TFTC: A Bitcoin Podcast #722: AI Privacy Is The New Bitcoin Privacy with Mark Suman
51 snips
Mar 2, 2026 
Mark Suman, co-founder of Maple AI and builder of privacy-preserving, agentic AI tools. He discusses the rapid rise of AI agents and the privacy risks of closed-source models. He explores open-source tooling gaps, how agents will change app interactions, and why private, decentralized AI and open hardware matter for freedom and future wearables.
AI Snips
Chapters
Transcript
Episode notes
Open Source Models Are Now Competitive
- Open source models have caught up to frontier labs and now match their intelligence for many tasks.
- Mark Suman cites models like Kimi K 2.5 and improved reasoning, math, and coding as evidence the gap is closed.
Isolate Agentic AI Like You Would A Virus
- Treat agentic tools as if they have full root access and isolate them from sensitive networks and machines.
- Run agents in a VPS or separate locked-down computer and avoid using them on the same Wi‑Fi with private devices.
Tooling Not Models Is The Bottleneck
- The main remaining gap for open source AI is robust hosting/middleware (the harness) for tool calling and system prompting.
- VLM-style universal hosts struggle; proprietary frontiers win by building specialized middleware around models.