AI Now Writes All Code at Anthropic, Engineer Reveals
100% AI-Written Code at Anthropic: A Technological Milestone
In a revolutionary shift that underscores the accelerating role of artificial intelligence in software development, AI now writes 100% of the code at Anthropic, according to senior engineer Boris Cherny. This groundbreaking revelation, shared during a podcast interview featured in Fortune, signals a dramatic change not just within Anthropic, but potentially for the entire tech industry.
Cherny, now working at Roon, revealed that during his time at Anthropic—a leading AI research company and OpenAI competitor—all production code was written by AI assistants. Human engineers have transitioned into a supervisory role, focusing more on reviewing, testing, and refining code outputs driven entirely by artificial intelligence.
Human Developers Shift to Code Reviewers
This isn’t a layoff-driven restructuring or an experiment limited to isolated projects. According to Cherny, it’s standard practice at Anthropic for every line of code to be written by AI tools. Engineers no longer author original code themselves but instead interact with AI systems to generate, iterate, and finalize software.
This human-AI workflow is built on three key processes:
- Prompting: Human engineers describe the desired functionality in natural language.
- Generation: AI systems such as Claude (Anthropic’s own model) write the code based on those prompts.
- Review: Engineers read, test, and modify the AI-generated code as needed to ensure quality, security, and performance.
Cherny emphasized that modern AI models like Claude are now capable of writing complex internal tools and architecture-level solutions with minimal oversight. He pointed out that at Anthropic, engineers often never type new code, as AI systems handle even large-scale backend implementations.
The Tools Behind the Transition
Anthropic uses its own AI assistant, Claude, for software development. Similar to how developers might currently use tools like GitHub Copilot or ChatGPT, Claude is integrated into the coding workflow. But unlike these tools, which often serve as mere suggestions engines, Claude has reached a level of maturity where its output is production-ready with only minor human revisions.
This level of autonomy sets a new benchmark in AI-assisted software engineering. The tools increasingly mirror the intent and structure that experienced developers would produce, enabling scalable and consistent code across projects.
Integrated AI in Software Teams
The integration of AI into Anthropic’s development pipeline also includes mechanisms for documentation and code clarity. AI isn’t just writing code—it’s explaining itself, documenting its functions, and even suggesting better architectural patterns.
Engineers now act more as:
- Code editors and reviewers
- System architects ensuring holistic integration
- Test engineers validating application behavior
- Product collaborators influencing development workflows
This systemic change aligns with Anthropic’s larger philosophy: maximizing AI for aligned, safe, and scalable productivity.
Why Full AI Code Generation Matters
While partial code generation via tools like Copilot has been embraced across the tech world, a 100% AI-generated production codebase is a pioneering step. This shift introduces several implications for the future of software engineering:
1. Improved Development Speed: AI can write code significantly faster than humans, even across multiple languages and frameworks.
2. Standardized Code Quality: Code generated by AI is more consistent, reducing bugs caused by human variability.
3. Broader Talent Accessibility: Junior developers can achieve more with AI assistance, possibly flattening traditional learning curves.
4. Evolution of Engineering Roles: Developers no longer need to write every line themselves, freeing them to focus on design, user experience, and performance metrics.
Controversy and Concerns
Despite the enthusiasm, not everyone in the software community is ready for this level of automation. Some engineers argue that reliance on AI may lead to skills degradation for human coders. If professionals no longer write code by hand, will they retain the intuition needed for understanding edge cases or optimizing critical systems?
Others question whether AI-written code can remain secure and efficient without deep human inspection. Although preliminary feedback from Anthropic seems positive, the broader industry may take a more cautious approach.
Ethical and Security Considerations
Replacing traditional development methodologies with AI raises important concerns:
- Bias: AI may perpetuate systemic biases embedded in its training data.
- Security: Automated code generation could introduce repeatable vulnerabilities if not properly vetted.
- Transparency: AI can occasionally produce obfuscated logic that may be difficult to fully understand.
Cherny acknowledged that human oversight remains key to mitigating these issues. Engineers at Anthropic could flag, debug, and escalate any suspicious or suboptimal behavior.
The Future of AI in Software Engineering
The revelation that AI writes 100% of code at Anthropic suggests a near-future where AI-first coding environments become the norm, not the exception. OpenAI, Anthropic, and other leaders in generative AI are proving that advanced models can handle full-stack software development, not just autocomplete functions or simple CLI tools.
Additionally, this advancement may fuel a new generation of AI-native startups, where lean teams can produce enterprise-grade platforms using AI as the primary developer.
Industries Most Likely to Adopt Full AI Coding
- Fintech: Speed and compliance-driven environments can benefit from accurate, auditable AI code.
- Healthcare Tech: AI can handle extensive documentation, streamlining regulatory coding burdens.
- Startups & SMBs: Small teams can scale faster through automated development.
- Enterprise SaaS: Quick iteration and feature rollouts become more feasible with AI-generated modules.
Conclusion: Toward a Co-Creative Future
Boris Cherny’s insights offer a glimpse into a world where human ingenuity and AI capability coalesce in powerful ways. At Anthropic, AI doesn’t just assist—it leads software creation. This paradigm challenges long-held assumptions about software production, crafting a new mold for digital builders.
While concerns about control, security, and job displacement remain, the benefits in speed, scale, and precision are undeniable. As AI continues to mature, engineering teams worldwide may have to rethink what it truly means to be a “coder” in the 21st century. The future of software isn’t just about writing code—it’s about designing systems where humans and AI collaborate in real time to build the next wave of technological innovation.