AI Glossary

Claude Chat

The classic question-and-answer app. It runs in Anthropic's cloud and cannot reach the files on your computer. Best for quick, one-off questions.

Claude Cowork

An agentic desktop app that runs on your own machine and can read and write your local folders. This is the home of your Second Brain, where automations and deep work happen.

Claude Code

The most powerful environment, aimed at building projects and applications. It is where large, complex, multi-step work lives.

Projects (inside Chat)

A folder-like space inside Chat. Convenient, but encapsulated and locked into Anthropic's ecosystem: no visible memory, no easy backup, no portability.

Rule of thumb: when starting new work, always pick "Choose a different folder," never "Create new project."

Local directory

A normal folder on your computer. Files here are your asset: fully visible, easy to back up, and portable, so you could move them to any other AI (Gemini, OpenAI) in the future.

Second Brain

A structure of pre-compiled Markdown files that act as the AI's memory. Open a fresh session and the context is already there, no reprocessing documents, no exhausting the context window.

Markdown (.md) as memory

Plain text with light formatting. It is the native format for AI memory and context: readable by you, portable, and lightweight, so knowledge accumulates day by day into a living base.

Context window

How much the model can "hold in mind" at once. Claude Opus 4.8 reaches about 1 million tokens (roughly 750,000 words). Very long sessions degrade quality; switching sessions backed by a Second Brain solves it.

Subagents / dynamic workflows

Given a big, complex problem, Claude can launch dozens or hundreds of parallel agents that coordinate, divide the work, and critique each other's output. It is how one request becomes a whole research team.

LLM OS

Andrej Karpathy's idea of treating the language model like an operating system, with the Second Brain as its memory and files. It is the mental model behind this whole approach.

Computer Use

Claude takes control of the computer itself, mouse, keyboard, any Windows app, to carry out tasks. We used it to auto-download your files out of Chat into a local work folder.

Browser Use

Claude operates inside Chrome through an extension: navigating sites, reading Outlook on the web, filling forms. Related to Computer Use, but scoped to the browser.

"Ask before acting" vs "Act without asking"

The two permission modes. Start in "Ask" to learn the dynamic; switch to "Act without asking" once you trust it, so Claude does not stop for approval on every click.

Daily Briefing

The automation we built: a deep daily search on digital-infrastructure topics that saves Markdown files to your work folder and generates an HTML newspaper-style panel. The knowledge base grows every day.

Power move: ask Claude what it has compiled over the last 30 days for an instant trend summary.

Connectors

Integrations that let Claude reach external apps such as Gmail, Google Drive, Outlook, Slack and more. In the session we connected Gmail and Google Drive successfully.

MCP (Model Context Protocol)

The open standard that lets Claude talk to external tools and data sources in a consistent way. Connectors and specialized search engines plug in through MCP.

Exa MCP

A semantic search engine built for AI agents, stronger than a plain Google query for deep research. It powers the Daily Briefing and any serious web search.

Microsoft 365 / Outlook

The corporate connector needs your IT team's approval, so it is blocked for now. Until Scala's IT authorizes it, Computer Use or Browser Use can reach Outlook as a workaround.

Gmail & Google Drive

Both connected during the session as live sources of email and files, so Claude can read and work with them directly.