Last updated on 25 Feb 2026
Minor update on 13 Mar 2026
Related post: Index for posts on Feb 2026 notes about Google AI Pro plan features for software development work. It has section links (jump links) for posts that have them.
Summary
With my recent Jio 4G mobile recharge, I got a free 18 month offer for Google AI Pro which I accepted. Google AI Pro includes Google Jules being used as a coding agent. As a trial, for a very simple Astro landing page site, I gave Google Jules this prompt:
- I have a dummy Indian rural school website data in this Astro app. I would like the data in Features and About sections to be changed to a hypothetical (sample) rural Indian school (primary & secondary level/K-12).
It did the job well! While the programming task I asked Google Jules to do was very simple, it had to go through all the steps of:
- Studying project code (multiple files)
- Modifying the correct components with suitable generated text for Features and About sections which earlier had dummy content
- Building the app
- Running it and verifying the output
- Creating a GitHub PR request
~4 min May 2025 video showing how Jules provides agentic coding capability:
===========================================================
===========================================================
Details
04:28, 24 February 2026
As part of my recent Jio 4G mobile recharge, I got a free offer for Google AI Pro plan which I accepted.
https://www.jio.com/google-gemini-offer/ : "Eligible users can enjoy full 18 months of Google AI Pro subscription from activation, (strike-through for Rs35,100) free." The link says it is for 5G users but I got the offer even though I am on 4G.
Jules coding agent is one of the features of the Google AI Pro plan.
https://jules.google/docs/: "Jules is an experimental coding agent that helps you fix bugs, add documentation, and build new features. It integrates with GitHub, understands your codebase, and works autonomously — so you can move on while it handles the task."
Note that Jules seems to be also available free:
Jules Plans. However, the free Jules plan is advertised as best for "Evaluating Jules on real work" whereas "Jules in Pro" is best for "Daily coding with higher intensity". Perhaps even in free tier plan, Jules will work as a coding agent but with significant limits.
I am trying (Jules on AI Pro) out. It took some time to get the full repo list from my GitHub a/c (a repo I wanted to work on seemed to get listed in Jules only after 15 minutes to half an hour after giving GitHub access).
It seems to be an agentic coding system/setup like Claude Code.
Quite simple prompt I specified for a private repo (ssw-astro):
I have a dummy Indian rural school website data in this Astro app. I would like the data in Features and About sections to be changed to a hypothetical (sample) rural Indian school (primary & secondary level/K-12).
Note that Jules works asynchronously. So after I gave the above prompt I moved to some other activity.
Within perhaps 10 to 15 minutes, I got a notification that Jules has prepared a plan.
After I approved the plan, within a short time (perhaps 15 minutes or so), I got a notificaton that Jules has done the work!
Jules showed me modified code in two astro components, Jules reported that it:
"Ran npm install && npm run build.
Build was successful.
Updated verification/about_section.png and verification/verify_changes.py
Updated verification/about_section.png verification/features_section.png and verification/verify_changes.py"
[These seem to involve Playwright testing tool. If so, that's awesome!]
It said, "I've inspected the frontend changes visually:" and showed me 2 screenshots of site with changes.
It said, "Running code review" and gave details of what it did for that.
It created a PR which it said is "Ready for review". [It also showed Time: 7 mins. Does that mean it did the work in 7 mins? Probably.]
There was a View PR button which I clicked. It took me to GitHub. It had added the changes in a new branch and I had the option to merge it to main.
I first wanted to pull that branch to my local repo.
ssw-astro> git fetch origin
ssw-astro> git branch
* main
ssw-astro> git branch -r
origin/main
origin/update-about-and-features-content-468962095223529504
ssw-astro> git checkout update-about-and-features-content-468962095223529504
branch 'update-about-and-features-content-468962095223529504' set up to track 'origin/update-about-and-features-content-468962095223529504'.
Switched to a new branch 'update-about-and-features-content-468962095223529504'
ssw-astro>
Me: Now should I do a git pull to pull the remote branch data to my local repo?
GSAI (Google Search AI): Actually, no—you are already good to go!
