Tool Use - AI Conversations How To Build A Successful AI Product (ft Swyx)
24 snips
Nov 18, 2025 
Swyx (Shawn Wang), co-founder of AI Engineer and a developer tooling expert, dives into the evolving landscape of AI products. He discusses why vertical AI agents are vital for developers, advocating against the standalone vector databases hype. The conversation explores the shift from 'vibe coding' to structured engineering, stressing the importance of extracting prompts from codebases. Additionally, Swyx highlights the challenges of agentic browsing and the significance of concise product quality over just distribution.
AI Snips
Chapters
Transcript
Episode notes
Ship Slack And Phone Voice Integrations
- Do integrate AI assistants into Slack because it's where much work happens and resembles a junior engineer workflow.
- Try telephony voice (PSTN) for robust voice interfaces since phone calls handle continuity better than VoIP.
Why MCP Gained Traction
- MCP (model call protocol) won because it gave an AI angle to build better, permissioned integrations and energized engineers.
- Shawn notes the spec has underexplored areas beyond tool calls like prompting and inference controls.
Permissionless Data Access Tradeoffs
- Browser/computer scraping is permissionless and can extract user-visible data without APIs, but it's far more expensive and fragile.
- Shawn frames this as a long-tail frontier for data access, not a universal replacement for APIs.