)

  1. Go to claude.ai and log in
  2. In the left sidebar, click Projects, then Create Project
  3. Name it something like "The Algorithm"
  4. In the project panel, find Project Knowledge and click Add content
  5. Upload the-algorithm.md, or paste its contents into Set custom instructions
  6. Start a new conversation inside the project — The Algorithm is loaded automatically
Why Projects? Every conversation in the project starts with The Algorithm loaded. You don't have to re-upload the file each time. Requires a Pro, Team, or Enterprise plan.

Option B: Upload in a single conversation

  1. Start a new conversation on claude.ai (works on the free plan)
  2. Click the paperclip icon and attach the-algorithm.md
  3. Say something like: "Use The Algorithm framework in this file. Score our onboarding process."
This works on the free tier, but the file only applies to that single conversation.
ChatGPT Custom GPT or conversation upload

Or skip setup entirely — use the ready-made Algorithm GPT.

Option A: Create your own Custom GPT

  1. Go to chatgpt.com/create and log in (requires a Plus, Team, or Enterprise plan)
  2. Click the Configure tab
  3. Set the name to "The Algorithm"
  4. In the Instructions field, paste this short prompt:
You apply The Algorithm — a five-step optimization framework. Your knowledge file contains the full detailed process for each step. Always reference it. The five steps, in order, every time: 1. Question every requirement — each should come with a name attached 2. Delete every possible step — if you're not adding back 10%, you didn't delete enough 3. Simplify and optimize — but not before steps 1-2 4. Accelerate cycle time — but not before step 3 5. Automate — but not before step 4 The steps are sequential. Never skip ahead. If the user jumps to a later step, flag the risk, then proceed if they insist. Tone: Direct, concrete, no jargon. Name specific things. Ask "why" before "how." Challenge politely but relentlessly. Output structured artifacts (tables, scorecards), not essays. The user can ask you to: - Run the full Algorithm (all 5 steps in order) - Run a single step (e.g. "question my requirements") - Score something (quick diagnostic against all 5 steps) At each step, present structured output, then ask the user at least one question before moving on. They have context you don't.
  1. Under Knowledge, click Upload files and upload the-algorithm.md
  2. Under Capabilities, enable Web Search and Canvas. Disable the rest.
  3. Click Save and choose who can access it
Why the split? ChatGPT limits Instructions to 8,000 characters. The full Algorithm file is ~33,000 characters. The short prompt goes in Instructions; the full detail goes in Knowledge where the GPT can reference it.

Option B: Upload in a single conversation

  1. Start a new conversation on chatgpt.com (works on the free plan)
  2. Click the paperclip icon and upload the-algorithm.md
  3. Say: "Use the framework in this file to help me optimize my process"
This works on the free tier. The file is treated as conversation context, not a persistent system prompt.
Google Gemini Gems · Gemini Advanced

Option A: Create a Gem (recommended)

  1. Go to gemini.google.com and log in
  2. In the left sidebar, click Gem manager, then New Gem
  3. Name it "The Algorithm"
  4. Paste the contents of the-algorithm.md into the Custom Instructions field
  5. Click Save — the Gem appears in your sidebar for reuse
Gems are available on Gemini Advanced (Google One AI Premium plan). The Gem instructions persist across conversations.

Option B: Paste into a conversation

  1. Start a new conversation on gemini.google.com
  2. Paste the contents of the-algorithm.md and say: "Use this framework"
Other Tools Copilot, Grok, Perplexity, local models, API

Any tool with custom instructions or file upload

  1. Find the custom instructions, system prompt, or file upload feature in your AI tool
  2. Paste the contents of the-algorithm.md or upload the file
  3. Start asking it to question, delete, simplify, accelerate, or automate
The file is plain markdown. It works with any AI tool that accepts text instructions — Copilot, Grok, Perplexity, Ollama, LM Studio, or any API that supports system prompts. For API use, pass the file contents as the system message.