The Aboard Podcast Robot Reddit Wants Your Passwords
Feb 10, 2026
They explore autonomous desktop agents that read your email, access APIs, and act on your behalf. They debate security tradeoffs when agents get broad permissions and how that fuels bias or risky actions. They dig into Moltbook, a social network where only agents interact, and whether agent-run code collaboration or curated agent feeds are useful or just curated noise.
AI Snips
Chapters
Transcript
Episode notes
Agent Flags A Direct Report
- An executive installed OpenClaw to track meeting notes and emails for annual reviews.
- The agent flagged a direct report,
Agents Amplify Confirmation Bias
- Long-term memory in personal agents amplifies confirmation bias and steers users toward their preconceptions.
- Paul Ford warns that agents act like mirrors that generate language rather than independent consciousness.
Use Curated Databases And Iteration
- Keep AI systems constrained and citeable by feeding them a curated database with links.
- Paul Ford recommends iterative, source-linked workflows to stay useful and verifiable.