Claude Writes 95% of Code

Situation

In 2023, Anthropic was experiencing rapid growth as an AI company developing Claude, their large language model. As part of their own development process, they began using their AI assistant internally for software engineering tasks. The company faced the challenge of maximizing engineering productivity while maintaining high quality standards during a period of intense competition in the AI space.

Actions

Implementing AI-Assisted Coding

Anthropic deployed Claude to assist their own engineering teams
Engineers adopted a workflow where Claude generated the majority of code
The team focused on being "ambitious" with their prompts, asking for complete solutions rather than small snippets
When Claude didn't succeed on first attempts, engineers would try multiple times (up to 3-4 attempts) rather than giving up
Engineers maintained oversight and quality control while letting Claude handle implementation details

Expanding Beyond Engineering Teams

Anthropic discovered that non-engineering teams could also benefit from code-writing capabilities
Legal and finance teams began using Claude Code to automate document review and data analysis
The company encouraged teams to take risks with AI tools even when it felt uncomfortable or unfamiliar

Results

Dramatic productivity increase: Teams produced 10-20× more code with the same number of people
95% of code written by Claude in many projects
Broader technical capabilities: Non-technical teams gained ability to perform technical tasks
- Legal team used Claude to redline documents
- Finance team ran BigQuery analyses of customer and revenue metrics
Maintained hiring pace: Despite automation, Anthropic continued aggressive hiring due to the expanding scope of work

Key Lessons

Persistence pays off with AI tools: Trying multiple times with the same prompt can yield success due to the stochastic nature of AI outputs
Ambition level matters: Teams that ask for complete solutions rather than small pieces get more value from AI coding assistants
AI expands the pie: Rather than replacing jobs, AI initially enables teams to accomplish much more, focusing human attention on higher-value work
Non-technical teams can leverage technical AI tools: With proper interfaces, legal and finance teams can perform technical tasks previously requiring engineering support
Adoption requires risk-taking: Teams must be willing to experiment with AI tools even when they feel uncomfortable or unfamiliar
AI doesn't immediately reduce hiring needs: The expanded capabilities create new opportunities that still require human oversight and direction