I was sitting in my car this morning just staring at the dashboard and I had this weird thought it feels like most of us are sitting in the driver's seat of a Ferrari this machine engineered for you know absolute peak performance but we're driving it around a golf course at five miles an hour we have this massive engine under the hood but we're just we're afraid to do anything more than tap the accelerator that is wow that's painfully accurate, especially if you look at
how the average person is using Gemini 3 .0 right now. We treat it like a search engine from, what, 2010. Ask a question, get an answer, close the tab. But the reality is this thing is designed to run your life while you sleep. It's not a chatbot anymore. It's a workflow engine. Welcome to the deep dive. I want to slow things down today. We usually rush through the news, but today we have a piece of source material that I think requires us to actually, you know, pause
and think. We are looking at a guide by Max Ann titled, You're Using Gemini 3 .0 Wrong. It was published just yesterday, February 8th, 2026. And the premise is exactly what you just said. The AI landscape has shifted right under our feet, and most of us just completely missed it. It really has. And that shift is subtle, but it's profound. We're not just generating text anymore. We are building deliverables. We're
building systems. Right. So our mission today is to unpack seven specific hidden features that are buried, and I mean deep in the settings menu of Gemini 3 .0. These are features that, according to Ahn, change the whole game. But before we get to the how -to, I want to talk about the friction. Ann calls it the amnesia problem. This is the core argument of the whole guide, really. It definitely resonated with me, but I have to play devil's advocate for a second. Isn't amnesia
just a feature of software? I mean, I open Word, it's blank. I open Excel, it's blank. Why is this a problem for AI? Because Word and Excel are tools you use to output your thoughts. The AI is supposed to be a partner that contributes thoughts. Think about it this way. If you hired a new executive assistant, a human being, and every single morning when you walked into the office, you had to reintroduce yourself. Hi, I'm Dave. I wrote a podcast. Here's our style
guide. Here's who our sponsors are. And then the next day you do it all over again. Okay, yeah. I'd fire them by lunch. Exactly. Yeah. You'd never get any deep work done because you're spending 90 % of your energy just on onboarding. But that is the default setting for Gemini. It starts from zero. Every time. It has no memory of your style, your context, your projects. So real work, complex work, it exposes those gaps immediately. You feel that friction because you're
just constantly repeating context. So if the intelligence, I mean, if the intelligence isn't the problem, but the memory is, what is the architectural fix here? Because I don't want to just give it all my data and hope for the best. The fix is that you have to stop treating it like a blank slate. You have to enable the persistent productivity system settings. You build a system of record inside the AI. Okay, let's get into the mechanics of that. This leads us to the first big feature,
max and highlights, custom gems. Now, I'll be honest, gems sounds a bit like a mobile game currency to me. It does, doesn't it? But stick with me on this one. So in this context, it's something very different. Right. Think of a gem as a cloned version of the AI that you have trained for one very specific purpose. You aren't just opening a chat. You are defining three things.
A roll. a task and an output can you walk us through a concrete example the guide mentions a meeting summary assistant but that sounds pretty basic i can already ask it to summarize meetings you can but how often do you get a summary that's actually you know usable usually it's too long or completely misses the point so with the gem you don't just paste notes and say summarize this you build the gem once you tell it roll
You are my executive assistant. Task. When I paste meeting notes, extract key decisions, action items with owners, and any open questions. Output. Keep it under 300 words. Use bullet points. Okay, so it's a saved prompt. I feel like I've seen that before. It's more than a prompt. This is the part that people miss. You can upload reference files to the gem. Up to 10 documents. What kind of documents? Like a PDF of my company handbook or something? You could, but that's kind of the
boring way to do it. The guide suggests something much smarter. Don't just upload a generic style guide. Upload the last five emails you wrote that you were actually proud of. Upload a project brief that got approved immediately. That's interesting. So you're giving it training data on your best work. Exactly. You're saying, don't just write like a professional. Write like me. on a good day so it mimics your specific voice your formatting even your shorthand That's a huge difference.
Usually I find myself typing the same three paragraphs of instructions. You know, don't use emojis. Keep it brief before I even paste the content I want processed. Right. And that's that friction we talked about. With a gem, you click the preset in your sidebar, paste the notes, and it just executes. It already knows the rules before you type a single word. So effectively it becomes a clickable preset that houses your specific
institutional knowledge. Precisely. All right. Let's pivot. So gems solve the issue of repeating instructions, but that doesn't solve the issue of the AI being isolated from your actual data. It still doesn't know what I'm working on unless I paste it in. Right. By default, Gemini is sitting in a clean room. It can't see your Gmail, can't see your drive. It doesn't know you have a dentist appointment at 2 p .m. It's smart, but it's blind to your actual life. And this brings us to feature
number two, connecting Google Workspace. There's a toggle in the settings to change that. Yes, under Connected Apps. Once you flip that switch, you unlock the ability to use the at symbol to direct its attention. I have to admit, when I read this part of the guide, I was a little skeptical. It felt a bit gimmicky. But then I tried the example. at Gmail draft a reply to Sarah about the Q1 report. And what happened? Well, it worked. But more importantly, it didn't just write a
generic Dear Sarah email. It actually found the thread. That's the whoa moment. It reads the thread to understand the context of what Sarah asked. Yeah. It looks at your previous replies to understand the tone you usually take with her. And then it drafts the response. It's not just writing text. It's performing an action based on history. And it's not just email. The guide lists at Drive to find files, at Calendar to check schedules, and even at Workspace to
search everything at once. Think about the panic search. You know that feeling, right? You're on a Zoom call. Someone asks, hey, where did we land on that budget proposal from last Thursday? Oh, God. And you start sweating. You open three tabs. Is it in Drive? Is it in Slack? Is it in an email? Exactly. Instead of panic searching, you just type into the chat. At Workspace, find the deck from my Thursday client call. And it pulls it up. It just eliminates that cognitive
load of, where did I put that thing? Does this fundamentally change the search behavior of the user? Yes, completely. It eliminates tab switching by bringing the data directly into the chat interface. Let's talk about the third feature, which I think is my favorite concept in the whole guide, scheduled briefings. Max Ann describes this as work happening while you sleep. This is where we move from being reactive to proactive. Usually you have to initiate
the interaction with the AI. You ask, it answers. Schedule briefings flips that. You write a prompt, but you attach a time trigger to it. So give me the mechanics. How does that look in practice? Do I need to know how to code or something? No, and that's the beauty of it. It's surprisingly natural language. You literally type every Monday at 7 a .m., summarize my calendar, and highlight meetings that need preparation. I have to be honest here. I still wrestle with prompt drift
myself. I have all these good intentions to ask for a summary or check my tasks, but I get busy The day starts and I forget to ask until I'm already late for the meeting. That is exactly what this solves. It automates the asking. Yeah. You can set it up for daily briefings on unread emails from your boss or a Friday 5 p .m. summary of everything you completed that week so you can log your hours. It shifts the dynamic completely. You aren't chasing the information. The information
is waiting for you when you wake up. Precisely. It's the difference between pulling data and having data pushed to you. And I assume there's a dashboard for this. Or does it just run for hours? and I can never stop it. No, thankfully. There's a specific scheduled actions panel in the settings where you can pause or edit them. Okay, feature number four. This connects back to the memory issue we started with, but on a macro level. It's called personal instructions.
Think of this as global memory. So we talked about custom gems, which are for specific tasks, like a summary assistant. Personal instructions are for you as a user. Across every interaction, every chat, every day. So this is where I tell it things like I'm a podcaster or I prefer American English spelling. Exactly. Instead of explaining I like bullet points or keep it professional, in every single chat, you save it once in the
settings. The source lists examples like use a direct professional tone, keep replies under 200 words, or I work in B2B sauce marketing. But hold on. People change. My job changes. If I set this up to say I'm a marketing manager and then I switch to being a, I don't know, a yoga instructor, isn't this going to screw up all my results? That's a valid fear. But the guide mentions a really important command here,
the forget function. It's granular. You can tell Gemini, forget my memory about formatting preferences or forget that I work in marketing. And it wipes just that specific part while keeping the rest. That's smart. It lets the system evolve with you. What's the psychological benefit of this feature beyond just saving keystrokes? It stops the feeling of talking to a stranger. The tool
adapts to your identity. We're going to take a quick break, but when we come back, we're going to look at the flip side of memory, when you don't want the AI to remember you, and a feature for the heavy researchers out there that bridges the gap between your notes and the internet. Stay with us. Mid -roll sponsor break placeholder. Welcome back to the Deep Dive. We're unpacking the hidden features of Gemini 3 .0 based on Max
Anne's new guide. We just spent a lot of time talking about memory, context, and letting the AI know everything about you. But the guide brings up a really important counterpoint in feature number five, privacy. Right, because sometimes memory is a liability. It is. Sometimes I just want to ask a dumb question. Or I'm brainstorming a surprise party for my wife, and I don't want party planning to suddenly become part of my professional work profile. Exactly. Or you're
researching a medical symptom. You don't want that influencing your future business drafts. The guide calls this temporary chat mode. It's essentially incognito mode for your workflow. Yes. When you toggle this on, the history is deleted after 72 hours. And more importantly, that data isn't used to train the model and it doesn't get saved to your long -term memory settings. It's a safety valve. You can ask the dumb questions or the sensitive questions without polluting
your long -term profile. Exactly. It keeps your global memory clean. Now, feature number six is a heavy hitter. This is for the people who are doing deep work. The guide talks about the integration of Notebook LM. Now, I know Notebook LM as a separate app. I've used it to organize research papers. Is this different? It's the same brain, but now it's integrated directly into the interface. It's designed for when you need to synthesize a massive amount of information.
The guide says you can attach a notebook with up to 300 sources, PDFs, videos, Google Docs, entire websites. That's a library. That's not just a file. That's a massive archive. It is. But the magic in Gemini 3 .0 is the combination. In the old world, you had your static notes in one app and live search engine in another. Here, you have both. So you can ask a question that bridges both worlds. Yes. The example in the guide is perfect, you can ask. What are the latest
trends in AI video? That's a live web search. And then in the same prompt, add, and how do they compare to the research in my attached notebook? And that's the static knowledge base. That is incredibly powerful. You're cross -referencing the entire internet against your private archives in seconds. It creates a synthesis that neither tool could do on its own. The live web doesn't know your notes, and your notes don't know what happened five minutes ago. Putting them together
is where the insight happens. So it essentially bridges the gap between your private archive and the public internet. Exactly. Deep structured research meets real -time data in a single answer. We are down to the final feature, number seven, and the bonus. This is about moving beyond text. We're talking about canvas and formulas. I feel like canvas is a term that gets thrown around a lot in tech. What does it mean here? Canvas is a shift in interface. Usually you chat, you
get a column of text. It's linear. But if you're building a presentation or a coding project, linear text is terrible. Canvas opens a dedicated window on the side to build actual artifacts. The guide mentions asking it to... create a 10 -slide presentation. And it doesn't just outline the slides in text. It builds the slides. It handles the formatting, the structure, the flow.
And because it's a dedicated window, you can edit the text directly, just like you would in a document editor, while still chatting with the AI to make big changes. So I can highlight slide three and just say, make this punchier. Exactly. It's collaborative editing, not just command and response. It also mentions building web apps or interactive quizzes from dense PDFs. Turn this policy document into a quiz and it builds a playable interface. It's moving from
tell me to make this. Which brings us to the bonus feature AI formulas in Google Sheets. Oh man, this one hits home. I live in spreadsheets and they are usually a mess. We all do. This feature is a superpower for messy data. It's the AI function. Walk me through a painful scenario. Let's say I just got back from a trade show. Okay, so you have a CSV export of 500 people. The names are mixed with the job titles. Some
are in all caps, some are lowercase. And the interests column is just a paragraph of text. Usually cleaning that up is two hours of manual copy pasting or writing complex nested formulas that break if you breathe on them wrong. Right, and I usually just give up and do it manually. With this, you just type AI. Extract the job title and format as title case A2 and drag it down the column. It processes every row individually using the LLM. That is wild. It treats every
cell like a tiny prompt. Exactly. You can use it to translate, to extract sentiment from customer reviews, to reformat dates, anything. But there's a catch, right? It can't do a million rows instantly. Correct. The guide notes that it currently processes only the first 200 cells at a time due to usage caps. Still, for a lot of tasks, 200 rows of automation... thinking is a lot better than zero. Absolutely. It turns a two -hour headache into a two -minute drag and drop. So let's zoom out.
We've covered seven features, custom gems, workspace connections, scheduling, personal instructions, privacy, Newtbook LM, and Canvas. What is the big idea here? If you take one thing away from this deep dive, what should it be? The big idea is the shift in mindset. We are moving from a search engine mindset to a productivity system mindset. The search engine is passive. You go to it when you have a question. The system is
active. It surrounds you. It feels like the goal is to set up a system that remembers how you work so you don't have to repeat yourself. That's it. Automation in context. Yeah. If you were repeating yourself to your AI in 2026, you were using it wrong. The friction isn't the tool's intelligence. It's the user's default settings. Maxanne ends the guide with a bit of a challenge. She says, stop reading. Start setting this up.
It's good advice. It feels a little unfair how much time this saves once it's running, but you have to do the initial lift. You have to go into the settings, turn on the workspace connection, and write your first gem. It's the difference between driving the Ferrari in first gear and finally shifting up. Well said. Thanks for listening. Go fix your settings. We'll see you in the next deep dive. Thanks, everyone.