⚡ [AIE CODE Preview] Inside Google Labs: Building The Gemini Coding Agent — Jed Borovik, Jules

Jed Borovik’s story begins with the release of Stable Diffusion, which he describes as his “first Gen‑AI moment” and the catalyst for moving from search freshness work to building AI‑powered coding tools. After nine years at Google, he joined Google Labs, an organization tasked with creating products that sit outside Google’s core offerings. Labs operates as a true AI product org, tightly coupled with DeepMind for model development while also leveraging Google’s massive internal data and infrastructure. This unique position lets the team take ideas from raw pixels through training to a finished developer experience.

The flagship of that effort is Jules, an autonomous coding agent built on the Gemini model family. Unlike typical plug‑in assistants, Jules provisions its own compute environment, allowing it to run for hours or days and perform complex, multi‑step refactorings without constant user supervision. The agent is exposed through a REST API, a dedicated CLI, and now integrates with the Gemini CLI, making it reachable from any developer workflow. Real‑world usage includes triggering Jules from GitHub Actions to automatically generate and merge pull requests, demonstrating how an ambient AI can become a seamless part of the software delivery pipeline.

Early versions of the system relied on heavy scaffolding and traditional RAG pipelines that stitched together embeddings, chunking, and external tools. As Gemini models improved, the team discovered that “less is more”: simpler prompts and fewer sub‑agents achieved higher reliability, and alternative retrieval methods like semantic‑aware grep replaced brittle embedding chunks. The transition from a preview in the Trusted Testers program to a production‑grade product was marked by strong reception at events such as the AIE Code Summit, where AI engineers gather to exchange ideas beyond academic conferences. The summit underscores the growing need for industry‑neutral venues that accelerate AI engineering collaboration.