with-agents

Build Crew - Series

Navigating the AI Agent Landscape

Amp, Inc. Episode 2 7 of 8
Open original playlist (opens in a new tab)
Episodes Browse 8 summaries 7 of 8

Build Crew Live: Episode 2

In episode 2 of "Build Crew Live", Thorsten Ball, Nico, Ryan Carson, and Tim—an Amp product manager who started building with agents—discuss AI agents. They cover changing product and engineering roles, small focused threads, and preserving context across longer projects.

Key Points Covered

  • Build Crew as a learning community: The panel describes Build Crew as a place for users of Amp, Cursor, Claude Code, Codex, and other tools. Members compare workflows, do standups, and learn from one another [00:00:00]-[00:00:56], [26:04]-[28:03].

  • PMs and non-engineers can become builders: Tim describes moving from product and filmmaking, rather than professional coding, to using AI agents to build software. The panel argues that when agents lower implementation costs, the old product/design/engineering triangle starts becoming a more fluid "blob" [04:10]-[09:27].

  • Use agents for bounded work you already understand: Thorsten recommends small, known tasks, such as fixing a VS Code integration bug. The human should know the desired shape and be able to review the result quickly [10:20]-[13:22].

  • Manual coding decreases, but review responsibility increases: The group discusses how much code they write by hand and when they still edit load-bearing logic themselves. They say reviewing generated code becomes the main skill [14:17]-[23:04].

  • Agent-friendly verification matters: Ryan uses Playwright, unified frontend/backend logs, and predictable dev accounts. These let the agent test its work and read relevant output instead of relying on manual screenshots or scattered logs [17:08]-[20:17].

  • Custom slash commands provide deterministic context: Nico demonstrates slash commands that fetch Linear issues and GitHub PR comments into a prompt. His goal is to provide exact structured context rather than ask the agent to discover everything through broad, noisy MCP-style tools [28:03]-[38:45].

  • Break big work into artifacts and fresh threads: Tim describes a "loop of doom" when a project becomes too large for one conversation. He creates PRDs, divides them into granular tasks, writes explicit artifacts, and restarts with clean context when needed [41:39]-[57:45].

  • Context hygiene is a discipline: The panel uses the "don't cut fruit on the chicken board" metaphor for agent context: carry forward distilled plans and decisions, not every failed attempt from a messy thread [49:33]-[57:45].

Full Video: Watch on YouTube(opens in a new tab)

Build Crew: https://buildcrew.team(opens in a new tab)