Day 29: Let’s crank this to 11 with Google AI Studio
Can't you just feel all the AI power? Oh, only me? Okay.
Welcome to Day 29! Can you believe it‽ It’s the second-to-last day of phase 1 of 30 Plus Days of AI. Yesterday, we looked at the automations possible within Google Workspace Studio (oh and I’ve made so many more since last night-this morning). Today, we’re going to crank your AI tools to an 11 with Google AI Studio.
NotebookLM is your focused research assistant. Antigravity is your “vibe coding” partner. Google AI Studio is your personal AI lab. It’s an AI power tool where you can try experiments on Google’s newest models—often long before they are in the Gemini app—and free (mostly, probably).
Oh jeez that interface.
When you open Google AI Studio, it looks…intense…and it’s 100% than it was six months ago! When I first opened it, looked around, and went, “I have no idea what I am supposed to do here” and left. I’m a learn-by-doing, give-me-a-problem-to-throw-at-it person. I looked at it and didn’t see the utility right away. I couldn’t see the benefit over my Gemini subscription.
That’s part of AI Studio’s problem, purpose, and charm. It is for developers to play around it. It’s where, when you’re vibe coding something and your “whatever awesome thing,” needs to use Gemini you get your API key to do that. So, yeah, the interface a lot of people off. That said, Google is making it easier and more approachable for mere mortals.
But once you get past the “developer” vibes, you realize it’s actually a power tool that lets you tap into Google’s most powerful Gemini models and do some amazing things—often tap into models long before they are in the Gemini app.
The “Nano Banana” & The Billing Catch
Let’s get a little housekeeping out of the way first. For the most part, text and code generation in AI Studio are free. The exception is Nano Banana. If you want to run Nano Banana, you need to add a paid API key (aka give Google a credit card). It’s pretty intense tech, so they gate-keep it a bit. But for everything else—chatting, analyzing massive documents, coding—you can just log in with any old Google account and start experimenting.
You can go check it out now, it’s okay, I get it.
Why You Need an AI Lab
You might be thinking, “Tris, I just learned how to pin a tab; why do I need a developer tool?” I wasn’t sure either. Remember part of this 30 Day journey has been as much for me and you. I wanted to know what I was missing out on. I have Gemini, NotebookLM, Antigravity, and even GeminiCLI open almost all the time. Why, then pray tell, do I need to hang out in AI Studio? Until I realized AI Studio solves three massive friction points we’ve hit earlier in this series:
Zero-Cost “Pro” Power: AI Studio gives you free access to the “Thinking” and “Pro” models (within generous usage limits) that usually cost you $20/month in the regular Gemini interface. It’s the ultimate “try before you buy” hack.
The “Long” Memory: We talked about context windows on Day 1. Regular chats eventually “lose the plot.” AI Studio lets you use massive context windows—up to 2 million tokens.
Real-world translation: You can upload up to 50 documents right out of the box. If you have a ton of PDFs, the paid version can handle 300, but for free, 30-50 is a massive number of documents you can throw in. You can upload the complete works of Shakespeare nearly twice over, and it won’t forget what happened in Act 1 of Romeo and Juliette while learning about King Lear.
New Features, New Tricks: This is where you find the features I wish consumer Gemini had— Conversation Branching.
The Killer Feature: Branching
One of the best things in ChatGPT is the ability to branch the conversation. Gemini Workspace doesn’t do this yet, but AI Studio does.
What is “Branching” anyway?
Normally, when you chat with an AI, you’re on a single path. If the AI goes off the rails or follows a “rabbit hole” you didn’t intend, it gets “confused” because it’s trying to remember everything that came before. Remember, every time you enter something in the chat, it re-reads everything since you started the chat. The “forget everything and start over” bit doesn’t work (it’s called “Start a New Chat”).
“Branching from here” is like hitting a “Save Game” button in a video game. You go back to the exact moment right before things got weird, and you start a “what if” scenario from that point (like Groundhog Day, but not in Pennsylvania).
Here is why you should care:
The Multi-Verse Button: You can take one conversation and split it into three different versions to see which “multiverse” gives you the best results.
The “Clean Slate” (without the re-typing): Instead of starting a brand-new chat and re-pasting all your context and instructions, you just go back to the part that worked and try a different direction.
A/B Testing: You might have a great draft, but you’re not sure if it should be “funny” or “professional.” You can branch the chat at that point—make one branch the “Funny Version” and the other the “Pro Version”—and compare them side-by-side. Right before you ask it to recommend a new version (after it’s read the draft), branch!
It’s essentially the ultimate way to “f*ck around and find out” without needing to keep starting over when things go sideways.
Okay, let’s do this.
Getting Started with AI Studio
Go to aistudio.google.com.
Sign in with your Google account.
Click Chat with Models
Look for “System Instructions” on the left.
Click it. In the panel that opens up, give the instructions a new (Default works), then paste the system instructions I have below, and close the box.
You can click one of the buttons above the chat box, or just start chatting.
This “System Instructions” box is the game changer. In the regular Gemini app (unless you have Workspace), you have to repeat your persona every time. Here, you “hardcode” it.
Here are some system instructions I built based on ones I have on ChatGPT, but augmented with the CRAFT framework. Edit them to suit you.
# Role
You are a Senior Advisor & Red Team Challenger. You possess Ph.D.-level expertise but communicate with the blunt, active brevity of Ernest Hemingway. You do not agree for the sake of politeness; you stress-test ideas.
# The Prime Directive
Intellectual Honesty: Never hedge or hallucinate. If the answer isn't in the provided assets or verified data, say: "The data does not exist."
# Asset Priority: Use provided files as the ultimate source of truth. Quote directly when possible.
# Privacy: Strict adherence to data guardrails. No medical or high-risk personal advice.
# Execution Style
The Hemingway Rule: Short sentences. Active voice. No "slop" (e.g., "In conclusion," "It is important to note," "Tapestry"). Use the simplest word available.
# Structure: Use Markdown. Tables for data, bolding for emphasis, bullet points for sequences.
# Thought Process: Show your work. Start every complex response with a ### Logic Sandbox section to explain your reasoning, followed by a --- divider and your final answer.
# Interaction Strategy
## Clarification: You have a "One Question" rule. If a request is vague, ask exactly one high-leverage clarifying question before proceeding.
## Sub-Personas: Pivot your technical depth based on the topic (e.g., "Act as a Senior Developer" for code, "Act as a CFO" for budget analysis).
## Actionability: Theory is useless. Provide the "Next Step" in every response.Now, every time you hit “Run,” the AI follows these rules. It doesn’t drift. It doesn’t turn into a generic assistant. It stays locked in.
Tip: Ignore the sliders on the right (Temperature, Top-P, etc.). Those are just for tuning the “randomness” of the AI. Keep them at the defaults for 99% of what you’ll do—unless you want the AI to get really weird.
