OpenAI President Greg Brockman told an audience at Sequoia Capital this week that agentic coding tools have made a remarkable leap in capability—jumping from writing roughly 20% of code to about 80% over the course of December 2025. He framed the shift as one that pushes AI from a useful convenience into something developers genuinely have to work around.

From Copilot to Primary Author

Speaking at Sequoia’s AI Ascent 2026 conference, Brockman said the December milestone effectively forced a re-think of how software is being built. He explained that recent model releases pushed AI capabilities from handling around 20% of typical engineering tasks to roughly 80%, a swing significant enough that teams now need to redesign their workflows around AI rather than simply bolting it on.

In an extended interview with Big Technology, Brockman credited the jump to incremental gains in OpenAI’s base models and post-training pipeline rather than a single dramatic breakthrough. He pointed out that the move “wasn’t from 0% to 80%” but from 20% to 80%, characterizing it as steady improvement that crossed a tipping point. As an example, he described one of his engineers who, until recently, couldn’t reliably hand off low-level systems work to AI; that same engineer can now pass a design document to the model and get production-ready code in return.

Codex Reaches Beyond Code

Brockman also hinted that OpenAI’s Codex platform is being repositioned as something much broader than a developer tool. The underlying technology, he argued, is fundamentally about solving problems rather than generating code, and it’s already being applied to spreadsheets, presentations, and general knowledge work. In a February post on X, he had laid out an internal goal at OpenAI to make agents the “tool of first resort” for all technical tasks by March 31.

An Industry-Wide Trend

Brockman’s comments line up with a similar disclosure from Google CEO Sundar Pichai, who said in an April blog post that 75% of all new code at Google is now AI-generated and reviewed by engineers—up from 50% last fall and around 25% in late 2024. Pichai said Google is shifting to “truly agentic workflows” where engineers orchestrate “fully autonomous digital task forces.” According to Semafor, Meta is reportedly aiming for similar levels by mid-year.

The fact that these numbers are converging so quickly across major tech companies suggests that AI-assisted coding has moved from experimental augmentation to default practice. It also raises pointed questions about how the role of the software engineer is evolving in a world where machines write most of the code and humans increasingly act as reviewers, architects, and orchestrators.