The Paradigm Shift
AI moved from the chat box to the operating system.
"It's not that we didn't think of letting AI operate computers — it just wasn't smart enough before. Now it is."
What Changed
Around fall 2024, something fundamental shifted in how people work with AI.
Before: The Chat Box Era
AI was a separate application. You:
- Opened a chat interface
- Copied text from your document
- Pasted it into the chat
- Waited for a response
- Copied the answer back out
- Repeated for every paragraph
You did most of the work. AI just gave you answers.
After: The Agent Era
Modern AI models crossed the intelligence threshold. They can now:
- Read your entire document directly
- Edit files on your computer
- Run commands in the terminal
- Follow complex, multi-step instructions
- Ask for clarification when needed
AI does most of the work. You supervise.
The Intelligence Threshold
Why didn't we do this before? We could have built AI tools that operate computers — the idea isn't new. But earlier models weren't reliable enough.
When AI makes mistakes:
- In a chat box → you catch it and rephrase
- Operating your files → it could break things
The threshold was crossed when models became:
- Reliable enough to follow complex instructions
- Self-aware enough to ask before risky actions
- Capable enough to understand context from many files
This isn't just "better AI" — it's a qualitative change in what's possible.
Both the Medium and the Message Changed
| Aspect | Before | After |
|---|---|---|
| Interface | Chat box | Terminal / IDE |
| Workflow | Copy-paste-wait | Point and approve |
| Your role | Operator | Supervisor |
| AI's role | Answerer | Agent |
| Output | Text in chat | Changes to your files |
The medium changed: from isolated chat to integrated tool.
The message changed: from "give me an answer" to "do this task."
What This Means for Research
For humanities scholars, this shift has practical implications:
- Scale — AI can process your entire archive, not just one paragraph
- Consistency — Same instructions, same behavior, every time
- Traceability — Every change is tracked in version control
- Iteration — Refine prompts like you refine code
You're not "using AI" anymore. You're building AI-powered workflows.
The Tools
Several tools embody this new paradigm:
| Tool | Description |
|---|---|
| Claude Code | Anthropic's terminal-based AI agent |
| Cursor | AI-integrated code editor |
| Windsurf | Another AI-first editor |
| Aider | Open-source AI coding assistant |
They all share the same core idea: AI reads your files, follows your instructions, and makes changes directly — with your approval.
Next: Learn the basics of Terminal/CLI — your interface to these new tools.