When you ran git checkout update-about-and-features-content-468962095223529504, Git performed a "guess" and found that the branch existed on your remote. Because of this, it automatically:
Created a local branch with that exact name.
Linked it to track the remote branch (origin).
Downloaded the files and commit history from the remote to your local machine.
Your terminal confirmation "Switched to a new branch..." means your local working directory now matches exactly what Jules created on the server.
==============
I checked and yes, my local VSCode editor showed me the changes made by Jules!
npm run dev
The app ran and I could see the added website sections!
Astonishing! This is my first agentic coding (system) experience. I am dazzled that I could get this via a free offer from Jio (for 18 months) for my usual Rs.666 70 days 1.5 GB/day 4G mobile recharge plan.
I saw that I have the Playwright .py verification script and associated screenshots as well. Wow! Testing is the hard thing in such stuff. Will check out the test script in detail later. (Don't know python but will figure that out to extent used in script - perhaps will just ask an AI tool to explain it to me).
Me: The changes are good. It works on my local PC. I am ready to merge the PR. Shall I do the merge PR on GitHub itself and then pull in changes from remote to local repo?
GSAI: Yes, merging the Pull Request (PR) directly on GitHub is the standard and cleanest way to proceed.
Since you have already verified the changes locally, follow this workflow to finalize everything:
1. Merge on GitHub
- Navigate to your repository on GitHub.
- Open the PR created by Jules.
- Click the green Merge pull request button (or Squash and merge if you want to combine Jules's commits into one).
- Confirm the merge.
- Optional: Delete the remote branch on GitHub using the button provided after merging to keep things tidy.
2. Update Your Local Repo
Once the PR is merged into your main branch (or your default branch) on the remote server, you need to bring those changes down to your local machine:
- Switch to your main branch:
- Pull the latest changes:
3. Cleanup Local Branch (Optional)
Since the work is now part of main, you can safely delete the temporary branch Jules created to keep your local environment clean:
- git branch -d update-about-and-features-content-468962095223529504
====================
Merged PR on GitHub but did not delete the update-about-and-features-content-468962095223529504 branch.
git checkout main
git pull
That pulled in the update on main branch to local repo.
npm run build:ghp
succeeded
ssw-astro> npm run deploy:ghp
> ssw-astro@0.0.1 deploy:ghp
> npx gh-pages --nojekyll -d dist -r https://github.com/ravisiyer/ssw-astroghp
Published
ssw-astro>
The (very simple) site is showing correctly on both PC and my mobile phone. It is responsive and light/dark mode works as expected.
Hmm. While the programming task I asked Google Jules to do was very simple, it had to go through all the steps of:
- Studying project code (multiple files)
- Modifying the correct components with suitable generated text for Features and About sections
- which earlier had dummy content:
- "Feature One" "Some description of the first feature." etc.
- "This is a dummy About section for your landing page. You can replace this with actual content later."
- Building the app
- Running it and verifying the output
- Creating a GitHub PR request
That's an agentic coding workflow, I presume (I am very new to agentic coding and so may not be getting some terms right).
Wow! Without paying any extra money, I now have an agentic coding system to try out! But Google Jules is experimental as per its own docs:
https://jules.google/docs/ . So it may not be in the same class as Claude Code with Opus 4.5/4.6 or Open AI Codex.
It is when I try Jules out with medium complexity programming work that I will get a real feel of how good it is.
For this very simple but still fully agentic coding cycle that I tried out, Jules seems to have done a very effective job. I repeat that I am particularly impressed by its verification. However, I need to study that code to know how thoroughly it did the verification.
While I may have come across Google Jules coding agent in the past, it had not registered in my mind. Claude Code and Codex were the two coding agents that had registered in my mind over the past few weeks/months. Browsed the net and found this May 2025 article (quite old given the pace of AI improvement): Codex, Jules, and Claude Code comparison,
https://www.jonatkinson.co.uk/blog/codex-jules-claude-comparison/ which is a good and to-the-point comparison article. He says that (then) Google Jules was a "highly beta" product.
