Ryan details how Notion leverages custom AI agents—like the 'hot potato' agent behind Project Afterburner—to automate daily standup pre-reads by aggregating data from Slack, GitHub, Honeycomb, and meeting transcripts. This cuts prep time, reduces burnout, and ensures equitable visibility of all engineers’ work. He champions spec-first development: using Whisper to capture ideas and Codex to author and implement specs directly, with versioned specs acting as living documentation and changelogs. Internally, Notion’s 'Boxy' system enables engineers to @mention Codex in Notion comments and receive verified PRs with UI screenshots in under 21 minutes. Ryan stresses that fast CI is non-negotiable—not just for human velocity, but as a force multiplier for autonomous agents (citing Stripe’s 1,300 weekly AI PRs). He also argues that engineering leaders must keep coding to stay grounded in tooling, automation, and AI realities—and that prompting AI to defend its reasoning improves output quality. Ultimately, the shift isn’t about replacing humans, but redirecting attention toward systems thinking, architecture, and high-leverage decisions.