26:19The Next New Thing · Reaction
Make Money, Cut Costs, and Run Your Life With Meta's Muse
A creator reacts clip by clip to Meta's free AI agent, grading the calls, purchases, and file reads it actually completed on camera.
September 29thThe argument in one line.
The GrokBot workflows that actually save time and money are the ones triggered by webhooks the instant a real event happens, not the ones polling on a timer, and every irreversible action still needs a human approval step.
Read if. Skip if.
READ IF YOU ARE…
- You already run or are building AI agent automations and want concrete prompt patterns, not just the concept of 'AI agents.'
- You're deciding between scheduled polling and event-triggered webhooks for an automation and want real cost and delay tradeoffs, not theory.
- You want to see several real people's actual GrokBot setups critiqued by someone more technical, to copy what works and skip what doesn't.
SKIP IF…
- You've never used an AI agent platform and want a from-scratch introduction; this assumes you already know what a routine or a bot is.
- You're looking for an official GrokBot tutorial from the product itself; this is two outside creators reacting to other people's builds, not a walkthrough from the maker.
The full version, fast.
Two hosts react to eleven real GrokBot setups other creators posted publicly, grading what works and what doesn't. The recurring lesson: routines that poll on a timer, like checking email or a meeting app every few hours, waste tokens and add delay, while webhook-triggered routines wake the agent the instant a real event happens. The strongest builds pair automation with a human-approval gate on anything irreversible, like a Stripe refund or a sent email, and delegate bounded projects to sub-agents so the main bot stays available. Zapier's MCP server extends GrokBot's reach to 9,000+ apps it doesn't natively connect to. Build event-triggered, not scheduled, and keep humans approving anything that can't be undone.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:18
01 · Cold Open: 11 GrokBot Use Cases
Andrew Warner frames the episode: eleven real GrokBot workflows other creators have published, each with his and Andrew Kordampalos's take on what's worth copying.
00:18 – 02:42
02 · Meeting Notes to Five Agents
A Granola meeting note is auto-routed through a GrokBot routine to five named agents (competitor research, carousel ideas, an invoice) with no extra prompting; the hosts flag the 3-hour polling schedule and the lack of a human checkpoint before agents act.
02:42 – 04:12
03 · Better Meeting Automation
The fix for the prior segment's timer problem: Zapier can ping GrokBot the instant a meeting ends, and GrokBot accepts webhooks as a trigger so it wakes immediately instead of on its next scheduled sweep.
04:12 – 06:00
04 · Local Computer Access
GrokBot reads, renames, and moves files on the actual desktop it's running on, diagnoses a slow computer by identifying which app is eating memory, and can restart a crashed local server, all triggerable from a phone.
06:00 – 08:51
05 · AI Recommendation Tracking
An agent asks ChatGPT, Claude, and Perplexity the kinds of questions a cold prospect would ask, without naming the product, to check whether and how it gets recommended, then reports what to change.
08:51 – 11:24
06 · Zapier + GrokBot
Sponsor segment: Zapier's MCP server exposes 9,000+ apps as callable tools to GrokBot, Claude, and ChatGPT; a real story of a $140/month plan running up $265 in overages in two days from unmonitored timed routines.
11:24 – 14:06
07 · Subscription Audit
A 'Penny Pincher' bot scans receipts and billing email for subscriptions to cancel, renegotiate, or pause; the hosts debate scraping email receipts versus a direct bank connection as the more complete source.
14:06 – 17:06
08 · Sub-Agents
A bounded project (build a landing page, report only when done) is handed to a spun-up sub-agent so the main agent stays available; the hosts warn that sub-agents chatting with each other on their own can spiral out of visibility.
17:06 – 21:18
09 · Webhooks
A webhook is explained as a doorbell: hit a URL and the waiting bot wakes instantly with whatever payload arrives. Demoed live with a form submission, with WhatsApp and Telegram raised as extensions.
21:18 – 25:30
10 · Refunds With Approval
'Holly Helpdesk' sweeps support inboxes hourly, fixes routine issues, and requires explicit human approval via a Stripe agent action before issuing a refund; also reviews the week's tickets for gaps in support docs.
25:30 – 28:03
11 · Facebook Marketplace
Listing an item becomes one prompt covering five chained tasks: price-check, scan local competition, draft a description in the seller's own voice, post, and auto-discount weekly; it negotiates with buyers without escalating every message.
