My Experience Coding a Complex Project from the Sofa with Google's Jules
So, I Tried Coding with an AI from my Sofa
I've been hearing a lot about these AI coding assistants, and I decided to give Google's Jules a try. My goal was to see if I could actually build something complex without being glued to my desk. The idea of coding from my sofa sounded pretty great, so I jumped in.
What I Liked About It
The whole browser-based, asynchronous coding thing is actually pretty cool. Here are a few things that stood out to me:
- Real Flexibility: I wasn't chained to my main development machine. I could work on my laptop, and I even tried it on my tablet, which was surprisingly effective for small tasks.
- Working at My Own Pace: It felt less like a frantic sprint and more like a calm, focused session. I could think through a problem, give Jules a task, and then come back to it later. This was a game-changer for collaboration with a friend on a small side-project. We weren't constantly stepping on each other's toes.
- Focusing on the Fun Stuff: I could hand off the boring, repetitive tasks to Jules. This meant I could spend more time thinking about the bigger picture and the creative parts of the project.
The Workflow
My typical workflow with Jules looks like this:
- Give Jules access to a repo on GitHub: This is the starting point for any project.
- Write a prompt for that repo: I'll give Jules a clear and concise task to work on.
- Check the plan, approve or rework with an instruction (auto approval after 2min): Jules will create a plan, and I can either approve it or ask for changes.
- Check the code, rework with more instructions: I'll review the code and provide feedback for any necessary changes.
- Let Jules create a pull request: Once I'm happy with the code, Jules will create a pull request.
- Merge (and maybe delete the branch): I'll merge the pull request and, if I want, delete the branch.
Pricing
The beta period for Jules is over, and now you'll need a plan to use it. Here's a breakdown of the different options:
Plan | Jules | Jules in Pro | Jules in Ultra |
---|---|---|---|
Best For | Evaluating Jules on real work | Daily coding with higher intensity | Power users & agent-heavy workflows |
Daily Tasks (rolling 24 hours) | 15 | 100 | 300 |
Concurrent Tasks | 3 | 15 | 60 |
Model Access | Gemini 2.5 Pro | Higher access to the latest model (starting with Gemini 2.5 Pro) | Priority access to the latest model (starting with Gemini 2.5 Pro) |
You can find more details about the different plans and their limits on the Jules usage limits page.
Some Problems I Ran Into
It wasn't all perfect, though. I did hit a few snags:
- Understanding the AI: Sometimes, Jules would misunderstand my instructions, and I'd have to rephrase them a few times. It's like learning to communicate with a new team member who's super smart but also very literal.
- Debugging Can Be Tricky: When the AI-generated code had a bug, it was sometimes harder to debug than my own code. I had to spend extra time understanding what Jules was trying to do before I could fix it.
My Tips for Working with Jules
If you're thinking of trying out Jules, here are a few things I learned that might help:
- Start Small: Don't throw a massive, complex project at it right away. Start with a small, well-defined task to get a feel for how it works.
- Be Super Clear: The more specific and detailed your instructions are, the better the results will be. Think of it as writing a very detailed ticket for a junior developer.
- Give it Project-Specific Instructions: I found it really helpful to have Jules create an
AGENTS.md
file. This is like a set of ground rules for the AI. It helps Jules to understand the project faster and make less mistakes, which helped keep the project consistent. - Bundle Your Tasks: To get the most out of it, I'd group a few related tasks into one request. For example, "implement this feature, write tests for it, and add a quick note to the README." This is also a good way to work around the daily task limits of Jules.
- Use it for Reviews: I even asked Jules to review my code and suggest improvements. It's like having a second pair of eyes, and it caught a few things I missed.
All in all, it was a pretty positive experience. It's not going to replace human developers anytime soon, but it's a powerful tool that can definitely change how you work. I'm still a fan of coding from the sofa.
Happy Coding!