Latest / AI Fire Daily / #42 Robin: The Chat-to-Action Pipeline - Why Hermes AI & NotebookLM Just Killed the App Switch
Transcript
- 0:00What if a single passing thought on your commute
- 0:03could automatically build a fully researched,
- 0:06organized project plan before you even sit at
- 0:09your desk? I mean, that sounds total like science
- 0:11fiction. Right. Beat. It really does. But you
- 0:15can actually build this today. You just need
- 0:17to connect your personal archives to an AI action
- 0:20layer. Yeah, it's a massive shift in how we approach
- 0:23our daily work, honestly. Welcome to the Deep
- 0:25Dive. Today, our mission is unpacking a fascinating
- 0:28workflow. We're looking at how two specific tools,
- 0:32Hermes and Notebook LM, create a 247 personal
- 0:36research assistant. And we're moving from a world
- 0:38where we just, you know, passively query information
- 0:41to one where our systems actively execute tasks
- 0:44for us. We're going to explore what these tools
- 0:46do. We'll look at why combining them is such
- 0:49a breakthrough and dive into the standout features.
- 0:51Plus the real world applications. Exactly. And
- 0:54crucially, we'll cover the severe security risks
- 0:56you need to navigate. But before we build the
- 0:59machine... We need to understand the parts. Right.
- 1:02So let's start with the research brain of the
- 1:04operation. That is Notebook LM. Most people are
- 1:07probably familiar with it by now. Yeah, it's
- 1:09this incredibly powerful platform. It's designed
- 1:12specifically to store and synthesize your personal
- 1:16sources. So you upload your own stuff. Exactly.
- 1:19Dense PDFs, long strategy documents, or like
- 1:23web links. And then it interacts with you based
- 1:25entirely on those specific materials. It essentially
- 1:28walls itself off from the broader internet. It
- 1:30does. It turns your private data into highly
- 1:33accurate summaries. It builds intricate mind
- 1:36maps. It even generates those those incredibly
- 1:40popular podcast style audio overviews. Which
- 1:44are wild to listen to. They really are. But the
- 1:46key is that it grounds every single response
- 1:48in your provided data. Right. And then we have
- 1:52the second player in this workflow, Hermes. Yeah.
- 1:54So if Notebook LM is the brain, Hermes is, well,
- 1:57it's the ham. The action layer. Exactly. Hermes
- 1:59operates as the control center for this entire
- 2:02workflow. It doesn't just passively retrieve
- 2:04information. It actually triggers actions. Right.
- 2:06It triggers external workflows, like it can draft
- 2:08your scripts, save your notes into your system,
- 2:10or even schedule follow -up reminders. So to
- 2:13synthesize this, it's like having a genius researcher
- 2:17locked in the archives. That's Notebook LM. And
- 2:20at the front desk, you have a highly capable
- 2:22executive assistant. That's Hermes. And finally,
- 2:26they have a walkie -talkie to talk to each other.
- 2:29That is a perfect analogy. Notebook LM does the
- 2:32heavy lifting of reading dense texts, and Hermes
- 2:35parses your intent and delivers the results.
- 2:37I do have to ask, though, about a common problem.
- 2:40How do we know the AI won't just invent fake
- 2:43facts? Ah, hallucinations. Yeah, that's a massive
- 2:46issue when automating research. Right. Because
- 2:49you aren't watching it work. But the answer lies
- 2:51in Notebook LM's core design. It has a highly
- 2:55constrained retrieval system. It forces the model
- 2:58to restrict its answers exclusively to the documents
- 3:01you uploaded. So if it's not in the text. It
- 3:03just refuses to answer. It won't guess. So it
- 3:06only reads your uploaded documents, not the whole
- 3:07internet. Exactly. And that hard limitation is
- 3:10what makes it trustworthy. Now, because both
- 3:13of these tools are so powerful alone, we have
- 3:16to ask why we need them to talk to each other.
- 3:18Right. Why bother connecting them? Yeah. Beat.
- 3:21And the answer is friction. Friction is the enemy
- 3:24of execution. Just think about the standard notebook
- 3:27LM workflow right now. You have an idea. You
- 3:29have to break your focus. Totally. You manually
- 3:32open a browser, navigate to the site, log in,
- 3:34scroll to find the notebook, type your query.
- 3:37And then copy and paste the output. Right. Back
- 3:39into whatever app you were originally using.
