Google Jules

Google Jules

Autonomous Coding Agent https://jules.google/ https://x.com/julesgoogle

Features

• Google’s asynchronous coding agent: you give it a task, it clones your repo into a cloud VM, makes a plan, works, and opens a PR for you to review/merge
• out of public beta (mid-2025), powered by Gemini 2.5 Pro then Gemini 3 Pro; plans scale with tier (Free ~15 tasks/day & 3 concurrent → Pro/Ultra up to 300 tasks/day & 60 concurrent)
• proactive features (Dec 2025): Suggested Tasks (scans your code and proposes improvements, e.g. #todos), Scheduled Tasks (cron-style recurring work), and a Render integration that auto-fixes failed deployments (reads logs → writes fix → opens PR)
• the 2026 research direction is “proactivity, not just autonomy” — Google Labs is benchmarking agents on an “insight policy” (when to surface a diagnostic vs stay silent), published as “Agentic Coding Needs Proactivity, Not Just Autonomy”

Summary

Jules is Google’s answer to Devin: a cloud VM agent you delegate bounded engineering tasks to and walk away from, with the PR as the handoff. Its differentiators in 2026 are (a) it ships with the Google AI Pro/Ultra subscription rather than as a standalone enterprise product, and (b) the proactivity push — Suggested/Scheduled tasks make it the “agent that watches your repo and keeps it healthy” rather than only the “agent that does what you ask.”

It’s deliberately lighter than Devin: no big enterprise VM fleet or VPC story, just clean async task→PR loops on top of Gemini. If your day is GitHub issues and you’re already on a Google subscription, Jules is the zero-friction delegation layer.

2026.09.08

Status: active — GA and moving to proactive. jules.google is live; the current frontier model is Gemini 3 Pro. The June 2026 research post (Measuring What Matters with Jules) shows Google is investing in goal-based evaluation for coding agents. No shutdown/rebrand signals — it’s a core Google Labs developer product.