‘I Don’t Write Code Anymore’: Anthropic, OpenAI Engineers Say AI Now Writes 100% of Their Code

Why It Matters

Full‑code generation could dramatically cut development cycles and reshape software talent demand, accelerating AI‑centric product delivery across the tech sector.

Key Takeaways

• Anthropic and OpenAI claim AI writes all production code.
• Engineers report shipping dozens of PRs generated entirely by LLMs.
• CEO predicts industry-wide AI coding within 6‑12 months.
• AI still produces dead code; engineers use self‑review tools.
• Hiring shifts toward AI‑oriented generalists over language specialists.

Pulse Analysis

The rise of large language models such as Claude Code and OpenAI’s own Codex has moved code generation from assistive suggestions to end‑to‑end production. Early adopters at Anthropic and OpenAI now treat the model as a virtual developer, feeding high‑level specifications and receiving complete pull requests in minutes. This leap mirrors the evolution of AI‑assisted tools that began with autocomplete and has been accelerated by improvements in model reasoning, context windows, and integration with version‑control pipelines, enabling teams to iterate faster than ever before.

For the workforce, the shift rewrites the traditional apprenticeship model. Junior engineers, who once learned by writing and debugging code, now need skills in prompt engineering, model supervision, and architectural oversight. Companies are already favoring adaptable generalists who can steer AI outputs rather than specialists locked into a single language stack. The rapid adoption also exposes a training gap: while 89% of employees use AI at work, only a third receive formal guidance, raising concerns about code quality, security, and long‑term skill erosion.

Looking ahead, industry leaders predict a six‑to‑twelve‑month horizon before AI handles most software engineering tasks across enterprises. Adoption will likely be uneven, with regulated sectors demanding rigorous validation and open‑source projects experimenting with AI‑first workflows. Key challenges include mitigating hallucinated logic, ensuring compliance with licensing, and integrating AI‑generated code into existing CI/CD pipelines. Organizations that invest early in AI governance, upskill their staff, and embed model‑review loops will capture the productivity upside while safeguarding against the hidden costs of automated code.