Jed Borovik

Jed Borovik worked on Jules, Google’s autonomous coding agent, as a Product Lead at Google Labs. In a November 2025 Latent Space interview, he explained the team’s direction: build for increasingly capable agents that can work independently, and simplify the surrounding software as Gemini models improve.[1][3]

Borovik described a separate computer and working environment as central to Jules’ autonomy. An agent working alongside a developer locally shares that developer’s workspace; an independent environment supports the team’s goal of agents working for hours or days on their own.[3]

He also described how giving Jules its own infrastructure and an API enabled work triggered by events. One example he recalled involved a user triggering Jules to update a website, then using a GitHub Action to merge its pull requests automatically. Borovik framed the API and command-line interface as ways to make the agent available wherever developers already worked.[3]

Borovik described Google Labs’ contribution as product development in close collaboration with DeepMind. That collaboration spanned interfaces, infrastructure, models and training, allowing the team to work across the full AI product.[3]

According to Borovik, early Gemini models required complex agent scaffolding—the software coordinating the model’s work. The Jules team experimented with specialized subagents for code edits, subproblems, integrations and review. As models improved, he found that less scaffolding was needed to compensate for their limitations. He distinguished those experiments from elaborate systems of simulated professional personas that the team had not pursued, and emphasized that a simpler harness could still be sophisticated.[3]

His discussion of code retrieval followed the same concern with complexity. Borovik highlighted the difficulty of maintaining embeddings and choosing how to divide code into chunks, and discussed agent-based search as approaches evolved beyond treating embedding-based retrieval as the default. He described retrieval as active research for the Jules team.[3]

Before joining Google Labs and Jules, Borovik worked on Google Search, particularly news and freshness. His recollection of that earlier period explains the career choice behind his coding-agent work: while still on Search, he experimented with agents in his own time and sought a role where he could build them professionally.[3]

Borovik recalled Stable Diffusion as his first significant encounter with generative AI. Watching creators react either as though their craft was threatened or as though they had gained a tool for better work, he applied that choice to programming. Conversations questioning whether students should study computer science prompted him to choose the creative-tool perspective and pursue coding agents.[3]

The AI Engineer Code Summit 2025 archive lists Borovik as an emcee with a Google DeepMind affiliation. In the interview, he welcomed major technology conferences in New York and recalled the city’s student hackathons as a place where he developed his enthusiasm for programming.[2][3]