Matt Paige and Thomas Schlossmacher discuss a shift from typing to talking as AI makes voice dictation accurate enough to use without constant corrections, arguing speech is faster and more natural and helps maintain thought flow when interacting with AI tools. Schlossmacher is building Resonant, a Mac voice dictation tool designed to run on-device so nothing goes to the cloud, motivated by privacy concerns and data retention/training practices of cloud-based alternatives like Whisper Flow. They explore the tradeoffs of local vs server inference, noting current consumer hardware can struggle to run full speech-to-text plus LLM post-processing fast enough, but expects improvement in 1–2 years. Schlossmacher explains differentiators like taste/brand, his design workflow using inspiration sources and ShadCN, his path into AI-assisted building, his stack (Claude Code, Next.js, Convex, Vercel), and a vision for proactive, context-aware agent features and potential open-sourcing and enterprise/self-hosted options, with beta/free access at https://www.onresonant.com/. Key moments: Making the Switch Why Build Resonant Local LLM Reality Check Standing Out in AI Designing Resonant Brand Building