Gemma Lara Savill
A snapshot of AI in my day-to-day workflow
By Gemma Lara Savill
Published at October 11, 2025
My journey into Generative AI is leading me to my next big project: building a custom chatbot from the ground up. Before I get started, I wanted to hit pause and reflect on how AI is already a quiet partner in my daily life as a developer.
Gemini Inside the IDE
Being an Android Developer, I now have Gemini Chat and Gemini Agent (preview) integrated right into Android Studio.
I use Gemini Chat constantly - to refine the wording of Jira tickets, write clearer PR descriptions, or even generate tidy commit messages. It's handy having it right there in the IDE.
I've also added a few custom rules for Gemini inside Android Studio. They describe things like my architecture preferences, how I like to keep code clean, and using Gherkin-style names for tests.
It gets a bit strange when Gemini repeats those rules at the beginning of every chat - I imagine they're prepended to the prompt. I've also noticed that when the context window runs short, some of those rules seem to drop off.
Still, I really like the idea. It lets me customize the assistant's behavior to match my development style, and that makes the suggestions feel more relevant.
Gemini Agent, on the other hand, I haven't used much yet. I tried it a few times, but found it a bit slow and the window too small.
Lately, I've been wondering if part of that is simply where the Agent window sits inside Android Studio. I usually keep it docked on the side, which means it competes for space with my code. I might experiment with placing it below the editor instead - maybe window placement has more to do with how we adapt to new tools than we realize.
I can't wait to use Gemini with Compose, but since I haven't worked on UI features recently, that one might have to wait for a personal project. There are some built-in features that can really boost my productivity, plus I LOVE Compose.
Discovering Gemini CLI
The real game changer for me has been Gemini CLI in the Android Studio terminal. I have it installed globally so I can use it in the terminal anywhere, but in the terminal right there in the IDE is just right.
It can access the full project context. To make sure the context is clear you can just run:
geminito start Gemini CLI, and then
/initnow you can review the generated GEMINI.md, tweak it, and keep it handy for deep-work sessions. I usually stash that file away and pull it out when I need to focus deeply on a project.
For me, Gemini CLI feels far more powerful than the built-in Gemini in Android Studio. It's backed by Gemini Pro with a massive context window, and that really shows.
Everyday Uses
Gemini CLI has become my go-to for debugging and analysis. I've used it to analyze logs, investigate build errors, as it can even run Gradle commands internally to check builds and tests.
It can also run git commands - although that makes me slightly nervous, so I've avoided that feature locally for now.
Gemini as PR Reviewer
At work, we're experimenting with Gemini AI as a PR reviewer in our main Android repo. It's only been a week, but it genuinely makes me smile.
We haven't fine-tuned its behavior yet; we're all still getting used to it. At first it felt verbose, even overwhelming - one teammate even marked one of its comments as spam, which made me laugh. We're all adapting to this new way of working.
Over time, I've found myself enjoying these automatic code reviews. Some of Gemini's comments have been genuinely helpful:
- Mismatched Jira ticket numbers between PR description and branch name
- Leftover println debug statements
- Missing default value in a null check
- A test missing a null case
Of course, not every suggestion makes sense. One example: Gemini proposed merging a series of async calls into a single awaitAll block - which would've meant if one call failed, all would fail. Not ideal. Yes, I could have wrapped it in a supervisor scope, but I preferred a simple, KISS approach.
What's interesting is that we're all slowly starting to respond to the comments - accepting some, dismissing others, or committing changes accordingly.
AI for User Sentiment
Another area where AI has subtly integrated into my daily work is in tracking user feedback. At work, we have a dedicated Slack channel that receives all our Google Play store reviews.
Slack has recently added an AI summary function for channels, which has become incredibly handy. Instead of manually scrolling through hundreds of reviews, I use the AI to quickly get a summary of user sentiment for the last 7 days. This allows me to immediately gauge whether recent changes have been well-received or if there's a critical new bug that needs attention, saving a significant amount of time.
For my personal projects
For personal projects, like my AI studies and experiments, I mainly use ChatGPT 5 and Gemini CLI. I really enjoy Chat GPT's study mode. I am also discovering Notebook LLM, and I love the diagram feature. This helps me learn new concepts and connect the dots between ideas.
Looking Ahead
I've adapted to using these AI tools as an individual, and they've already changed the way I work day-to-day.
The next challenge will be integrating this technology into our team dynamics - learning how to collaborate effectively with AI, fine-tuning its behavior, and deciding when to trust its suggestions.
It's exciting to see this new way of working unfold. With thoughtful adaptation, it has the potential to make our development process more efficient - and maybe even more enjoyable.
