Paul Graham’s 232 essays, applied to your actual business.
Your AI may already know a surprising amount about what you are building from your conversations, project files, README, documentation, codebase, or other context.
Install PG Founder and say:
PG me.
That’s it.
PG Founder uses whatever context your AI already has, searches the bundled corpus of 232 Paul Graham essays, finds the ideas that genuinely matter for your situation, and connects them directly to your business.
No questionnaire if enough context already exists.
If almost nothing is known, it asks one thing:
What are you building? A few sentences is enough.
PG Founder is not:
- a Paul Graham impersonator
- a quote bot
- an essay summarizer
- a generic startup-advice bot
- a YC application simulator
It works more like this:
Your business / product context
↓
Search Paul Graham's 232 essays
↓
Find the ideas that actually matter
↓
Look for non-obvious connections
↓
Apply them to your specific situation
↓
Insights, questions, blind spots and next moves
The interesting part is not simply finding famous essays like Do Things that Don't Scale.
PG Founder can surface ideas from anywhere in the corpus — including essays about users, product, growth, startup ideas, persistence, attention, ambition, identity, design, competition, wealth, work, learning, management, independence, decision-making, and life — when those ideas materially change how your business should be viewed.
Generic startup advice is a failure.
This:
Talk to users.
is not enough.
A useful answer should look more like:
Your product only becomes valuable if sponsorship managers actually change where they spend after seeing competitor activity. So the useful test is not whether they like the dashboard — it is whether the intelligence changes a real buying decision.
The essays are the lens.
The business-specific implication is the product.
Once installed, try:
PG me.
or:
Analyze my business through Paul Graham's essays.
or:
Use whatever you already know about my product and PG me.
If the AI has little context:
I'm building X for Y. PG me.
You can also ask narrow questions:
Should I hire now?
Am I overbuilding this?
Should I keep working on this product?
My competitor just shipped the feature I was building. What should I do?
I have five ideas and keep switching between them. PG me.
Users keep asking for a feature I don't think belongs in the product.
Analyze this through the PG corpus.
Broad questions trigger a wider corpus analysis.
Narrow questions should get a direct answer instead of a giant startup teardown.
The repository includes:
all_paul_graham_essays.md
This contains the bundled 232-essay corpus used by the skill.
For normal use, PG Founder reads from the local corpus.
No web access is required.
The smaller collection inside references/ is only a routing layer that helps the agent find promising themes and search terms quickly.
It is a map.
It is not the knowledge boundary.
For broad analysis, PG Founder may surface any relevant essay from the complete corpus.
PG Founder is designed to use whatever context the host AI already exposes.
Depending on the environment, this may include:
- the current conversation
- available memory/context
- project files
- README
- documentation
- strategy notes
- attached files
- repository structure
- code
- metrics
- previous decisions in the current context
It does not magically gain access to information the host does not expose.
The more useful context the host already has, the less you need to explain.
This means the experience can be especially useful inside:
- an AI you have discussed your company with for months
- your actual product repository
- a Claude/ChatGPT project containing your business documents
- an existing coding-agent session working on your product
PG Founder is an Agent Skill built from Markdown and supporting reference files.
There is:
- no backend
- no database
- no API
- no account
- no package dependency
- no web requirement
Repository:
https://github.com/tejas3732/pg-founder-skill
Cursor supports Agent Skills directly.
You can install PG Founder globally so it works across all your projects, or inside one project.
Use this if you want PG Founder available in every Cursor project.
mkdir -p ~/.cursor/skills
cd ~/.cursor/skills
git clone https://github.com/tejas3732/pg-founder-skill.gitYou should now have:
~/.cursor/skills/pg-founder-skill/
├── SKILL.md
├── all_paul_graham_essays.md
├── references/
└── ...
Restart Cursor if needed.
Then in Cursor Agent chat:
/pg-founder-skill
or:
Use PG Founder and analyze this project.
or:
Use whatever you know about this repo and PG me.
From your project root:
mkdir -p .cursor/skills
git clone https://github.com/tejas3732/pg-founder-skill.git .cursor/skills/pg-founder-skillThen say:
PG me.
Useful prompts:
Use PG Founder. Inspect this repo first and then PG me.
Use the full bundled Paul Graham corpus, including essays outside the curated references.
What stands out about this business?
For remote/cloud use, install the skill inside the project rather than only in a user-level local folder.
Claude Code supports Agent Skills natively.
mkdir -p ~/.claude/skills
cd ~/.claude/skills
git clone https://github.com/tejas3732/pg-founder-skill.gitThen start Claude Code:
claudeInvoke:
/pg-founder-skill
or:
PG me using whatever you know about this repository.
mkdir -p .claude/skills
git clone https://github.com/tejas3732/pg-founder-skill.git .claude/skills/pg-founder-skillThen run Claude Code inside the repo and invoke the skill.
Useful prompts:
Use PG Founder. Read the README and relevant product files first, then connect this project to the most relevant ideas across the full PG corpus.
PG me. Don't limit yourself to the obvious startup essays.
Updating:
cd ~/.claude/skills/pg-founder-skill
git pullClaude Code and Claude.ai are different environments.
If you primarily use Claude in the browser:
- Create a Claude Project.
- Add your business/product documents.
- Put the contents of
SKILL.mdinto Project Instructions. - Add
all_paul_graham_essays.mdand the reference files if your project accepts them. - Start chats inside that project.
