Move over, “Copilots.” The era of the AI Coding Agent is here. While traditional AI assistants wait for you to type a line of code so they can suggest the next, Jules is Google’s new autonomous teammate that handles entire engineering tickets from start to finish.
In this post, we’ll dive into the architecture of Jules and walk through a hands-on POC using Spring Boot to see how it handles real-world tasks.
What is Jules?
Jules isn’t just a chat window; it’s an asynchronous digital worker. It operates on a Perceive–Plan–Execute–Evaluate loop, running in a secure Google Cloud VM (Ubuntu) so your local environment stays clean.
The Core Architecture
Asynchronous “Fire and Forget”: Assign a task and go grab coffee. Jules works in the background.
Deep Context: Powered by Gemini 2.5/3 Pro, it reads your entire repo, including specialized AGENTS.md files for project-specific rules.
Secure Sandboxing: Every task runs in a dedicated VM to install dependencies and run tests safely.
Built-in Critic: An internal reasoning loop double-checks the logic before Jules ever presents a Pull Request to you.
Setting Up Your First Agent
Getting started is as simple as connecting your accounts:
Login: Use your Google Account.
Connect GitHub: Grant Jules access to your specific repositories.
Choose Your Interface: Use the Web UI for a visual experience, or the CLI/REST API for power-user workflows.
The 4 Flavors of Autonomy
Jules offers four distinct modes depending on how much control you want to retain:
Start (The Sprinter): No plan needed. Jules executes the prompt immediately based on its best judgment.
Review (The Guardrail): Jules generates a plan and waits for your “thumbs up” before touching the code.
Interactive Plan (The Architect): Jules creates a detailed, step-by-step roadmap for you to tweak before execution.
Schedule Task (The Night Owl): Set it and forget it. Jules takes full charge and executes tasks at a predefined time while you are away.
Hands-On POC: The Dog API Project
We created a repository called siddhu-jules-repo and put Jules to the test using a Spring Boot Maven application.
Phase 1: The “Start” Mode 
Prompt: “Create a maven springboot java application that will consume the Dog API and expose them as rest api. Response must be in JSON format.”
In this mode, Jules scaffolded the entire project, added the necessary dependencies (Spring Web), and created the controller/service layers to fetch data from https://dogapi.dog/api/v2/breeds.
Phase 2: The “Review” Mode 
Prompt: “Create another method that will consume the facts api and return JSON data.”
Jules analyzed the existing code, drafted a plan to add a new endpoint for https://dogapi.dog/api/v2/facts, and paused. We reviewed the diff, approved it, and Jules merged the changes.
Phase 3: “Interactive Plan” 
Prompt: “Modify the repo to add a rest api for fetching a random image of a dog using dog.ceo api.”
Here, Jules provided a deep breakdown of the file changes required. This is perfect for complex refactors where you want to ensure the AI isn’t hallucinating a weird architectural pattern.
Phase 4: “Schedule Task” 
Prompt: “Integrate the application to return response images based on server response codes using http.dog.”
We scheduled this task and walked away. Jules independently integrated the logic to map HTTP status codes to dog images (e.g., 404 results in a specific “not found” dog image) and completed the task on its own schedule.
Key Features You’ll Love
Audio Changelogs: Don’t want to read a wall of text? Jules can generate an audio summary of its commits.
Web Browsing: If it encounters a library it doesn’t know, it proactively searches the web for documentation.
Massive Scalability: You can run dozens of Jules agents simultaneously on one repo. Imagine a whole team of AI developers crushing your technical debt overnight! 
Resources
CLI Reference: jules.google/docs/cli/reference
API Reference: jules.google/docs/api/reference
Source Code:- https://github.com/shdhumale/siddhu-jules-repo.git