- 3:41It's exhausting. I have to admit, I still wrestle
- 3:44with having 20 tabs open just to research a new
- 3:46microphone. Huh. We all do. The context switching
- 3:50is brutal on your working memory. Yeah. But Hermes
- 3:52eliminates this. Walk me through the actual sequence.
- 3:55Sure. So you just message Hermes in your chat.
- 3:58Okay. Beat. Hermes analyzes it and decides if
- 4:02Notebook LM is needed. It makes that choice itself.
- 4:05Yep. Beat, beat. Then Notebook LM processes the
- 4:08sources. Beat. And finally, Hermes returns the
- 4:11answer and helps you act on it. That's amazing.
- 4:13But why is a simple chat interface actually superior
- 4:16to a dedicated visual dashboard? Dashboards give
- 4:20you control. Because chat removes the barrier
- 4:22between having an idea and executing it. It captures
- 4:26your raw intent immediately. You don't have to
- 4:28navigate drop -down menus. Exactly. You just
- 4:31state your goal in plain English, and Hermes
- 4:33acts as the translation layer. Chat makes complex,
- 4:37multi -step research as easy as texting a smart
- 4:40friend. Yeah, intent -driven execution. It's
- 4:43a total paradigm shift. Now that the friction
- 4:45is gone, let's look at the wild features this
- 4:48actually unlocks. This is the fun part. Let's
- 4:50talk about voice notes. Okay. Imagine you're
- 4:53walking down a busy street. No screen. You just
- 4:56hold a button on your phone and ask for research
- 4:58on, like, Ralph Lauren jumpers. You're just rambling
- 5:01naturally. Rambling naturally. Hermes transcribes
- 5:04it, extracts the intent, and turns it into a
- 5:07notebook LM task instantly. You don't even touch
- 5:10a keyboard. Not at all. And it gets better. You
- 5:12know those podcast -style audio overviews and
- 5:14mind maps? Yeah, they're great. You can retrieve
- 5:16those directly in the chat. Hermes pulls those
- 5:18heavy files right into your thread. You just
- 5:21ask? and the image renders right there but here's
- 5:24where it gets crazy you can introduce external
- 5:27connective tools like nan okay let's clarify
- 5:31that n8n a tool that visually connects different
- 5:34apps together perfect it's the digital plumbing
- 5:37and you can also use mcp mcp a universal plug
- 5:40that lets ai control other software right so
- 5:43you use mcp to connect hermes to your nnn workflows
- 5:46now your assistant is actively moving data across
- 5:49your digital life whoa Two sec silence. Imagine
- 5:53turning a passing thought into an automated daily
- 5:55morning briefing. Just waking up to a fully formatted
- 5:58intelligence report. That's incredible. But I
- 6:01have to push back here, Beat. Is it truly safe
- 6:04to let an AI agent use MCP to control other apps?
- 6:08That is a very valid concern. The text issues
- 6:11a strong warning about this exact thing. You
- 6:14must be incredibly careful. Especially with authentication,
- 6:16right? Yes. Do not ever paste private credentials,
- 6:19API keys, or cookies into insecure web -based
- 6:23environments. Because if it's compromised, they
- 6:25have your digital identity. Precisely. Never
- 6:27paste them into random tools. Keep your digital
- 6:30keys safe. Never paste them into random or unverified
- 6:33tools. Exactly. Security has to be the top priority.
- 6:35Let's take a quick break here. Insert mid -roll
- 6:38sponsor read here. Welcome back. So knowing these
- 6:42capabilities is really exciting, but we need
- 6:45to ground this in reality. Yeah, we have to talk
- 6:47about the setup. Exactly. How hard is the actual
- 6:51plumbing to put this together? I won't sugarcoat
- 6:53it. It is unglamorous. There are some highly
- 6:56necessary, tedious steps. Walk us through it.
- 7:00First, you have to download the notebook LM skill.
- 7:03Because there's no official API yet, this is
- 7:06a workaround. You're manually installing it into
- 7:08your environment. Right. Then comes the browser
- 7:10authentication. This is the tricky part. Because
- 7:13you're dealing with live sessions. Yeah, you
- 7:15have to connect it to your Google account by
- 7:16extracting live access tokens and cookies directly
- 7:19from your browser's developer tools. Which goes
- 7:22back to our critical safety warning. Exactly.
- 7:24You have to handle those cookies very carefully.