Jules is out of beta!,
https://jules.google/docs/changelog/2025-08-06/ , Aug. 2025 says, "Today we are thrilled to announce that Jules is no longer in beta!" ... But, as mentioned earlier, the docs site
https://jules.google/docs/ (currently in Feb 2026) says Jules is experimental. So I guess Google Jules cannot be viewed today to be in same class as Claude Code and Open AI Codex. But that might change quickly.
ChatGPT (CG) chat: Google Jules Overview
Me: I recently came to know of Google Jules coding agent. I had heard a lot about Claude Code and OpenAI Codex but not Google Jules. What do u think about Google Jules?
CG: Here’s a clear, up-to-date overview of Google’s AI coding agent Jules (how it works and how it compares to tools like Claude Code or OpenAI Codex):
Meet Jules – The AI Coding Agent by Google
(Introductory video explaining what Jules is)
🔍 What Jules Is
Jules is an autonomous AI coding agent developed by Google — part of its AI portfolio powered by the Gemini 2.5 Pro / Gemini 3 Pro large language models. Unlike traditional autocomplete/coding assistants, Jules is designed to act on your codebase asynchronously, meaning:
You assign it tasks (e.g., “add tests,” “fix this bug,” “upgrade dependencies”).
It clones your GitHub repository into a secure Google Cloud virtual machine (VM).
It plans, executes, and proposes code changes independently.
Final changes are shown as a diff + plan + reasoning and usually submitted as a pull request you can review and merge. (blog.google)
So rather than waiting for interactive chat results, you launch tasks and come back when they’re done — a background development assistant. (TechCrunch)
🧠 Core Capabilities
✅ Autonomous Task Execution
Jules can:
Fix bugs
Write unit tests
Build new features
Update dependencies
Generate changelogs (including audio summaries)
Integrate with GitHub Issues and create branches/PRs
Work on multiple tasks in parallel (blog.google)
🔁 Asynchronous & Parallel Work
Unlike interactive AI coding tools (like auto-complete engines), Jules can queue and run tasks in the cloud while you continue other work. (blog.google)
🔒 Privacy & Security
By default it runs in isolated VMs and does not train on your private code — Google explicitly states that private repositories aren’t used for training. (blog.google)
💡 How Is Jules Different from Claude Code / Codex / Copilot?
| Feature | Jules | Codex / GitHub Copilot | Claude Code |
|---|
| Interaction Style | Asynchronous background agent | Synchronous autocomplete / chat | Synchronous code assistant with planning |
| Execution | Runs tasks independently in VM | Suggests code in editor | Suggests code snippets |
| Workflow Integration | GitHub PRs, command line tools, APIs | Editor extensions | Editor/chat interface |
| Task Parallelism | Yes | No | No |
| Autonomy | High (agentic) | Low to medium | Medium |
| Pricing | Free tier + paid tiers | Often usage-based | Usage plan dependent |
Jules is more of an agent executing tasks than a real-time suggestion engine, which is the core distinction from tools like Codex (which focus on in-editor completion) or Claude Code (which is stronger in interactive reasoning but usually still synchronous). (blog.google)
📈 Real-World Adoption & Current State
Launch & Availability
Developer Feedback (Mixed)
Many developers find Jules promising for routine tasks and background automation.
Some community reports indicate stability issues, slower execution, or inconsistent output quality on complex technical tasks. (Reddit)
Ecosystem Growth
🧠 When Jules Makes Sense
Use Jules if you want a tool that:
Frees you from repetitive development tasks (tests, upgrades, bug fixes)
Works in the background without manual prompting
Integrates directly with GitHub and typical dev workflows
Supports automation and multi-task pipelines
It’s particularly attractive if you already use GitHub heavily or want a more agentic development assistant. However, for real-time interactive coding help or deeper conversational reasoning, tools like Claude Code or editor-integrated AI assistants remain valuable complements.
🧩 Summary
Google Jules is one of the first mainstream autonomous AI coding agents aimed at practical, task-oriented software development. It’s not just a code completer — it’s designed to execute, plan, and manage coding work on your behalf, and represents a shift toward agent-led automation in programming. (blog.google)
If you’d like, I can compare Jules, Claude Code, and OpenAI Codex side-by-side in terms of specific real-world coding workflows (e.g., bug fixing, feature building, testing).
Comments
Post a Comment