28:03 – 30:18
12 · Email & Calendar
The agent drafts full-thread email replies in the user's voice held for approval, edits the calendar from a screenshot, and takes voice instructions by phone; a host pushes back that it can't replace a calendar's at-a-glance view.
30:18 – 33:27
13 · Inbox Triage
A detailed inbox-triage prompt batch-archives safe email, summarizes what needs awareness only, and drafts replies for what needs a response, never sending without approval; a single correction about time window gets applied to the agent's other scheduled jobs on its own.
33:27 – 35:33
14 · Upgrade a Bot From X
A post spotted on X gets forwarded to the agent with one question: upgrade an existing bot, or build a new one? A separate 'Stashy' bot saves interesting posts for later recall.
35:33 – 38:42
15 · Recurring Appointments
'Booker' checks the calendar for a recurring need like a haircut, proposes an open slot, confirms with the user, then joins the booking site's online waitlist so the user is first in line by the appointment time; extended to a dentist and a residential service portal.
38:42 – 40:03
16 · Teach a Task
Recording a manual walkthrough once on a complicated site lets GrokBot turn it into a reusable skill; the hosts also compare notes on giving an agent limited password-manager access instead of full vault access.
40:03 – 40:24
17 · GrokBot for Teams
A short tease for a follow-up video about bringing GrokBot agents into Slack so they can work with other humans on a team, not just solo.
Lines worth screenshotting.
- A $140/month GrokBot plan ran up $265 in overage charges in just two days once scheduled routines were left checking things on a timer.
- GrokBot accepts webhooks as a trigger, so an event like a finished meeting or a form submission can wake it instantly instead of waiting for its next scheduled check.
- Scraping email receipts for subscriptions misses billing that financial institutions see but that never generated a matching email, making a direct bank connection the more complete source.
- Zapier's MCP server exposes over 9,000 apps and 40,000 actions as tools any MCP-compatible AI client, including Claude and ChatGPT, can call directly.
- Checking whether an AI engine recommends your product only works if you ask about the problem category, not your product by name, since naming it guarantees a mention.
- A support agent with a human-approval gate on refunds still earned unsolicited customer praise for responsiveness within a week of going live.
- An agent correctly inferred that a one-off 'only the last 24 hours' correction should also apply to its own recurring scheduled jobs, without being told to update them.
- GrokBot can read messages directly on WhatsApp, something a comparable agent tool couldn't do at the time of this review.
- Teaching an agent a complicated browser task once, by recording yourself doing it, turns that single walkthrough into a reusable, repeatable skill.
- An appointment-booking agent joined a haircut chain's online waitlist ahead of a confirmed time, so its user was first in line the moment they arrived.
- The stronger prompt pattern for a multi-step task was a numbered list of explicit steps, not a single sentence describing the end goal.
- Running every support approval through one continuous chat thread becomes hard to track once several refunds or fixes are in flight simultaneously.
Replace Timers With Event Triggers
AGENT WORKFLOW PATTERNS
Across eleven creators' GrokBot setups, the workflows that actually save time and money are event-triggered rather than scheduled, and keep a human approval step on anything irreversible like a refund or a sent email.
02Meeting Notes to Five Agents
- One Granola meeting note, routed through a GrokBot routine, auto-delegated to five differently-skilled agents (competitor research, carousel ideas, an invoice) without a separate prompt per task.
- Running the handoff routine on a 3-hour timer burns tokens on nothing happening and introduces a delay between the meeting ending and the work starting.
- Letting agents take action immediately after a meeting, with no human checkpoint, is a setup for compounding mistakes; a single main agent presenting choices to approve is safer.
03Better Meeting Automation
- A scheduled routine polling every few hours is a workaround; the fix is an event trigger that fires the instant the real-world event, like a finished meeting, actually happens.
- GrokBot accepts webhooks as a trigger, so a tool like Zapier can wake a dormant bot the moment an external event occurs instead of waiting for its next scheduled sweep.
04Local Computer Access
- GrokBot can access the local files on whatever computer you're actually using, not just a shared cloud machine, so it can analyze, rename, and move a specific file in your Downloads folder.
- It diagnosed a slow computer by identifying which running apps were eating memory, and it can restart a crashed local server, all triggered remotely from a phone.
