Ep. 63: What Should Your AI Have Access To? - Prompting Curiosity | Lyssna här

In this episode I cover how AI access and permissions have changed since ChatGPT launched back in 2022, from zero connections and basically zero worry to the current YOLO approach of vibe coders and early adopters. I share what I currently give Claude access to, what my biggest concern is, and offer a read/write/act framework that you can use to decide what your AI should touch.

Main Topics Covered

How AI access sentiment has changed over 4 years
2022: ChatGPT with zero connections
2023: Bard connects to Gmail, Docs, Drive, Maps
MCP and the "AI in my email" era
Agentic AI, Claude Code, and Claude Cowork
What I currently give Claude access to
Prompt injection: my biggest concern
Read/write/act framework
Privacy, client data, and benefit vs. risk
How I used AI this week: Claude-Descript connector and pulling quotes from transcripts

Links & Resources for This Episode

Chapters

(00:00:05) - Intro and Episode Overview
(00:01:32) - Early AI Access: ChatGPT and Bard Origins (2022-2023)
(00:04:16) - MCP and the Rise of Agentic AI (2024-2025)
(00:06:23) - Claude Cowork and Mainstream Agentic Adoption
(00:08:23) - Personal Evolution in Granting AI Access
(00:10:15) - Prompt Injection Risks Explained
(00:11:20) - A Framework: Read, Write, Act and Whose Data
(00:13:38) - Privacy Concerns vs Personal Comfort
(00:15:01) - Training Data Concerns
(00:15:28) - Handling Clients' and Others' Data
(00:17:08) - Healthcare, Therapy, and Sensitive Data Gray Areas
(00:18:34) - Benefit vs Risk: The Big Picture on AI Access
(00:20:26) - Final Thoughts on What Your AI Should Access
(00:20:58) - How I Used AI This Week: Claude and Descript
(00:23:10) - Outro and Sign Off

Rss