# Jules: reviews and analysis > Google's asynchronous agent: leave it tasks and it brings back finished changes. - Canonical: https://serchai.com/en/reviews/jules/ - Site: Serchai (https://serchai.com) — AI tools comparator - Language: en - Updated: 2026-07-25 --- ## Verdict Jules works asynchronously: you assign tasks from the web, it runs them in its own environment with your repository and presents the changes as proposals. The free plan with daily tasks makes it the cheapest way to try asynchronous delegation. On autonomy and judgment it trails the front-line agents, and ambiguous tasks trip it up. **Best for:** Anyone delegating small, well-defined tasks while doing something else. **Rating:** 3.8/5 ## Pros - Background tasks without occupying your machine - Free plan with daily tasks - Changes presented as reviewable proposals ## Cons - Less judgment than leaders on ambiguous tasks - Paid tiers live inside Google subscriptions ## Key facts - Price: Free - Free trial: No - Platforms: Web - Categories: [Code](https://serchai.com/en/best-ai/code/) - Official website: https://jules.google ## What the internet says (agentic sweep) Its model is asynchronous work: you hand it a task or an issue, it goes off to a cloud machine and comes back with a reviewable pull request. The free plan is among the most generous in the catalogue at 15 tasks a day, and the complaints concentrate on short context in large repos and uneven results next to interactive agents. - Sweep date: 2026-07-25 - Derived score: 3.8/5 ### Axes - Results: 3.7/5 - Control: 3.3/5 - Real price: 4.3/5 - Integration: 4.1/5 - Support: 3.5/5 ### Recurring themes in favor - A free plan with 15 tasks a day and three in parallel, enough to use it seriously without paying (strong theme) - It works on its own: clones the repo, runs the tests, iterates when they fail and opens the pull request while you do something else (strong theme) - It fits the GitHub issue and pull request flow, and has an API and a command line for automating it (present theme) ### Recurring themes against - Context runs short on large repositories, and the quality of its proposals visibly drops there (strong theme) - Output is uneven: the community compares it to a junior colleague whose work needs full review (strong theme) - Paid plan prices do not appear on the product site, which publishes only task quotas per day (present theme) ### Sweep sources - [Docs] https://jules.google/ — Web oficial con las cuotas de tareas por día de cada plan - [Communities] https://news.ycombinator.com/item?id=44813854 — Presentación del agente con 227 comentarios de la comunidad - [Communities] https://news.ycombinator.com/item?id=45466588 — Anuncio de la API con 76 comentarios sobre uso real - [Communities] https://news.ycombinator.com/item?id=48764451 — Pregunta abierta sobre quién lo usa de verdad y con qué resultado > **TLDR:** Jules works asynchronously: you assign tasks from the web, it runs them in its own environment on a copy of your repository and presents the changes as reviewable proposals. Its free plan with daily tasks is the cheapest way to try asynchronous delegation. On judgment it trails the front-line agents, so its ground is small, well-defined tasks while you do something else. ## What Jules is and how it works Jules is Google's bet on a specific working model: the agent you do not supervise while it works. Instead of the live conversation of terminal agents, the flow here is commission-based: you connect your repository, describe the task, and Jules takes it to its own virtual environment, where it clones the project, drafts a plan, executes it and returns with changes ready to review. The difference from a supervised agent is one of attention, not just architecture: Jules is built for leaving tasks like errands, several running in parallel, with review whenever suits you. The plan appears before execution, so you can correct the approach without spending a whole run. Underneath run the Gemini models, and around it sits the Google ecosystem, of which Jules is part with the free tier as the entry door. ## What it is like day to day The use that works is the errand list: the small bugs, the dependency upgrade, the missing test, the documentation tweak. Assigned in the morning, they work alone, and reviewing the proposals fits into an afternoon gap. For that queue of tasks that never wins a slot, a free agent quietly working through them in the background is exactly the tool. Calibration arrives fast: ambiguous or long tasks come back with detours that demand redoing, and the agent's judgment sits a step (sometimes two) below [Claude Code](https://serchai.com/en/reviews/claude-code/) or [Codex](https://serchai.com/en/reviews/openai-codex/), whose cloud agents play this same format with more accuracy. The practical answer is size: the smaller and more verifiable the task, the better Jules performs. Reviewing the plan before approving execution is the habit that adds the most yield: correcting a sentence in the plan costs seconds, correcting a crooked execution costs the whole task. ## Pricing and plans The free plan gives a handful of daily tasks with limited concurrency, enough for an individual developer's errand flow. The paid tiers, folded into Google's AI subscriptions, multiply daily tasks and concurrency for intensive use. The math is simple: free for trying asynchronous delegation and for the daily drip of small tasks. If the pattern works and you want more judgment, the natural jump is not paying for more Jules but trying Codex's cloud agents or [Devin](https://serchai.com/en/reviews/devin/)'s flow, which play the same card with more autonomy. ## Who it is for (and who it is not for) Jules is for developers with a queue of small tasks and an appetite to try delegation without spending: personal projects, maintenance, that backlog of minor improvements that never competes well for your attention. Also as first contact with the asynchronous format before deciding whether to pay for it. It is not for ambiguous tasks, nor for anyone already paying a subscription with a cloud agent included, which covers this ground with more judgment. ## Jules alternatives Codex's cloud agents are the same format with more accuracy, inside the ChatGPT subscription. Devin is the professional version of the ticket-based autonomous engineer. And [Gemini CLI](https://serchai.com/en/reviews/gemini-cli/) is the supervised sibling from the same free ecosystem. They are in [Jules alternatives](https://serchai.com/en/alternatives/jules/) and the ranking in the [best AI coding tools](https://serchai.com/en/best-ai/code/). ## Frequently Asked Questions ### What does asynchronous mean here? That you do not supervise it while it works: you assign the task, it runs in its own cloud environment and presents the changes when done. Your attention only enters at commissioning and at review. ### Is the free plan good for anything? Yes: the handful of daily tasks covers an individual developer's errand drip, and it is the cheapest way that exists to test whether the asynchronous delegation format fits you. ### Which tasks suit it? Small, verifiable ones: specific bugs, dependencies, missing tests, documentation tweaks. Ambiguous ones come back with detours, and there the front-line agents earn their cost. ### Jules or Codex's cloud agents? Same format, different league: Codex lands more medium tasks and ships inside paid ChatGPT. Jules wins on price (free) and loses on judgment. Starting with Jules and moving up if the pattern works is a reasonable path. ## Alternatives - [Devin](https://serchai.com/en/reviews/devin/) — Cognition's autonomous engineer, now across four surfaces: Desktop (the former Windsurf), Cloud, CLI and Review. (3.4/5) - [OpenAI Codex](https://serchai.com/en/reviews/openai-codex/) — OpenAI's coding agent, included with ChatGPT: local in your terminal or in the cloud. (4.1/5) - [Gemini CLI](https://serchai.com/en/reviews/gemini-cli/) — Google's terminal agent, open source with a generous free tier. (3.3/5)