- It can't yet click around a desktop with a mouse the way a browser-use agent can; file and process management is its current ceiling on local control.
05AI Recommendation Tracking
- AI search increasingly replaces Google for product discovery, so being invisible in AI answers is now as costly as not ranking in search results.
- The AI-recommendation test only works if you ask about the problem category, not your product by name: naming it guarantees a mention and hides whether cold prospects would ever surface you.
- An agent can run this check across ChatGPT, Claude, and Perplexity in a browser it's pre-authorized into, then report which part of the product or site content needs to change to get recommended.
06Zapier + GrokBot
- A routine set to check email or a calendar 'every hour' still burns tokens waiting and means real delay between an event and the response, the same timer problem as the first segment.
- Zapier MCP exposes 9,000+ apps and 40,000+ actions as callable tools inside GrokBot, Claude, ChatGPT, or Cursor, covering integrations too niche for GrokBot's native connector list.
- An unmonitored usage-based plan can spike fast: a $140/month plan ran up $265 in overages inside two days once always-on routines were left checking things on a timer.
- Zapier can proxy access to accounts GrokBot isn't authorized for directly, such as a work email already approved for Zapier but not for a third-party agent.
07Subscription Audit
- An agent scanning receipts and billing email found a forgotten kids' app subscription to cancel, an insurance renewal worth renegotiating months in advance, and an unused Audible subscription worth pausing rather than canceling.
- Scraping email receipts under-counts real billing, since financial institutions see more charges than match an inbox; a direct, read-only connection to the bank is the more complete source of truth.
- Some vendors only disclose line-item detail in the email receipt, not the bank transaction, so the strongest setup checks both rather than picking one source.
08Sub-Agents
- Spinning up a dedicated sub-agent for a bounded project, like building a landing page with a stated URL and success criteria, frees the main agent to stay available for conversation.
- A well-formed delegation prompt states the outcome, the goal, and the success criteria up front, which let the agent hand a coding tool a strong enough brief to get a working result in just over a minute.
- Letting agents spin up and chat with other sub-agents on their own can spiral out of visibility fast; keeping delegation deliberate, not automatic, keeps you able to track what's running.
- Routing a sub-project into a separate tool's own virtual machine, like Cursor's, shifts that work's token cost off the main agent's subscription instead of adding to it.
09Webhooks
- A webhook is just a URL: hit it, and the bot waiting on it wakes up with whatever payload you sent, instead of needing to be polled or scheduled to notice.
- Most creators default to time-based routines because they're easier to set up, but timers mean waiting for the next scheduled check even when the triggering event already happened.
- Any event with a URL can wake an agent: a form submission, a Slack message, a CRM record, or, with a bridging service, a Telegram or WhatsApp message.
- GrokBot can already check WhatsApp directly, which a comparable tool couldn't do yet, making WhatsApp monitoring one of the clearer current differentiators between agent platforms.
10Refunds With Approval
- A support agent that sweeps the inbox hourly, handles routine fixes, and requires explicit approval before issuing a refund gets both speed and a safety net tied to an actual Stripe request.
- A hands-off support bot still produced a measurable result: unsolicited customer praise for support responsiveness within a week of turning it on.
- Having the agent review the past week's tickets and flag gaps in the support docs turns every resolved ticket into material that prevents the next one.
- Running everything through one continuous chat thread gets hard to track once several approvals are in flight at once; separable, disposable threads per issue would scale better.
11Facebook Marketplace
- Listing one item well is actually five chained tasks, price-check the model, scan local competition, draft a description in the seller's own voice, post, and auto-discount weekly, given as a single prompt.
- The agent consulted the seller's own 'writer' bot to match how they'd written past listings, instead of generating generic marketplace copy from scratch.
- The edge over a platform's own marketplace assistant is negotiating with buyers proactively instead of returning to ask permission for every reply, which matters once a listing gets more than a couple of messages.
- Structuring the prompt as a numbered list of explicit steps, not just a stated end goal, was the stronger pattern for this kind of multi-step task.
12Email & Calendar
- An agent can draft full-thread email replies in the user's own voice and still hold them for approval, which keeps 'my bot handles email' from meaning 'my bot sends email unsupervised.'
