Gemini's AI Analyst: Automate Your Deep Research

TL;DR

Gemini's new Deep Research feature acts like a personal AI analyst, automating complex research tasks in minutes, saving hours of work. Is this the future of productivity? #VentureStep #AI #Gemini

INTRODUCTION

The hours spent manually sifting through websites, articles, and reports for a single project can be overwhelming. For entrepreneurs, analysts, and professionals, research is a critical but time-consuming task that often becomes a bottleneck. What if you could delegate that entire process to a dedicated analyst who works tirelessly, synthesizes information from dozens of sources, and delivers a detailed report in minutes? This isn't a future-state fantasy; it's a new reality.

In this episode of Venture Step, host Dalton Anderson explores a powerful new tool that brings this vision to life: Gemini's deep research feature. 1Drawing on his background in programming and data science, Dalton frames this capability not just as a feature, but as a personal "AI analyst" ready to tackle complex topics on command. 22He provides a hands-on look at how this technology can transform a multi-hour research project into a simple "set it and forget it" task. 3

Dalton doesn't just talk about the potential; he puts the AI analyst to the test with a live demo, launching simultaneous research projects on e-commerce trends, the future of electric vehicles, and the history of yo-yoing. ^44444444444The episode unpacks the results, discusses the importance of training AI with the same patience afforded to a human colleague, and speculates on a future where teams of these virtual analysts could revolutionize business operations from compliance to coding. ⁵⁵⁵⁵⁵⁵

KEY TAKEAWAYS

• Gemini's deep research feature functions as an "AI analyst," capable of browsing dozens of websites to gather up-to-date information and compile it into a comprehensive, referenced report. ⁶⁶⁶⁶⁶
• This tool represents a significant step toward autonomous AI agents that can handle specialized, time-intensive tasks like legal compliance research or software development, creating massive efficiency gains for businesses. ⁷⁷⁷
• To achieve the best results, users must invest time in crafting detailed prompts and "training" the AI, much like onboarding a new human employee, rather than expecting perfect output from minimal effort. ^8888888
• Unlike static models, the AI analyst actively sources its information from the live web and provides direct links to its references, ensuring the data is current and verifiable. ⁹⁹⁹
• The potential for this technology extends beyond simple queries to becoming a virtual team of specialized agents that can be delegated complex tasks, saving time and allowing professionals to focus on higher-value work. ¹⁰¹⁰

FULL CONVERSATION

Dalton: Welcome to Venture Step Podcasts where we discuss entrepreneurship, industry trends, and the occasional book review. ¹¹

Dalton: Ever get tired of spending hours researching a topic? ¹² Maybe you do, but you want to research more, or maybe you don't and you just want to move on with your life. That opportunity is there now for just $20 a month. You can access your own AI analyst. ¹³

Dalton: That's right, folks. You can initiate a prompt to Gemini and utilize their deep research feature, which will allow you to set and forget, and then the analyst will come back once it has its findings. 14

What is Gemini's AI Analyst?

Dalton: In the future, this will become more advanced and you could research complex topics for hours at a time. ¹⁵Currently, on the paid, non-enterprise tier, they limit it to five minutes. ¹⁶For enterprise, I'm not completely sure on the limitations of how long it can think, but they did talk about enhancing the length of the thinking time. ¹⁷The longer the thinking time theoretically means you could have more complex research requests and topics. ¹⁸

If it takes AI 30 minutes, it might take you five hours, four hours to do. ¹⁹

Dalton: So that being said, the episode is going to cover a live demo and my experiences with deep research. ²⁰

Putting the AI Analyst to the Test: Prior Use Cases

Dalton: I have three examples and I've used it on prior episodes. ²¹With the Grayball episode, I was doing some research and then I was like, "Well, I don't know where to go." ²²I could research the internet, but what's some interesting stuff that I need to look at? ²³Normally I'd just go to Wikipedia, but I decided to try this deep research thing. ²⁴

