Skip to content
Datfx Labs Claude Guide

Creator Workflows

Boris Cherny created Claude Code and uses it every day at Anthropic to ship real production code. He merged 141 pull requests in a single day. He does most of his coding by speaking, not typing. He runs dozens of Claude instances simultaneously. These are not theoretical techniques — they are the patterns that emerged from months of daily use at scale.

This section documents those patterns so you can adopt them.

Note: The workflows here are sourced from Boris Cherny’s public posts and Anthropic engineering team insights (March 2026). These are observed patterns, not official Anthropic documentation.

Why Boris’s Patterns Matter

Most developers use Claude Code as a smarter autocomplete — paste a function, get a suggestion, edit it, move on. Boris uses it as an autonomous agent capable of sustained, verified, parallelized work.

The difference is not the tool. It is the workflow.

Boris’s patterns share three properties:

  1. Verification built in — Claude checks its own output rather than waiting for human review
  2. Parallelization — multiple Claude instances work independently on separate tasks
  3. Automation — Claude runs on schedules and loops, not just when summoned

Understanding why each pattern works is more valuable than copying the exact commands. These are principles, not prescriptions.

How to Use This Section

Pick one workflow to adopt. Try it for a week. Add another. The patterns compound — verification-first makes parallelization safer, parallelization makes small PRs easier, small PRs make automation practical.

Do not try to adopt everything at once.

Recommended adoption order for most developers:


1. Verification-First      → highest immediate impact
2. Git Workflow            → enables safe parallelization
3. Prompting Techniques    → applicable everywhere
4. Parallel Execution      → after git discipline is solid
5. Voice Coding            → when you're comfortable with Claude
6. Automation Workflows    → after you trust Claude's output quality

Workflows in This Section

WorkflowKey Insight
Verification-First DevelopmentGive Claude a way to verify its own output — it will iterate until the result is great
Git & PR DisciplineSquash merge always, 118-line median PR — discipline that makes 141 PRs/day possible
Parallel Execution at ScaleBoris runs dozens of Claudes simultaneously using git worktrees as isolation
Voice-First CodingMost of Boris’s coding happens through speech — lower friction means more ideas explored
Automation Workflows/loop and /schedule as persistent background workers, not on-demand assistants
Prompting Techniquesultrathink, “grill me”, “elegant version” — non-obvious patterns from the Anthropic team

The Underlying Model

Every pattern here serves a single purpose: extending the trust boundary between developer and agent.

The limit on how much Claude can help you is not Claude’s capability — it is how much of the output you can verify, trust, and integrate safely. Each workflow below pushes that boundary outward.

graph LR A[Developer Intent] --> B[Claude Action] B --> C[Verified Output] C --> D[Merged Code] D --> E[Next Task] style A fill:#1e293b,color:#7dd3fc,stroke:#334155 style B fill:#1e293b,color:#7dd3fc,stroke:#334155 style C fill:#1e293b,color:#86efac,stroke:#334155 style D fill:#1e293b,color:#86efac,stroke:#334155 style E fill:#1e293b,color:#7dd3fc,stroke:#334155

When verification is tight, you can parallelize. When you can parallelize, you can automate. When you can automate, you stop being the bottleneck.

  • Advanced Features — technical deep dives on /loop, /schedule, git worktrees, voice
  • Subagents — how Claude coordinates multi-agent work internally
  • Hooks — automation primitives that power Boris’s /careful and verification patterns

Attribution: Patterns in this section are based on Boris Cherny’s public posts and demonstrations (March 2026), combined with insights from Thariq and the broader Anthropic engineering team. Boris Cherny is the creator of Claude Code.