- Calendar actions like booking, editing, or adding an event can be driven from something as unstructured as a screenshot, not just a structured calendar API call.
- Voice conversations with the agent, checking messages and issuing instructions by phone, are now a real interface, not just chat and dashboards.
- An agent still can't replace the visual overview a calendar app gives you at a glance; chat-only interaction has a real ceiling for anything spatial or at-a-glance.
13Inbox Triage
- A well-specified inbox prompt, batch-archive what's safe without reading each one, summarize what needs awareness only, draft full replies for what needs a response, never send without approval, is reusable across any inbox.
- Telling the agent one missing constraint, only the last 24 hours, made it also go fix its own related scheduled jobs to use that same window going forward, without being told to touch those jobs.
- A single detailed run surfaced real value: dozens of safely archivable promo emails, cold pitches, and already-replied threads identified correctly, plus informational items that needed no action at all.
14Upgrade a Bot From X
- Instead of researching a new technique from scratch, the agent was handed a competitor's post and asked a binary question first: upgrade what we already have, or build something new?
- A dedicated 'stash' bot that saves any interesting post for later recall turns passive scrolling into a growing, searchable reference the main agent can pull from when relevant.
- Sending an agent a link to a how-to video and asking it to implement the technique is a legitimate, normal workflow, not something to be sheepish about.
15Recurring Appointments
- A recurring-need agent can proactively check the calendar, propose a specific open slot, and only book once you confirm, instead of picking a time unilaterally.
- Booking isn't just scheduling: the agent also joined a booking site's online waitlist ahead of the appointment time, so the user was first in line the moment they walked in.
- Any site with a working browser-based booking flow is fair game, even a bad one; the use case extended to a dentist and a residential service portal by automating around the janky interface.
16Teach a Task
- Recording yourself completing a complicated, non-standard browser task once lets the agent turn that single walkthrough into a skill it can reproduce reliably going forward, without needing a clean API.
- Full password-manager access for an agent is avoidable: moving only the specific passwords it needs into a separate shared collection limits exposure versus granting broad vault access.
Terms worth knowing.
- GrokBot
- The AI agent platform central to this video; runs persistent bots with scheduled routines, local file access, webhook triggers, and sub-agent delegation.
- Routine
- A GrokBot task that runs on a fixed schedule, such as every 3 hours, rather than being triggered by an event. Simple to set up but wastes tokens and adds delay.
- Webhook
- A URL that, when hit with a payload, instantly wakes a dormant agent to act, instead of the agent having to check on a schedule.
- MCP (Model Context Protocol)
- A standard that lets an AI client call external tools and apps directly. Zapier's MCP server exposes 9,000+ apps as callable actions to Claude, ChatGPT, Cursor, and GrokBot.
- Sub-agent
- A separate agent spun up and delegated an entire bounded project, reporting back only once it's done, so the main agent stays free for other conversation.
- AEO (AI Engine Optimization)
- Optimizing a product or brand so AI tools like ChatGPT, Claude, and Perplexity recommend it in answers, the AI-search equivalent of traditional SEO.
- Agent action
- An explicit approval step, such as a Stripe refund request, that an agent surfaces to a human before executing something irreversible.
Things they pointed at.
00:18toolGranola (meeting notetaker)
Lines you could clip.
10:20
“My $140 a month plan suddenly had overages of $265 for just two days of extra usage.”
17:10
“Just think of it like someone ringing on your doorbell.”
29:10
“Let's stop pretending that everything needs to happen via chat.”
21:10
“Grokbot can actually check your messages on WhatsApp versus Muse, which cannot check your messages on WhatsApp.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphor
Grokbot is the best and easiest to use agent out there, but I want to see what we could do with it. I've got a collection of videos that I've seen on YouTube. I've tried them.
I've got feedback on what works and what doesn't and how you can use Grokbot to make money, grow your business, or save money. Presented by Zapier, the AI automation company. Let's look at the first one.
This is from Brock. He uses it to have his Grokbot at the end of a meeting do all the tasks that came up in the meeting. Check out how he does it, and I've got an opinion on this.
Show you exactly how granola works. This is what a note looks like. And he's using granola as a meeting note taker.
So this is from a meeting that I actually had a little bit ago where I kind of break down kind of all the things that I need to work on. I just basically just gave it a brain dump of everything that I need to do over the next couple of days. And what it did is it automatically took the transcript, which I could go and click on, and it turned that transcript into a list of the main talking points and action items from this call.