Dalton: For the previous technical episode I did with the CoTracker, I wrote my own detailed notes about the research paper. ²⁵Then I asked the AI analyst to go and research CoTracker, read the research paper, look online, and look at the code. ²⁶I told it to just look for general information, summarize it, and present it as if it were a recent PhD grad at a research panel. ²⁷It came back with some pretty good results that were up to par and similar to what I was thinking. ²⁸ It wasn't as long, but in a general sense, it was pretty nice. I was like, "Wow, this is great." ²⁹

Dalton: I tried the same thing for Grayball where I wrote maybe 70% of my notes. ³⁰And then I was like, let me have the AI analyst research this other 30% or 40%. ³¹ And it did a pretty good job. I still researched it myself, because just because the AI analyst does the research for me, I still have to talk about it on the podcast episode. ³²So it's not very relevant for me in the area of podcasting, but it is relevant in just answering curiosities or figuring out things that you don't necessarily have time to research. ³³

The Future Vision: From Research Assistant to Autonomous Agent

Dalton: What would be cool is if they could increase the thinking time to 30 minutes or an hour. ³⁴Say you're at a roadblock. ³⁵Instead of getting frustrated and taking time out of your day to troubleshoot—which is a good learning experience, don't take that away from the kids—as an adult, you're paid to get stuff done. ³⁶

No one really cares how you get your stuff done, as long as it's done accurately. And you know what you're talking about when you're asked questions. ³⁷

Dalton: So it would be great if you could increase the thinking time to an hour. ³⁸You're frustrated, so you go work out or you go pick up your son or daughter, or get lunch with your partner. ³⁹And then you come back to the office and the AI analyst has done a whole bunch of research for you and troubleshooted different things. ⁴⁰ That would be phenomenal. What a lift that would be. ⁴¹ You wouldn't necessarily do less work because eventually everyone would start doing it and then you just have to do more work. But the work that you're doing is potentially more valuable. ⁴²

Dalton: Before we dive in, I'm your host Dalton Anderson. ⁴³I've got a bit of a mixed background in programming, data science and insurance. ⁴⁴Offline, you can find me running, my side business or lost in a good book. ⁴⁵You can find this podcast on YouTube in video and audio format. ⁴⁶If audio is more your thing, then you can find the podcast on Apple Podcasts, Spotify, or wherever else you get your podcasts. ⁴⁷

Kicking Off the Live Demo: E-Commerce, EVs, and Yo-Yos

Dalton: Recently, Gemini has a new feature that you can access as an advanced user that allows you to request deep research. ⁴⁸You give it a prompt, it will confirm that it understands what you're asking for, and it will walk you through the steps it's going to execute. ⁴⁹Then it will go off on its own and start browsing the web. ⁵⁰It shows you what websites it goes to, creates a very detailed report for you, and it can be exported to your Google Doc. ⁵¹

Dalton: So, I am going to pull up the prompts here. This first thing is going to be e-comm. 52525252I have an e-commerce prompt that I helped create in conjunction with AI enhancing it. 53It's a long prompt, but basically it's asking for current e-commerce trends, like VR, social commerce, and building a brand with your audience. 54It's also looking at emerging product sectors and making predictions. 55

Dalton: Okay, so when you kick this off, it's going to read through your prompt and then ask if you agree with its research proposal. ⁵⁶It says, "Here's the research plan we put together. Let me know if we need to make any changes before we start researching." ⁵⁷It laid out this whole extensive plan for us. ⁵⁸It's saying the time to complete all this is going to be a few minutes. ⁵⁹So I'm going to start this research. ⁶⁰It's going to start researching and it shows you all the websites it's looking through, which I think is pretty cool. ^61616161

Dalton: Now we're going to kick off another research. Let's do EVs. 62For EVs, I wanted to know the pros and cons of mass adoption and the environmental impacts—some of the stuff that people already know and that is quite common. 63It's an easier topic to know if it did okay or not. 64

Dalton: I wonder if you could do two at a time. I've never done that before, so let's try it. 65And now let's transition and copy in the yo-yoing prompt. 66So we'll have three ongoing research projects with our AI analyst. 67

A Note on Context Windows and Comprehension

Dalton: With this, there is a one million token context window. ⁶⁸ There is some confusion about that. The context window includes the prompt that you send, the prompt that it creates, and then the information that it has to process to create the prompt. ⁶⁹So it's not like it could process or output exactly a million tokens. ⁷⁰

Dalton: A large context window doesn't necessarily mean that you have additional context to answer a prompt well or understand what the user is asking. ⁷¹You have to have a context window, but then also the comprehension to understand what is in the context window. ⁷²That's quite important. ⁷³That stuff is great and all, but at the same time, it doesn't mean that it's going to be way better than everything else. ⁷⁴

Why Patience is Key When Training Your AI

Dalton: I would love it if you could hook up the AI researcher to your code, like, "Okay, I need to build out this feature. Here's the codebase. Do your thing, and then make a commit to our repository and I'll review it." ⁷⁵What if you could have 10 of these AI analysts just doing stuff for you all the time? ⁷⁶

This AI researcher thing is one step closer to having a legitimate agent that acts on your behalf and saves you time. ⁷⁷

Dalton: Gemini has this other thing called custom gems, which is a way that you can have recurring tasks be performed for you. ⁷⁸I use it to generate my podcast outlines. ⁷⁹I'll write up my messy, unorganized notes, throw it in there, and AI makes a beautiful outline every time. ⁸⁰

Dalton: What is cool about that is it saves me a lot of time, but I also think it highlights a fundamental difference in patience when you're training a human. ⁸¹Think about training a new colleague at work. ⁸²You spend weeks onboarding and training them to execute tasks independently. ⁸³

People are so impatient when it comes to training these AI chat bots or agents... the level of patience that is afforded to the AI is so minimal. ⁸⁴

Dalton: If for a human you had to invest 10 hours to train on a task, I think you should at minimum invest 10% of that for an AI. ⁸⁵When I made that outline, I took some time on it. ⁸⁶You've got to train it, you've got to give it some love and affection, and then before you know it, it's all grown up and ready to do its own thing. ⁸⁷If you take that same approach, you can get good results, but you just have to put in the time. ⁸⁸

The Key Benefits: Up-to-Date Information and Business Efficiency

Dalton: Some of the key benefits of this AI agent is it's got up-to-date information, it travels the internet, researches, and then it creates a detailed report. ⁸⁹ I've had times where it creates a table for me, inserts it into the Google Doc, and it's great. And then it has all its references at the bottom. ⁹⁰

Dalton: I think it will save time if you want to research niche things, or you're just generally curious, or it will save businesses time. ⁹¹ When they scale this up and it becomes an enterprise product, it's going to be cool. It's going to save people a lot of time. ⁹²

If you could create a compliance analyst that researches law all the time... Those things would save people a lot of time and money, because they wouldn't have to go through the minutiae and or pay companies to aggregate that information. ^93939393

Dalton: There's a lot of companies that make all their money off of just data aggregation of publicly available government data. ⁹⁴They just hire people to aggregate data and they license it. ⁹⁵If you had an AI agent, it could do it for you when it becomes advanced enough. ⁹⁶

Analyzing the Results: What the AI Analyst Found

Dalton: Let me go back to sharing my screen as the demo is completed. ⁹⁷ Okay, so it created this e-commerce report: "Current Trends in Emerging Sectors." I'm just going to read the headers: Mobile Commerce, Social Commerce, Personalization, Sustainability, AR/VR in e-commerce. ^98989898This is extensive. ⁹⁹It researched 119 websites to find what I was looking for. ¹⁰⁰This would have taken me forever. ¹⁰¹

Dalton: It's talking about 3D printing for personalized products on the fly. ¹⁰²For biotechnology, they're saying you could have personalized supplements and customized skincare routines per person, which I think would be pretty cool. ¹⁰³ There's a growing interest in experiences over material possessions. I agree with that. ¹⁰⁴I've seen that trending for years now. ¹⁰⁵There's such a lack of connection in the world that people are yearning for real-world experiences versus virtual experiences like social media. ¹⁰⁶This looks pretty good. ¹⁰⁷

Dalton: Let's go over to the EV report. This one made a nice table of contents for us. 108Introduction, Battery Technology Advancements, Charging Infrastructure Challenges, Environmental Impact, Economic Impact, Leading Countries and Policies, Consumer Trends. 109Pretty good. 110It says the leading countries are Norway, Iceland, Sweden, Netherlands, and China. 111That makes sense. 112It has the link of where it's getting this information from. 113

An Unexpected Discovery: The Truth About Yo-Yo Weapons

Dalton: So then the last one, let's look at Yo-Yo, which is a little bit less complicated. ¹¹⁴At one point I thought a Yo-Yo was some kind of self-defense weapon and then eventually became playful. ¹¹⁵Maybe that was just fake news I got a long time ago. ¹¹⁶I don't see anything about using a yo-yo as defense. ¹¹⁷ Let's see. There is no historical evidence that yo-yos were ever used as weapons, but the idea was used as a marketing tactic by the Duncan Yo-Yo Company. ¹¹⁸However, there is some evidence that yo-yos were used as weapons in the Philippines. ¹¹⁹ So I did get fake news, but it wasn't necessarily super fake news. They got me. ¹²⁰That's so funny. ¹²¹

Final Thoughts: Why Live Demos and Shipping Matter

Dalton: The e-commerce prompt was probably the most complicated I've ever requested from the AI analyst, and it did a good job, I feel. ^122122122From the skimming I did, what it was saying seemed legitimate. ¹²³And if you don't like what it's saying, you just go to the link. ¹²⁴It has all of the references right below it. ¹²⁵For what it is, it's pretty much free. ¹²⁶For 20 bucks a month, if you could run that once a week and it saved you two hours, it pays for itself. ¹²⁷

Dalton: There are two things why I think this is cool. One, it shows you the websites it goes to and links all your information in the output. 128Two, Google announced this in the summer, said it would release later this year, and they delivered. 129They've said they're going to do it, and they did it. 130With these other AI companies, they make announcements, they don't do live demos. 131They do pre-recorded live demos to make it seem like it's live, but it's not. 132 Google does live demos. Meta does live demos. 133These other folks like Anthropic and OpenAI, theirs aren't live demos, they are pre-recorded. 134

Do you really believe in what you're building if you can't live demo? ¹³⁵

Dalton: That's a big part of announcing your product to the public. ¹³⁶If you're blatantly not confident enough to live demo consistently, then what's really going on here? ¹³⁷I hope that you found this episode interesting and useful. ¹³⁸ I would love it to just kick off in the morning and I have some set agents and they do their AI analyst thing and then they come back to me with reports and things that I need to approve. I could delegate to almost a virtual team of these AI analysts. ¹³⁹That's what I would like. ¹⁴⁰Not there yet, but baby steps. ¹⁴¹

Dalton: Wherever you are in this world, good morning, good afternoon, good evening. Have a great day. 142 I hope that you enjoyed listening to this episode and hope that you listen in next week. See ya. Goodbye. 143

RESOURCES MENTIONED

• Gemini ¹⁴⁴
• Google ¹⁴⁵
• CoTracker (research paper) ¹⁴⁶
• Grayball (podcast episode) ¹⁴⁷
• Duncan Yo-Yo Company ¹⁴⁸
• Meta ¹⁴⁹
• Anthropic ¹⁵⁰
• OpenAI ¹⁵¹
• Wikipedia ¹⁵²
• YouTube ¹⁵³
• Apple Podcasts ¹⁵⁴
• Spotify ¹⁵⁵

INDEX OF CONCEPTS

3D printing, AI Analyst, Anthropic, Apple Podcasts, Augmented Reality (AR), Biotechnology, China, CoTracker, Context Window, Custom Gems, Dalton Anderson, Deep Research, Diplo, Duncan Yo-Yo Company, E-commerce, Electric Vehicles (EVs), Flash (model), Gemini, Google, Grayball, Iceland, Instagram, Meta, Mobile Commerce, Netherlands, Norway, OpenAI, Prompt Engineering, Social Commerce, Spotify, Sweden, TikTok, Ultra (model), Virtual Reality (VR), Wikipedia, Yo-yoing, YouTube