@pro.glitch: How to set up an AI agent safely on your computer (Claude Code permissions, 5 steps) How to set up an AI agent safely before you let it on your computer. AI agents like Claude Code open files, run commands and connect to your apps, so the setup decides what it can reach. Ten minutes of setup, and it only touches what you meant it to touch. In the video it's Claude Code in VS Code on a made-up bakery's order folder; the runs and the receipt are real. How to set up an AI agent safely: 1. Give it one folder. Open Claude Code in the folder for the job and nowhere else. It writes inside that folder. The first time it wants to read outside it, it asks; pick "No, and block reads outside the working directories from now on" and your tax returns and family photos stay out. 2. Change the starting mode. New Claude Code versions start in auto mode, where a background safety check approves most actions. For the first run on business files, switch to plan mode: it reads, shows you the plan, and nothing changes until you say yes. After that, manual mode asks before every edit and every command. 3. Keep secrets out of the folder. Passwords, card numbers, the salary spreadsheet: if the job doesn't need it, it isn't in the folder. Never paste a password into the chat. 4. Read before write. When you connect Shopify, Gmail or your accounting app, start with read access (like read_orders or gmail.readonly) before write access. Anything that leaves the building, an email, a payment, a post, waits for your yes. 5. Ask for a receipt. Put it in CLAUDE.md: end every job with what you read, what you changed and what you chose not to do. Two minutes to check. For anything you can't undo, work on a copy. Works for small businesses, bookkeeping, e-commerce, agencies, office work and anyone trying AI agents, Claude Code, Claude Cowork or ChatGPT agents on real business files. #aiagents#claude#claudecode#aiemployee

Professor Glitch

Open In TikTok:

Region: US

Tuesday 03 November 2026 17:50:43 GMT