Gemini Code Assist has specific strengths that most developers never discover because they treat it like a generic Copilot alternative. Here’s where it genuinely outperforms and how to get more from it.
1. Use Gemini in Cloud Console for Live Infrastructure Help
Most developers only use Gemini in their IDE. The Cloud Console integration is equally powerful:
Click the Gemini icon (
) in console.cloud.google.com and ask:
Why is my Cloud Run service returning 503 errors?
Check the logs from the last hour.
Which BigQuery queries cost the most this month?
Show me the top 5 by bytes scanned.
Gemini in Console has access to your actual project resources and can give answers based on real data, not hypotheticals. This is something no other AI coding tool can do.
2. Leverage the Massive Context Window
Gemini’s 1M+ token context window is its biggest technical advantage. Use it deliberately:
@codebase Analyze the entire authentication flow in this project.
Trace from the login form submission through middleware,
services, database queries, and session management.
Identify any security gaps or inconsistencies.
With other tools, this kind of codebase-wide analysis hits context limits. Gemini can genuinely hold your entire project in context and reason about cross-cutting concerns.
3. BigQuery SQL Is a Sweet Spot
Gemini understands BigQuery SQL dialect (not just standard SQL) at a deep level:
Write a BigQuery query that:
- Calculates monthly recurring revenue (MRR) by cohort
- Uses the subscriptions table (user_id, plan, amount,
started_at, cancelled_at)
- Handles mid-month upgrades and downgrades
- Partitions by signup_month for cost optimization
- Uses approximate_count_distinct for the user counts
It handles BigQuery-specific syntax: UNNEST, STRUCT, ARRAY_AGG, partitioning hints, materialized views, and scheduled queries. General-purpose AI tools frequently generate standard SQL that doesn’t run on BigQuery.
4. Firebase Security Rules Generation
Firebase security rules are deceptively complex. Gemini handles them because it understands the rules language:
Write Firestore security rules for:
- Users can read/write their own document in 'users' collection
- Users can create posts in 'posts' if they're authenticated
- Users can only edit/delete their own posts
- Admin users (custom claim: admin=true) can read/write everything
- Comments subcollection: authenticated users can create,
only author can edit/delete
- Rate limit: max 10 posts per hour per user
The rate-limiting logic in Firestore rules is especially tricky — Gemini generates working rules that other tools struggle with.
5. Android-Specific Compose Patterns
In Android Studio, Gemini understands Jetpack Compose at a level that general tools don’t:
Create a Compose screen with:
- Top app bar with back navigation and overflow menu
- Pull-to-refresh LazyColumn
- Each item is a card with image (Coil), title, subtitle,
and bookmark toggle
- Shimmer loading placeholders during initial load
- Empty state illustration when no items
- Use Material 3 design tokens
- Handle configuration changes correctly
Gemini generates Compose code that handles lifecycle correctly, uses remember and derivedStateOf appropriately, and follows Material 3 patterns — areas where ChatGPT and Copilot often produce code that compiles but has subtle lifecycle bugs.
6. Use Gemini for Terraform on GCP
Write Terraform for a production GKE cluster:
- Regional cluster in us-central1
- Node pool: e2-standard-4, autoscaling 2-10 nodes
- Workload Identity enabled
- Binary Authorization enabled
- Network policy enforcement
- Private cluster with Cloud NAT for egress
- Monitoring with Cloud Operations Suite
Gemini generates Terraform with GCP-specific provider resources and correct attribute names. It understands which features require specific API enablement and adds the necessary google_project_service resources.
7. Multimodal: Screenshot to Code
Gemini supports image input. In Gemini Chat:
- Take a screenshot of a UI you want to replicate
- Paste it into the chat
- Ask: “Recreate this UI using Jetpack Compose with Material 3” or “Build this layout using React and Tailwind”
The large context window means Gemini can analyze complex UIs with many components and generate comprehensive code rather than just the obvious elements.
8. Cloud Functions with Proper Triggers
Create a Cloud Function (2nd gen) in Python that:
- Triggers on Firestore document update in 'orders' collection
- Only fires when the 'status' field changes to 'paid'
- Generates a PDF invoice using ReportLab
- Uploads it to Cloud Storage
- Sends a notification via Firebase Cloud Messaging
Include the deployment command with proper flags.
Gemini generates Cloud Functions with the correct trigger decorators, event parsing (which differs between 1st and 2nd gen), and deployment configurations.
9. Compare GCP Services Through Chat
For a real-time data pipeline processing 100K events/second:
Compare Cloud Pub/Sub + Dataflow vs. Kafka on GKE vs. Cloud Tasks.
Consider: latency requirements (<1s), cost at this scale,
operational overhead, and team expertise (we know Python,
limited Java experience).
Gemini’s comparison of GCP services is more nuanced and accurate than general-purpose AI tools because it understands actual service limits, pricing tiers, and integration patterns.
10. Workspace API Integration
If you’re building tools that integrate with Google Workspace:
Write a Google Apps Script that:
- Triggers daily at 9am
- Reads all rows from a Google Sheet where 'Follow Up Date' is today
- For each row, creates a draft email in Gmail using data from the row
- Updates the sheet with 'Draft Created' status
- Sends a summary to a Slack webhook
Gemini understands the Apps Script environment, its specific APIs, quota limits, and deployment model. This is a niche but highly practical use case for teams using Google Workspace.
Where has Gemini surprised you with its output quality? Share your experiences. 