OpenAI President Reports AI Writing Up to 80% of Code
According to Business Insider, at a Sequoia Capital talk uploaded Thursday, OpenAI president Greg Brockman said, "If you look even over the course of December, we went from these agentic coding tools writing 20% of your code to writing 80% of your code." Brockman also told the audience that Codex, OpenAI's code-generation platform, has broadened from a developer tool to support "anyone who's doing work with a computer," Business Insider reports. The coverage includes counterpoints: Business Insider separately reported former Tesla AI head Andrej Karpathy saying AI-written code can be "bloaty" and "brittle," and still needs human oversight. The comments together frame rapid adoption claims alongside persistent quality and safety concerns documented in the same reporting.
What happened
According to Business Insider, at a Sequoia Capital talk uploaded Thursday, OpenAI president Greg Brockman said, "If you look even over the course of December, we went from these agentic coding tools writing 20% of your code to writing 80% of your code." Business Insider also reports Brockman saying that Codex, OpenAI's code-generation platform, has recently evolved from a tool primarily for software engineers to something that can support "anyone who's doing work with a computer." Business Insider separately reports that Andrej Karpathy said AI-written code is often "bloaty" and "brittle" and that humans still need to provide oversight.
Editorial analysis - technical context
Industry-pattern observations: Rapid improvements in code-generation models and agentic tools tend to shift developer time away from boilerplate implementation toward integration, review, and specification. Observers note common failure modes in AI-generated code include redundant or copy-pasted blocks, fragile abstractions, and non-idiomatic patterns, all of which increase the need for human-led validation, testing, and security review.
Context and significance
Editorial analysis: If agentic tools are producing a substantially larger share of code in practice, that alters where engineering value is created, from hand-writing routine logic toward designing prompts, writing tests, and specifying interfaces. At the same time, reported quality issues mean organizations must balance throughput gains against maintenance costs, security posture, and developer productivity metrics.
What to watch
Industry-pattern observations: Practitioners and observers should track three indicators: adoption metrics for Codex and competing tools; prevalence of AI-originated code in codebases as measured by provenance or attribution tooling; and changes in bug/incident patterns tied to AI-generated commits. Also watch for tooling that automates validation (static analysis tuned for AI outputs, specialized test-generation) and for policy or security guidance from major platforms.
Limitations of the reporting
What was reported above is drawn from Business Insider coverage of the Sequoia Capital talks. The reporting quotes Brockman and Karpathy directly; Business Insider does not provide independent quantitative telemetry in the article to verify the "80%" figure beyond Brockman's statement.
Scoring Rationale
Claims that AI now writes a majority of code indicate a notable acceleration in developer tooling and workflows, affecting testing, security, and productivity. The story pairs high-impact adoption claims with practitioner-quality concerns, making it important for engineers and platform teams.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems

