# Gemini 3: The Developer's Guide to the New AI Era > A week with Gemini 3 Pro: Deep Think on a real migration, a bug diagnosed from a 2-minute video, and the agentic stuff that still needs a babysitter. → I've been testing Gemini 3 Pro for the past week, and honestly? I'm annoyed it's this good. Not because it's bad. The opposite. I'd finally gotten comfortable with my Claude/GPT-4 workflow, and now Google drops this on November 18th and throws everything into chaos again. Classic. But here's the thing: after forcing myself to rebuild a feature I'd already finished (yeah, I rewrote working code just to test this), I get the hype. This isn't your typical "10% better at benchmarks" release. Deep Think and actual multimodality (not the fake kind) change how you can work. Let me show you what I mean. ## What's actually different this time ### 1. Deep Think, or why it actually pauses now You know how previous models would just... start vomiting tokens? Mid-thought, zero planning, just word prediction on steroids? Gemini 3 actually stops. It thinks. You'll see "[Thinking...]" in the API response, and at first I thought my connection dropped. Nope. It's genuinely spending compute time planning its answer before committing. **Where I've actually used this:** Last Thursday, I asked it to plan our migration from a monolithic Angular app to micro-frontends. Instead of immediately dumping code, it: - Asked about our deployment pipeline (I didn't even prompt for this) - Identified 3 legacy dependencies that would break - Suggested a phased rollout plan that actually made sense Could Claude do this? Maybe. But Gemini 3 did it without me babysitting the prompts. **The catch:** It's slower. If you're used to instant responses, the 3-5 second thinking pause feels eternal. But for architecture decisions? I'll take accuracy over speed. ### 2. Multimodality without the duct tape Every model claims "multimodality" now. Usually, it means they duct-taped three different models together and hoped for the best. Gemini 3 is different: one model handles everything. Text, code, video, audio, images. Same weights, same architecture. **Real test I did:** I recorded a 2-minute Loom video of a UI bug (button wouldn't disable after clicking). No transcript, no code snippets. Just me clicking around and complaining in Spanish. Uploaded it. Asked: "What's broken?" It responded: - Identified the event handler wasn't preventing double-clicks - Pointed to the React component by name (HOW?!) - Suggested adding a `isLoading` state with `disabled={isLoading}` I checked. It was right. The component name was correct. From a VIDEO. **The weird part:** It occasionally hallucinates component names if your video is too long (over 5 minutes). But for quick bug hunts, nothing else I've used comes close. ### 3. Agentic Capabilities (With a Giant Asterisk) Google's numbers look great on paper: SWE-bench Verified jumped from 59.6% to 76.2% (Gemini 2.5 Pro to Gemini 3 Pro). In practice? It's... complicated. **What works:** - Running terminal commands (I let it debug a Docker networking issue, and it actually used `docker inspect` correctly) - Editing files (made a 12-file refactor across my codebase without breaking tests) - Running tests and interpreting failures **What doesn't:** - It still sometimes tries to run commands that don't exist (`git commit -fix` is not a thing, Gemini) - Gets confused if your project structure is unconventional - Will confidently suggest deleting files that are actually critical (caught this twice; always review) Trust, but verify. Always. ## Where it slots into a real workflow ### 1. Stop Writing Boilerplate (Seriously, Just Stop) Forget "vibe coding". Let's talk about the stuff you actually hate doing. **My new workflow:** Instead of scaffolding another CRUD API by hand, I now do this: > "I need a FastAPI endpoint for user authentication. JWT tokens, refresh logic, PostgreSQL with SQLAlchemy. Follow our existing pattern from the `/products` endpoints." Then I upload our products module as context. Gemini 3 generates 90% correct code. The other 10%? Usually just import paths or environment variables. **Time saved:** What used to take 2 hours now takes 20 minutes. **The catch:** You need a consistent codebase pattern. If your project is chaos, Gemini 3 will reflect that chaos. ### 2. Code Reviews (But Not How You Think) I don't trust AI for full code reviews yet. But I do use Gemini 3 as a "first pass filter." **My setup:** 1. Developer opens PR 2. GitHub Action sends diff to Gemini 3 3. It flags **obvious** stuff: hardcoded credentials, missing error handling, SQL injection risks 4. Posts these as automated comments 5. I (human) do the real review for logic and architecture This catches the boring stuff so I can focus on whether the code actually solves the problem. **Warning:** It will sometimes flag false positives. Last week it complained about a `.env` file that was actually a `.env.example`. Don't blindly merge based on AI feedback. ### 3. Legacy Code (Where It Actually Shines) Remember that 6-year-old jQuery spaghetti nobody wants to touch? Gemini 3's 1M token context is perfect for this. **What I did last month:** Uploaded our entire legacy admin panel (22 files of vanilla JS horror). Asked it to: - Identify which files are actually used - Create a dependency graph - Rewrite the user management module in React It worked. Not perfectly (I had to fix event handling bugs), but it did 70% of the grunt work. **Pro tip:** Do this incrementally. Don't ask it to rewrite your entire app. Start with one isolated module, verify it works, then move to the next. ## So... Should You Care? Look, I'm not switching from Claude entirely. For brainstorming and writing, Claude still feels more "human." But for code? Gemini 3 is winning me over. The Deep Think feature alone saves me from second-guessing architecture decisions. The multimodality is genuinely useful (not gimmicky). And the context window means I can actually load real projects, not toy examples. **My advice:** - **Try it** if you're doing refactoring or working with legacy code - **Stick with Claude/GPT-4** if you need creative writing or nuanced explanations - **Don't trust it blindly:** I've caught it suggesting bad practices when it's outside its knowledge cutoff The future of development is probably agentic. But we're not there yet. Gemini 3 is a big step, though. Now excuse me while I go rewrite my CI/CD pipeline for the third time this year. **Related reading:** more on letting AI take the first review pass: [I tried using AI for code review on my side projects](/en/blog/gemini-flash-code-review-automation/). --- **P.S.** If you've tried Gemini 3, I'd love to hear what you think. Did it hallucinate component names for you too, or is that just me? Hit me up on [Twitter/X](https://x.com/garbarok).