Okay, pretty interesting. We all get this, but the next step I thought was the fascinating part. Hey, I want you to go and pull my last meeting inside of Granola.
Now take a look at this. Not only did it pull from the Granola note that I had and pulled all of the different action items from it, but it automatically routed from that note these different action items for these specific AI agents. And if I click on this, you can see that it messaged all five of these different agents from that one note that I created.
Agent Tracker, Short Form, Content Bot, Brand Partnership Agent, and Invoice Agent are now all working on these tasks that I got from Granola. And over on the left -hand side, as you can see, there's these little blue dots. And that just means these AI agents have been working.
So let's go take a look at what they did. If I click on my Content Bot, it set up a routine that automatically scrapes my YouTube competitors. That is pretty cool.
I literally didn't have to do anything. This is all from that. meeting that I had that Granola took these action items from and is going and doing these tasks for me now.
And it knows exactly which agent to delegate these to. And then if I click on my short form agent, it then went and researched some different Instagram carousels since it knows that I need to work on creating five different Instagram carousels for the week. So now I have all these ideas ready to go.
And all I have to do is pick which ones I want to move forward with. And it will automatically route this task to my IG carousel agent, which we created earlier on in this video. And if I come inside of my invoice agent, you can see it literally went and drafted this invoice here.
And this is ready to go and ready to be sent to this client over Gmail. And of course, I'll have a link if you want to see the rest of his video below. But I want to show you, Andrew, specifically the setup because I have an opinion on his setups.
Here's how he does it. Routine right here that is called granola action. So you know routines in Grokbot are basically their scheduled tasks.
And each bot has a set of scheduled tasks called routines. sweep. And this runs every three hours on weekdays.
So if I click on this, there's just a pretty basic set of instructions of exactly how this task is supposed to go down. And if I look at this, it's actually pretty in -depth, honestly, which is great because hopefully this will never make any mistakes. Okay.
Here's my problem with it. Every three hours is a little delayed. I think what he needs to have is as soon as the meeting is done.
So Granola needs to ping Slackbot, excuse me, not Slackbot, needs to ping his Grokbot and tell it, I have a new meeting with action items, get to work. And not to make this into a Zapier ad, but Zapier is the only way to do that, right? It doesn't work natively with Granola.
Yeah. And also Grokbot accept webhooks as a trigger. So you can wake it up from Zapier, do something and then start your routine within the app, which is.
amazing if you combine both yeah i think the problem with him having it run every three hours is it's just wasting tokens and it's leading to potential mistakes and also you want that immediacy the other suggestion that i have is He's having his agents immediately take action at the end of a meeting. I think that's just a recipe for disaster.
He should have one agent, like a main agent, come in and say, look, here are the tasks that I noticed in the meeting. Select from these buttons which ones you want us to act on, and we will take action. So this demos well.
I think there are a couple of twists, but I love this use case. Let's go on to the next because I have so many of them. All right, this is a faster one, and it has bigger implications.
It's about how Grokbot can use your computer. local file access so that means that yes they have a cloud computer that they all share but whatever computer that you're actually working on they can access local files so right here you can see that i gave it a file path which was in my downloads i said analyze this file and tell me what's in it it grabbed it it analyzed it i told it to rename it and then move it and it actually was able to rename it and move it so if i go into my new thumbs this is the file it was talking about right here it used to be in my downloads and it was just called like you know orange blah blah blah And then it looked at it, it renamed it to zero to pro, and then it moved it inside of this new thumbs folder.
And it did this locally on my own desktop. So the thing that I wish that Grokbot had that Codex has is the ability to actually start clicking buttons and use a computer just like a human being with a mouse. It doesn't have that, but it can move folders like this.
And the other thing it was able to do for me is... As I was in an Uber going home yesterday, I said, why is my computer so slow? I was able to go to Rockbot and say, go figure out the memory issue.
And it saw what was eating memory. It turned out it was the Codex app and my Chrome. And it helped that way.
Have you used it at all to control your computer? Yeah, and it's very handy because you can use it from any device. So you can actually use your mobile phone, ask questions.
about your desktop and it goes uses your desktop and make changes one day i had an app that i'm using to connect remotely and it was crashed i couldn't use it and i said what's wrong and it technically checked it and restarted the whole server for me and then i connected from the phone Really helpful feature. I forget to use it, and I'm glad to see some examples that will jog my memory for next time.