- 7:26Once you feed them into Hermes, you run a test
- 7:28prompt. Like what? You just say, find one of
- 7:31my last three notebooks. If it works, the bridge
- 7:34is up. But relying on an unofficial integration
- 7:37skill, isn't that system incredibly fragile?
- 7:41It is. Because it's unofficial, any future updates
- 7:44to the Notebook LM interface could temporarily
- 7:46break your entire workflow. Unofficial tools
- 7:49can break, so expect to tweak the plumbing occasionally.
- 7:53Yeah, that's the price of being an early adopter
- 7:55right now. Assuming you're willing to manage
- 7:57that setup, who is this actually built for in
- 8:00the real world? The applications are actually
- 8:01vast. Let's look at creators first. A creator
- 8:05can dump historical scripts and analytics into
- 8:08a notebook and have Hermes generate 10 new video
- 8:11hook ideas that match their specific voice. That
- 8:14cures blank page syndrome instantly. Totally.
- 8:17Or founders. They can upload 50 dense competitor
- 8:20PDFs. Nobody wants to read 50 PDFs. Exactly.
- 8:23So they ask Hermes to find positioning gaps in
- 8:26the market based on that research. What about
- 8:28students? Oh, it's huge for them. They turn dense
- 8:32academic papers into beginner -friendly study
- 8:34guides, pulling out only the key terms. And everyday
- 8:37shoppers. Yeah, say you're comparing $500 standing
- 8:41desks. You just feed the specs into the notebook,
- 8:43and Hermes cross -references the warranties and
- 8:46materials without you opening 20 tabs. It saves
- 8:48so much manual effort. And travelers. You're
- 8:52going to Rome. You put historical articles and
- 8:55maps into a notebook, and Hermes generates a
- 8:58day -by -day guide to the Colosseum. The underlying
- 9:01theme here is really interesting. Beat. It stops
- 9:04your second brain from becoming a graveyard of
- 9:07untouched notes. Oh, absolutely. We all have
- 9:09those folders of saved PDFs we never open. This
- 9:13actually makes them useful. But I have to ask
- 9:15thoughtfully here, is there a danger in letting
- 9:17AI synthesize everything? Do we lose our own
- 9:21judgment? That's a fair question. The cognitive
- 9:23friction of reading is important, but the AI
- 9:26just organizes and retrieves the data. So it
- 9:29preps the landscape. Right. It's meant to prep
- 9:31the information, not make the final human decision.
- 9:33It serves up the menu, but you still have to
- 9:36pick the meal. That's a great way to put it.
- 9:37You still make the call. So having seen the menu
- 9:40of possibilities, it's time for the ultimate
- 9:42cost -benefit analysis. Let's weigh it out. Before
- 9:46the listener decides to build this, what are
- 9:48the ultimate benefits? Well, the gains are pure
- 9:50speed, the reuse of forgotten knowledge, and
- 9:53drastically less context switching between apps.
- 9:56But the risks are real. You have that technical
- 9:59setup curve for beginners, the danger of leaked
- 10:02credentials if you mishandle cookies, and the
- 10:05inherent variability of AI output quality. It
- 10:08can still hallucinate if the grounding fails.
- 10:11What is the single biggest behavioral trap once
- 10:14someone gets this working? Over -reliance. Without
- 10:18a doubt. Specifically, letting automations run
- 10:20without human review. Like sending emails. Yes.
- 10:23Do not set up a workflow that auto sends an email
- 10:26or a client report without you reading it first.
- 10:28Never let the AI auto send an email or report
- 10:31without reading it first. Absolutely never. You
- 10:33are still responsible for the output. As we wrap
- 10:36up, let's look at the broader paradigm shift
- 10:38here. Beat. We are moving away from simple chatbots
- 10:41that just answer questions. Right. The old model
- 10:43was just a fancy search box. Exactly. And now
- 10:46we're moving toward integrated AI systems that
- 10:49research, organize, and execute physical digital
- 10:52tasks from a single interface. It's an active
- 10:55assistant, not a passive tool. It really is.
- 10:58So I want to leave you with a final thought to
- 10:59mull over. Beat. Think about the biggest, messiest
- 11:04folder of saved articles or PDFs you currently
- 11:06have sitting on your desktop. We all have one.
- 11:08Ask yourself, what would happen if that dead
- 11:11information could suddenly talk back to you and
- 11:13draft your next project? The possibilities are
- 11:16incredible once you remove the friction. Thank
- 11:18you for joining us on this deep dive. Stay curious.
- 11:20See you next time.