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

The most important part of a specialized conference isn't the talks, which are typically recorded, but the 'hallway track'—the unstructured conversations with speakers and other expert attendees. Maximizing this value requires intentionality and a clear goal for engagement, as these serendipitous connections are the primary reason to attend in person.

Early on, Google's Jules team built complex scaffolding with numerous sub-agents to compensate for model weaknesses. As models like Gemini improved, they found that simpler architectures performed better and were easier to maintain. The complex scaffolding was a temporary crutch, not a sustainable long-term solution.

The trend of 'vibe coding'—casually using prompts to generate code without rigor—is creating low-quality, unmaintainable software. The AI engineering community has reached its limit with this approach and is actively searching for a new development paradigm that marries AI's speed with traditional engineering's craft and reliability.

Coding is a unique domain that severely tests LLM capabilities. Unlike other use cases, it involves extremely long-running sessions (up to 30 days for a single task), massive context accumulation from files and command outputs, and requires high precision, making it a key driver for core model research.

Borovik's realization came from observing artists' split reaction to Stable Diffusion—fear versus embracing it as a new tool. He saw a direct parallel for software engineering, deciding AI was a tool to enhance his craft, not replace it, which spurred his move into building coding agents at Google.

Increased developer productivity from AI won't lead to fewer jobs. Instead, it mirrors the Jevons paradox seen with electricity: as building software becomes cheaper and faster, the demand for it will dramatically increase. This boosts investment in new projects and ultimately grows the entire software engineering industry.

For a coding agent to be genuinely autonomous, it cannot just run in a user's local workspace. Google's Jules agent is designed with its own dedicated cloud environment. This architecture allows it to execute complex, multi-day tasks independently, a key differentiator from agents that require a user's machine to be active.

Embedding-based RAG for code search is falling out of favor because its arbitrary chunking often fails to capture full semantic context. Simpler, more direct approaches like agent-based search using tools like `grep` are proving more reliable and scalable for retrieving relevant code without the maintenance overhead of embeddings.