Gemini CLI: Claude-Like Guidance, Unique Failure Modes

Gemini CLI's conversational UX feels remarkably like Claude Code—it explains decisions, walks you through setup, and genuinely guides rather than just executes. But it has unique failure modes you need to understand: execution loops, resource exhaustion, and no screenshot support.

I built a simple app tracker with Antigravity (Google's AI IDE) and wanted to add Firebase authentication and cloud sync. The catch? I'd never used Firebase before. Gemini CLI just released extensions—essentially MCP servers with custom instructions for specific tools—so I installed the Firebase extension to see if it could guide me through setup and handle the migration.

The first surprise was how conversational Gemini felt. When I asked if Firebase was the right choice, it didn't just start coding—it explained alternatives (Supabase, Auth0), trade-offs, and why Firebase fit my needs. It walked me through console setup, CLI authentication, and Firestore configuration step-by-step. This felt more like Claude Code's hand-holding approach than the "just execute" behavior I've seen in other tools.

The migration started smoothly. Gemini warned me it would need to refactor significant chunks of code to support Firestore's data model. Authorization worked. State management was updated. Then I tested the app—and drag-and-drop was completely broken.

That's when things got rough. Gemini entered execution loops, flashing through "execute, confirm, proceed" cycles. When I asked it to use Playwright for testing, it looped through opening browsers with failed test reports. Resource exhaustion errors appeared—not from hitting token limits, but from hammering the API so rapidly it triggered rate limiting.

Two discoveries changed everything. First: Gemini CLI can't see images. I'd been sending screenshots the entire time, and it finally told me "I can't directly view images." Unlike Claude Code or Codex, Gemini is completely blind to screenshots unless you install the nano banana extension.

Second: I asked it to enter "planning mode." Gemini doesn't have a built-in planning mode like Antigravity or Claude Code's /plan command, but when I said "I need you to make a plan, follow it step by step, and slow down to avoid quota limits," it created one. That turned everything around.

With a structured plan, progress became steady. The layout issues got fixed. Drag-and-drop slowly came back to life. I hit the token limit right near the end, but Gemini finished the last task before timing out.

Final result: A working app with Firebase authentication and cloud sync. It took about five hours and required significant human guidance to recover from loops, but the conversational UX is real. Gemini explains what it's doing, gives opinions on technical choices, and recovers when you provide structure.

What worked: Conversational guidance that rivals Claude Code. Clear step-by-step explanations even for tools I'd never used. Recovery from failure states when given explicit structure.

What didn't work: No screenshot support (solvable with nano banana extension). Gets stuck in execution loops that require manual intervention. Resource exhaustion handling is rough. Drag-and-drop broke during migration—though Antigravity also struggled with this, so it might just be hard.

Open questions: Would Claude Code have the same Firebase migration issues? Is drag-and-drop complexity model-specific or just genuinely difficult? Would starting fresh on a clean branch produce different results?

For now, I learned Firebase basics, got a working app, and discovered where Gemini CLI shines and where it needs work. If you want to see a Claude Code comparison on the same Firebase migration, let me know in the comments.

Tools mentioned:
Gemini CLI: https://ai.google.dev/gemini-api/docs...
Firebase: https://firebase.google.com
Antigravity (Google AI Studio): https://aistudio.google.com
Claude Code: https://claude.ai/claude-code
MCP (Model Context Protocol): https://modelcontextprotocol.io

CHAPTERS:
0:00 Introduction - Can Gemini CLI Migrate to Firebase?
0:32 The App & Project Setup
1:56 Getting Started with Firebase Extension
4:08 Firebase Setup & Configuration
5:17 The Migration Begins
7:07 First Problems Emerge
8:03 The Loop & Resource Exhaustion
10:10 The Screenshot Discovery
12:05 Planning Mode to the Rescue
13:18 Token Limits & Final Success
14:53 What Worked Well
15:38 What Didn't Work
16:59 Final Thoughts

#GeminiCLI #Firebase #AICoding #GoogleAI #DeveloperTools

Subscribe for more AI reviews, comparisons and walkthroughs