update
Apr 30, 2026
By Teun
Anthropic teams show how they use Claude Code at work
Anthropic said teams across the company use Claude Code for debugging, code navigation, testing, incident response, documentation, and custom automation. The company shared examples from engineering, design, security, marketing, data science, and legal teams, showing how the tool is used beyond traditional software development.
Anthropic says teams across the company are using Claude Code for far more than writing code. In a post dated July 24, 2025, the company described how employees use the agentic coding tool to debug production problems, explore unfamiliar codebases, automate repetitive work, and build custom internal tools.
The examples span technical and non-technical teams. Anthropic said new data scientists on its Infrastructure team feed Claude Code their entire codebase so the tool can read CLAUDE.md files, explain data pipeline dependencies, and show which upstream sources feed dashboards. The company said that replaces the need to rely on traditional data catalog tools during onboarding.
⚡ New to this?
This story shows how an AI coding tool is moving beyond software engineering and into everyday business work. Claude Code is Anthropic’s agentic coding assistant, meaning it can take a task, make decisions, run steps, and iterate with some autonomy instead of only suggesting text.
For non-experts, the main point is that AI tools are starting to help people who are not programmers build internal tools, write tests, handle documentation, and even troubleshoot incidents. That matters because it changes who can create software and how quickly teams can respond to problems.
🦞 OpenClaw angle
If you run self-hosted agents, copy Anthropic’s pattern: start by wiring the agent into your real work artifacts, such as runbooks, code comments, design files, and internal docs, instead of treating it like a generic chat bot. Build workflows where the agent can draft, test, and revise in a sandbox, then require human review before anything reaches production.
Also prioritize tasks that are repetitive or context-heavy, like incident triage, test generation, documentation cleanup, and bulk content variations. Those are the places where an autonomous agent can save time without needing full trust on day one.
On the Product Engineering team, Claude Code is used as a first stop for programming tasks. Anthropic said engineers ask it which files to inspect for bug fixes, features, or analysis, so they do not have to spend as much time gathering context before they start building.
Testing and code review are another major use case. According to Anthropic, the Product Design team uses Claude Code to write comprehensive tests for new features, while also automating pull request comments through GitHub Actions. The Security Engineering team said it moved from a manual, stop-and-start process of writing code and then trying to add tests, to using Claude for pseudocode and test-driven development.
The company said Claude Code also helps teams work in unfamiliar languages. Anthropic said members of its Inference team use it to write test logic in native languages like Rust after explaining what they want to verify, even when they are not fluent in that language themselves.
For production incidents, Anthropic said Claude Code helps teams reason through stack traces, documentation, and system behavior in real time. The Security Engineering team said this cuts the time needed to trace control flow through code during incidents. Anthropic said problems that once took 10 to 15 minutes of manual scanning now resolve about three times faster.
The Data Infrastructure team shared a specific outage example. When Kubernetes clusters stopped scheduling pods, the team fed Claude Code dashboard screenshots. Anthropic said the tool guided them through Google Cloud’s interface until they identified pod IP address exhaustion, then gave commands to create a new IP pool and add it to the cluster. The company said that saved 20 minutes during the outage.
Claude Code is also being used to speed up feature development and prototyping. Anthropic said Product Design employees feed it Figma files, then set up autonomous loops where the tool writes code, runs tests, and iterates with limited human review. In one case, the team had Claude build Vim key bindings for itself.
The company said design teams are also using Claude Code earlier in the process, not just during implementation. Anthropic said Product Design uses it to map error states, logic flows, and system statuses during design, which helps catch edge cases before development starts.
Some of the most notable uses come from people who are not writing in the target language themselves. Anthropic said data scientists use Claude Code to build React applications for visualizing RL model performance, even when they do not know TypeScript. In sandbox environments, the tool writes the visualization code from a one-shot prompt, with small tweaks if needed.
Anthropic also pointed to documentation work. The company said technical knowledge is often scattered across wikis, code comments, and team members’ heads, and that Claude Code can pull that material together through MCP and CLAUDE.md files. The Inference team said this lets people without ML backgrounds get explanations of model-specific functions faster, while Security Engineering uses the tool to turn multiple documents into markdown runbooks and troubleshooting guides.
Outside engineering, Anthropic said the Growth Marketing team built a workflow that processes CSV files with hundreds of ads, flags underperformers, and generates new variations within character limits. The company said the system uses two specialized sub-agents and can generate hundreds of ads in minutes. The team also built a Figma plugin that can generate up to 100 ad variations by swapping headlines and descriptions.
Anthropic said its Legal team used Claude Code to prototype phone tree systems that help employees find the right lawyer inside the company. Across the examples, Anthropic said the strongest results came when teams focused on the human workflow around the task, not just code generation itself.
