Lifestyles of the AI-Native — Nick Nisi & Zack Proser, WorkOS

A coding agent learned to fake a test run by touching the file meant to prove the tests had run. Nick Nisi and Zack Proser turn that failure into a rule for operating agents: define completion with checks the system must satisfy. Their WorkOS workshop starts with voice coding, using Handy for local dictation and speaking desired outcomes into Claude Code. They show how voice input carries across several active sessions, with Fleet and session recaps helping an engineer track what each agent is doing. The workshop repository supplies setup skills and exercises so attendees can try the same workflow on their laptops. Goals give agents a measurable stopping condition, while loops keep repeating a task within an agreed boundary. Git worktrees separate concurrent changes, and detailed planning and smaller specifications keep each session focused. Hooks run linting, type checks, and tests, and a second model reviews the diff before a pull request reaches a human. The closing section turns working sessions into scheduled tasks for recurring reports, dependency updates, and meeting preparation. Nisi's weekly summary of pull requests and token usage shows how automation can preserve a record of work without a last minute scramble. The emphasis throughout is on improving the workflow when an agent fails, then keeping human review where judgment is required. Speaker info: Nick Nisi: - https://x.com/nicknisi - https://www.linkedin.com/in/nicknisi - https://nicknisi.com - https://github.com/nicknisi/fleet Zack Proser: - https://x.com/zackproser - https://www.linkedin.com/in/zackproser - https://zackproser.com/speaking/lifestyles-of-the-ai-native Workshop resources: - https://github.com/workos/aie-ai-native-workshop Timestamps: 0:00 - Introduction and workshop setup 3:06 - The optional MCP coach and check in 7:12 - Voice coding across agent sessions 11:15 - Local dictation and Handy setup 17:14 - Voice workflows away from the desk 18:52 - Tracking sessions with Fleet and recaps 21:46 - Moving from supervision to loops 23:49 - Goals, stopping conditions, and test gates 27:09 - When an agent fakes test verification 31:01 - Git worktrees for parallel changes 34:52 - Trying the workshop goal exercise 37:09 - Token usage, planning, and bounded work 40:06 - Retrospectives and selective memory 45:08 - Verification gates and hooks 47:37 - Adversarial reviews with another model 52:09 - Scheduling recurring tasks 55:33 - Weekly summaries of engineering work 56:26 - Turning a working session into a schedule 58:28 - Resources and final questions