The biggest breakthrough in my own AI usage was starting to use repositories as habitats/factories for my AI agents to work in (January 2025). This sounds weird if you haven’t tried it, but works incredibly well.
Below is my attempt to explain what I mean and how to do it yourself using free tools. There’s a copy-paste prompt at the end for you to get started with.
Cheers, Graham
Most people use AI as a brain.
They ask it questions. They get answers. They paste text into emails, documents, or slides.
That’s useful. But it’s also why so many people feel like AI is impressive but somehow… flimsy. Nothing really sticks. Every day feels like starting again.
There’s another way to use AI that feels completely different. AI as a factory.
The difference isn’t intelligence. It’s environment.
The brain mode
When you use AI as a brain, you’re having a conversation.
But once the conversation ends, the work mostly evaporates. The AI doesn’t remember exactly what you decided last week. It can’t refine yesterday’s output automatically. It can’t check its own work against reality.
It’s like having a very smart consultant who never builds anything.
The factory mode
A factory is not smarter than a brain. It’s organized.
A factory has:
a workspace
tools
raw materials
processes
quality checks
The key difference is this: factories accumulate progress.
Once you build a production line, it keeps producing. Improvements compound. Mistakes get fixed once instead of repeatedly.
When you give AI a proper workspace (not just a chat window) it stops being a clever assistant and starts being an engine.
Why a “workspace” matters
A workspace is simply a place where:
work can be stored
changed work replaces old work
multiple AIs can collaborate
results can be checked
improvements persist
This is why people who use AI inside a shared workspace (like a GitHub code repository) experience a completely different reality from those who only use chat.
The AI isn’t just thinking. It’s doing. And that doing compounds over time.
What this unlocks
You don’t need to be a programmer to benefit from this mindset.
Imagine:
Instead of asking AI to rewrite the same report every month, you have a living template that improves itself.
Instead of manually tracking tasks, the system updates your process documents automatically.
Instead of fixing the same mistake over and over, the AI fixes it once and never repeats it.
Instead of “ideas,” you get systems that quietly run in the background.
This is how AI stops feeling like magic and starts feeling like practical leverage.
Why most people never get here
Because it requires one mental shift:
Stop treating AI like a brain you chat with.
Start treating it like a machine you operate.
That means:
giving it a place to work
letting it change things directly
allowing structure to form
caring more about results than clever answers
For many people, that feels unfamiliar. But once you experience it, going back to chat-only AI feels like trying to run a business with sticky notes.
The quiet advantage
The people who figure this out won’t necessarily talk about it.
They’ll just:
get their work done faster
fix problems once and never again
build things that don’t fall apart
and quietly move ahead while others are still “prompting”
AI didn’t just give us a smarter brain.
It gave us the ability to build our own virtual factories: cheaply, incrementally, and independently.
Copy-paste prompt to start your first AI factory with free tools:
I want to move from using AI as a “brain” (chatting, drafting, thinking) to using AI as a factory that accumulates work over time. Please help me do the following, step by step, assuming I am not technical.
First, explain in very simple terms what a GitHub repository is and why it matters as a workspace for AI. Then walk me through setting up my first GitHub repository for free, including exactly what to click and which choices to make. After that, help me connect an AI coding agent (such as Cursor) to that repository using their free tier.
Next, ask me questions about the kind of work I do (for example writing, research, analysis, operations, reporting, coordination, or other work) and help me identify one small, concrete workflow that could be automated or systematized using AI plus a repository. Then help me create a very simple starting structure inside the repository (folders or files) that supports that workflow.
Finally, show me how to let an AI coding agent safely make changes directly in the repository and improve the system over time, rather than starting from scratch each time.
Important constraints: do not assume I know programming; avoid technical jargon unless absolutely necessary and explain it plainly when you use it; focus on practical steps rather than theory; and remember that the goal is not to build something perfect, but to build something that persists and improves instead of disappearing when the chat ends.
Treat this as a guided setup session, not a lecture. Ask me questions when needed and adapt your guidance based on my answers.
Absolutely brilliant! This fundamentally changes the way I think about AI as a productivity tool to get work done faster. I'll definitely be trying this
This factory metaphor lands way better than the usual 'AI as a tool' framing. The compounding progress point is what most people miss, they keep rebuilding from scratch instead of letting systems accumulate improvements. I've noticed the same pattern with teams using AI in code reviews versus teams just usng it for one-off code generation, the delta in output quality over a few months is massive.