Kath Korevec

Kath Korevec is a member of product staff at OpenAI working on Codex, after leading development of Google Labs’ Jules coding agent. She builds tools that reduce the cognitive overhead of software development, from developer dashboards and documentation to autonomous systems capable of identifying and completing useful work.

From developer experience to coding agents

Korevec grew up in Alaska, experimented early with HTML, and redesigned a college library’s website around how visitors navigated its physical space. Before moving to San Francisco in 2006, she worked as a pastry chef and helped a car dealership publish its inventory online.

She subsequently held a front-end role at Google, became a technical strategist and architect at Hot Studio, and moved into product management at SAY Media and Disqus. At Heroku, she helped bring command-line capabilities into the company’s developer dashboard. As a senior director of product at GitHub, she led developer-documentation initiatives grounded in customer research. She later joined Vercel before returning to Google; her account of her career also describes her commitment to mentoring women in technology.

At Google Labs, Korevec led product work on Jules, launched in public beta in May 2025. The asynchronous coding agent works with existing repositories to tackle tasks including bug fixes and tests. Her AI Engineer Code appearance outlined a more ambitious model: agents that anticipate useful work while keeping developers in control.

Designing software that earns a developer’s attention

• Proactive coding agents: Repository-aware systems can identify missing tests, outdated dependencies, unsafe patterns, and unfinished tasks, then suggest fixes with visible rationales and confidence signals. Developers decide which work proceeds.

• Developer attention as a product constraint: Korevec’s case for proactive assistance rejects workflows that force people to continually prompt, monitor, and coordinate supposedly autonomous tools. Effective systems combine observation, personalization, timely intervention, and integration with existing development environments.

• Critic-augmented generation: She coauthored Google’s explanation of Jules’ adversarial code-review mechanism, which evaluates proposed patches, identifies implementation problems, and enables revision before human review.

• Human-controlled agent memory: Editable project memory, browser-based verification, explicit task controls, and connections between code, design, and behavioral data help agents develop context without displacing human judgment.

Korevec’s own projects include an office flip-disc display and an Atom extension for designing beaded earrings. Building firmware for an animatronic installation reinforced her practical standard for coding agents: they should absorb debugging and library maintenance so developers can concentrate on creative decisions. At OpenAI, she applies that sensibility to Codex and serves as an OpenAI Build Week judge.