Anthropic uses Claude Code well beyond its engineering teams. In its July 2025 report "How Anthropic teams use Claude Code", the company describes engineers using it to understand code, debug production problems and write tests, alongside data scientists, marketers and lawyers who use it to build their own tools. The measured results include stack-trace problems resolved about 3x faster, roughly 20 minutes saved during an outage, and an 80% cut in research time. The video below reads Claude4World's case study of that report, with each figure checked against Anthropic's original.
Anthropic's key lesson: "Claude Code works best when you focus on the human workflows that it can augment." Its most successful teams treat it as a thought partner rather than a code generator.
Engineering workflows
- Product Engineering calls Claude Code its "first stop" for any programming task, including working out which files matter in an unfamiliar part of the code.
- New data scientists on the Infrastructure team have Claude read the repository's CLAUDE.md files and explain data pipeline dependencies, which helps them get productive quickly.
- Product Design runs automated pull-request comments through GitHub Actions, with Claude handling formatting issues and test case refactoring.
- Security Engineering moved from "design doc, janky code, refactor, give up on tests" to asking Claude for pseudocode and guiding it through test-driven development.
Debugging and incident response
Security Engineering gives Claude stack traces directly. Problems that used to take 10 to 15 minutes of manual scanning now resolve about three times as quickly. During one outage, the Data Infrastructure team fed Claude screenshots of its monitoring dashboard; Claude traced the problem to pod IP address exhaustion in Kubernetes, saving the team about 20 minutes in the middle of the incident.
Claude Code for non-engineers
- Data scientists who are not fluent in TypeScript build entire React applications to visualize reinforcement-learning model performance.
- Inference team members without machine-learning backgrounds cut research from about an hour of searching to 10 to 20 minutes, an 80% reduction.
- Growth Marketing built an agentic workflow that reads a CSV of hundreds of ads, finds the underperformers and writes new variations within strict character limits, producing hundreds of new ads in minutes instead of hours.
- Product Design feeds Figma files into Claude Code and sets up loops where it builds and tests the interface on its own.
- The Legal team prototyped a phone tree that helps employees reach the right lawyer at Anthropic.
How to apply these patterns to your team
- Write a CLAUDE.md file for each repository so Claude Code starts every session with your conventions and architecture.
- Write the tests first and let Claude Code implement against them, so it has a clear definition of done.
- Paste stack traces and dashboard screenshots into Claude Code during incidents instead of scanning logs by hand.
- Give marketing, legal and operations teams access, and let them build small tools for their own repetitive work.
- Measure one task before and after adoption, so the decision to expand rests on numbers rather than impressions.
Our take: don't start by asking what Claude Code can replace. Pick one repetitive step your team does every week, hand it to Claude Code, and time the result. For the design side of agent workflows, see our guide on how to use Claude Opus 5.5 and our courses on agentic design patterns.
