When you write software every single day, you eventually need tools that handle real background tasks, offload heavy computing from your local machine, and process raw text files across different systems.
Rather than treating "AI" as a single chatbot window, modern developer infrastructure separates tasks by compute tier, autonomy level, and context size. Here is how Google's developer tools actually fit into my daily production workflow:
| Tool | Execution Environment | Daily Engineering Responsibility |
|---|---|---|
| Jules | Cloud VM / Background | Automating GitHub issues, dependency upgrades, running test suites, submitting PRs |
| Google Antigravity | Local Workspace / Swarms | Multi-file code refactors, Godot game scenes, parametric 3D CAD scripts |
| Google AI Studio | Web Dashboard & API | Token pricing modeling, prompt experimentation, instant zero-config prototype hosting |
| Google Colab | Hosted GPU/TPU VMs | Heavy numeric data processing, machine learning training, persistent Drive storage |
| Gemini Code Assist | In-Editor Plugin | Inline code completions, fast localized refactoring, documentation lookup |
| 2M Token Context | Long-Context Inference | Full-repo architectural audits, cross-file git commit history analysis |
When a repository piles up minor bug reports, missing test coverage, or outdated dependencies, I assign the issue to Jules so it can handle the tedious fixes without slowing down my local setup.
Assigning a GitHub issue to Jules starts an isolated Google Cloud virtual machine that clones the repository, checks the dependencies, writes the necessary fix, executes the test suite, and opens a clean pull request. With the 100 task runs per 24 hours included in the Google AI Pro plan, I can delegate routine background maintenance across several repositories at once, and I can trigger or inspect any of these tasks directly from my terminal with Jules Tools.
When a task spans across dozens of files at the same time, I open Google Antigravity across my workspace directory.
Rather than giving single-line completions, Antigravity starts multiple AI agents that work in parallel across the codebase. This approach is especially useful outside standard web development, since 3D CAD tools like FreeCAD and OpenSCAD use parametric code scripts, while game engines like Godot or Unity store scene structures, node trees, and project settings in plain text files.
The agents update 3D geometry scripts and game scene configs as easily as backend Python or TypeScript files, and I set up custom rules in Google Antigravity to review every proposed diff before applying changes.
Before writing actual application logic for an automation script, I test everything in Google AI Studio to figure out the exact token consumption and operating costs beforehand.
Tuning model parameters, testing system instructions, and checking exact token counts directly in the web UI prevents unexpected API bills before anything goes live. When I need to build a quick internal prototype or a small utility, AI Studio lets me publish a working web application straight from the dashboard, saving me from setting up web servers, hosting environments, or deployment pipelines.
I run resource-heavy Python scripts, large data processing tasks, and machine learning experiments in Google Colab rather than melting my local CPU.
Colab gives you hosted Jupyter notebooks backed by Google Cloud virtual machines with GPU and TPU access. By mounting my Google Drive with one simple command, the notebook can read raw data files and write persistent outputs directly to cloud storage, which keeps my laptop quiet and responsive while cloud servers handle the heavy lifting.
For small single-file edits, quick refactorings, or inline diffs inside VS Code or JetBrains IntelliJ IDEA, I rely on Gemini Code Assist right inside the code editor.
It provides immediate contextual suggestions and accelerates localized edits without requiring a full agent swarm or context switch away from the active buffer.
I mostly use Gemini Spark as an editing partner for drafting technical articles and structuring ideas, though it can also handle background workspace tasks.
When combined with the 2-million-token context window in Google AI Studio, I can paste an entire application codebase alongside git commit histories and full framework documentation into a single prompt to analyze cross-file dependencies and system relationships in one pass.