OpenAI president says AI is now writing 80% of the company’s code

OpenAI President Greg Brockman: AI Now Writes 80% of the Company’s Code

May 1, 2026 – 8:13 am

Greg Brockman, OpenAI president, stated during Sequoia Capital’s AI Ascent 2026 conference on Thursday that AI generates roughly 80% of the company’s code. (Business Insider)

"It’s hard to know what percent is not [written by AI]," Brockman acknowledged, echoing his previous remarks on the Knowledge Project podcast.

This comment fits into a broader narrative from AI lab leaders who highlight self-reinforcing productivity metrics. However, the evidence supporting AI coding efficiency remains contested.

Two Interpretations of the 80% Figure:

1. Lines of Code: AI tools generate 80% of all lines of code committed to OpenAI’s codebase.
2. Coding Involvement: AI is involved (through autocompletion, refactoring suggestions, or generation followed by human revision) in 80% of the coding work. Brockman’s qualifier suggests the latter interpretation aligns more closely with his statement.

A Consistent Pattern Across AI Labs:

Brockman isn’t alone in claiming high AI-coding figures. Similar statements have been made by:

• Dario Amodei, Anthropic CEO (90% of code written by AI)
• Cursor (achieved $2 billion in annual revenue within three years on AI-assisted coding)
• GitHub Copilot (4.7 million paid subscribers with 90% adoption among the Fortune 100)
• Anthropic ($30 billion run-rate revenue, concentrated in coding, enterprise search, and productivity).

These labs claim their AI models are transforming software engineering workflows.

Context and Criticism:

Brockman further explained in a Big Technology podcast interview (April 2026) that models have reached an "inflection point" where they can handle roughly 80% of typical engineering tasks, demanding a complete retooling of work processes around AI.

However, there is significant debate about the validity of internal AI-coding productivity claims. A February 2026 paper from the National Institute of Standards and Technology (NIST) raised concerns about:

• Overstating AI capabilities: Benchmarks may not accurately reflect real-world performance.
• Bias in data: Training data can perpetuate existing biases, leading to incomplete or inaccurate results.
• Limited generalizability: Models trained on specific tasks may not transfer effectively to new scenarios.