Скачать или смотреть The Claude Code Workflow a $4B Unicorn Copied From Me

An AI-assisted development workflow that was organically adopted by engineers at a $4 billion unicorn — and I didn't know anyone was using it until they asked me to present it division-wide.

In this video, I walk through my complete workflow: custom commands, specialised subagents (VAPI expert, Pipecat expert), and how I manage context to get consistent results from Claude Code.

We build a real feature live: JSON-based assistant configuration for a voice AI pipeline using Pipecat.

TIMESTAMPS
0:00 - Introduction & why this works
03:41 - Kicking off the plan-task command
06:59 - What the subagents produced
08:30 - Implementation notes & idiomatic code
11:11 - Project-specific learnings (the "dollar store memory" system)
12:26 - Other commands: root cause analysis, create-PR, create-subagent
17:18 - Implementation & corrections
20:44 - Live demo: switching TTS providers via config
22:22 - Summary and close

KEY CONCEPTS
Context management: getting the right information to the AI at the right time
Task decomposition: how complex a task can you give at once
Subagents as researchers, not implementers
Building project-specific "memory" through documentation

RESOURCES
Pipecat: https://github.com/pipecat-ai/pipecat
Claude Code: https://claude.ai/code