Then say:
Analyze this business using the PG Founder instructions and the Paul Graham corpus in this project.
or:
PG me.
Codex supports Agent Skills directly.
Use:
~/.agents/skills/
for global skills, or:
.agents/skills/
inside a repository.
mkdir -p ~/.agents/skills
cd ~/.agents/skills
git clone https://github.com/tejas3732/pg-founder-skill.gitThen invoke:
$pg-founder-skill PG me using whatever you know about this project.
or:
$pg-founder-skill Analyze this business through the full Paul Graham corpus.
mkdir -p .agents/skills
git clone https://github.com/tejas3732/pg-founder-skill.git .agents/skills/pg-founder-skillThen:
$pg-founder-skill Use the repository context and PG me.
Useful prompts:
$pg-founder-skill Inspect this product and analyze it through Paul Graham's full essay corpus.
$pg-founder-skill Search beyond the curated essay map and find the most relevant non-obvious connections to this business.
Updating:
cd ~/.agents/skills/pg-founder-skill
git pullIf your ChatGPT environment exposes native Skills, install the complete PG Founder skill bundle there and invoke it from a conversation or workspace that already has useful business context.
Typical usage:
Use whatever you already know about my business and PG me.
or:
Analyze my business through Paul Graham's essays.
If native Skills are not available in your ChatGPT account, use a project/persistent workspace:
- Add
SKILL.md. - Add
all_paul_graham_essays.md. - Add
references/. - Add whatever business/product files are useful.
- Use
SKILL.mdas the reasoning instructions.
Then say:
Use the PG Founder instructions and the essay corpus provided here.
PG me.
The important part is that ChatGPT can only use conversation, memory, project, and file context actually exposed to it.
| You use | Recommended setup |
|---|---|
| Cursor every day | ~/.cursor/skills/pg-founder-skill/ |
| Cursor only for one product | <repo>/.cursor/skills/pg-founder-skill/ |
| Claude Code everywhere | ~/.claude/skills/pg-founder-skill/ |
| Claude Code for one repo | <repo>/.claude/skills/pg-founder-skill/ |
| Codex everywhere | ~/.agents/skills/pg-founder-skill/ |
| Codex for one repo | <repo>/.agents/skills/pg-founder-skill/ |
| ChatGPT with Skills | Install/upload the complete skill bundle |
| ChatGPT without Skills | Project/files + SKILL.md instructions |
| Claude.ai browser | Claude Project + instructions + corpus |
Install it once and use it across multiple businesses/projects.
Example:
PG me.
Install PG Founder inside one product repository.
This is especially useful with coding agents because they can combine:
actual product
+
codebase
+
README/docs
+
Paul Graham's 232 essays
Then:
Inspect this project and PG me.
If installed with git clone:
cd ~/.cursor/skills/pg-founder-skill
git pullcd ~/.claude/skills/pg-founder-skill
git pullcd ~/.agents/skills/pg-founder-skill
git pullFor project installs, cd into that skill folder and run:
git pullUse whatever context you already have about my business and PG me.
Inspect the README, docs and relevant product structure first.
Then use PG Founder to analyze the business — not just the code.
I'm building an AI tool that helps accountants recover unpaid invoices.
PG me.
Users keep requesting this feature.
Use PG Founder to help me decide whether the underlying need matters and whether this implementation belongs in the product.
I've worked on this for 18 months. Growth is flat.
Use the full PG corpus and tell me what I may be refusing to see.
Use PG Founder and inspect what I've actually built.
What am I overbuilding relative to what users currently need?
I keep switching between projects.
Use Paul Graham's essays across the full corpus — not only the startup essays — and tell me what this pattern suggests.
Not:
Paul Graham says to talk to users and do things that don't scale.
But:
Three essays point toward the same issue from different directions. You are treating publisher discovery as the product, while your buyer's actual job is deciding where the next dollar should go. That means the recurring value probably isn't “more inventory”; it is detecting a change that causes a buying decision.
The goal is a connection you might not have made yourself.
pg-founder-skill/
├── SKILL.md
├── README.md
├── all_paul_graham_essays.md
├── references/
│ ├── essay-index.md
│ ├── frameworks.md
│ ├── users-product-growth.md
│ └── founder-focus.md
└── examples/
├── b2b-saas.md
└── marketplace.md
The behavior.
The full 232-essay corpus used as the primary knowledge source.
A lightweight routing layer.
The references are a map.
They are not the territory.
Example applications of the skill.
It does not blindly paste all 232 essays into context.
The intended process is:
Understand the business
↓
Identify likely themes
↓
Use the reference map as a starting point
↓
Search the full corpus
↓
Expand into adjacent / contrary ideas when useful
↓
Retrieve only relevant passages
↓
Connect them to this business
PG Founder should be:
- specific
- practical
- intellectually honest
- willing to challenge your initial interpretation
- willing to find conflicting ideas
- willing to say an essay does not apply
- useful beyond venture-backed startups
It should work for:
- startups
- SaaS
- apps
- marketplaces
- agencies
- small profitable businesses
- open-source products
- side projects
- founder decisions
It should not assume every business wants venture-scale growth.
The goal is simple:
Take Paul Graham's body of work and make it useful at the moment you are making a real business decision.
SKILL.md is the source of truth for behavior.
PG Founder is an independent project and is not affiliated with or endorsed by Paul Graham or Y Combinator.
Essay titles and source links point back to the original work on paulgraham.com.