Okay, this is one that will help you get ranked in AI results. One of the things that I know about Zapier is it keeps getting recommended by agents because you all are very good at AEO. Here's how people can watch how agents recommend them.
I like this from AI Labs. Previously, if you had an app and wanted people to know about it, you mainly had to make sure your product ranked well on search engines like Google because that's where people were searching for solutions to their problems. But now people also use AI to search for things, so you need to make sure your product shows up there as well.
More people are now asking AI what products they should use, and these models search the web before giving them an answer. So if your product isn't showing up in those answers, a lot of people might never find out about it. That's harder to check than a normal Google search.
With Google, you can search for something and see where your product appears in the results. But with AI search, it's harder to know whether your product is actually being recommended. You can create a bot that goes to chat GPT, Claude, and Perplexity and asks the kinds of questions a real person might ask when they're looking for a product like yours without mentioning your product by name.
That part matters because if you ask an AI about your product by name, it can just tell you about it. But people who don't already know about your product aren't going to search for it by name. what you really need to know is whether your product shows up when nobody has mentioned it.
So when you ask it to check, it will open up all the AI tools in the browser, give them prompt and review their answers. Based on their answers, it will give you a thorough report and will also mention what part of your app would need to be changed. It's a little disconcerting that the AI labs videos don't have a face.
It feels... It makes me anxious somehow. But this is a very good use case because what you could do is tell Rockbot, go and make up the kinds of questions that our people will ask and see where we show up.
And you've got all these resources here. You might as well have it spin it up and use its search time, right? Yeah, and it also has a browser that you can pre -authorize all those tools and use your accounts, which makes it very handy for it.
And I've done that for one big important reason. I like Claude for certain things, certain type of programming better than Grokbot. I like ChatGPT or OpenAI software for others.
And so what I did was I basically installed them via CLI on all these devices, on all the Grokbot agents. And I said, if you ever need it, now you have access to it. And it's incredibly helpful.
And I didn't know how to do it. But I just told the agent, you do it. I'm doing something similar.
I have installed the Grok terminal CLI because it has some specific connectors that Grokbot doesn't have. And when I want to use that connector, let's say my banking connection, I ask it to prompt the terminal because the connector is already there. Okay, I love that idea a lot.
Okay, one of the things that people do with this is they basically take this video and they give it to their agents. Now their agents will be able to use it, and we'll talk about how to do that. But first, I've got to tell you, this interview is sponsored by Zapier.
Here's a problem that I had, and now that I've got someone from Zapier on, we're going to talk about how Zapier solves it. I, two weeks ago, ran out of my subscription usage on my GrokBot, and it said, do you want to stop? And I go, God, no.
Actually, for a moment, I literally stopped, and a few seconds later, I said, no, I need this. So I turned it back on, and I said, okay, let's just pay per use. My $140 a month plan suddenly had overages of $265 for just two days of extra usage.
I came back on Monday and did what everyone on Twitter told me to do, which is say to the agent, what happened? What burned up all these tokens? And the agent said, look at all these routines that you have me doing.
That's what burns through tokens. And I realized, okay, I'm having it do things like what Brock showed earlier, which is like every three hours, check granola or check this, check email, check calendar. And it's wrong to say every hour, go check my calendar because I need it immediately when something shows on the calendar to take action.
And it also burns through tokens. And so can you talk about how Zapier MCP solves that problem and webhook specifically?
Yeah, the beauty is that you can, take all those routines that use your usage and export them and build them again in Zapier. So they can run independently of your usage, but at the same time, when you want them inside your GrogPod, let's say, you can always send data either through the webhook routine that you can trigger anytime you want, or you can actually connect the Zapier.
connector within GrokBot, and then it gets access to all the 9 ,000 apps that we already have. I didn't have to use a connector. I just used the MCP like that.
Okay. I like that even better. Exactly.
And of course, there are many popular connectors already on Grogbot. But what about all the other smaller apps that are not so popular and Grogbot doesn't have a connector for them? And that unlocks everything.
And also a great use case that I catch myself many times in that problem is that your, let's say, your work email is not... uh grokpot is not authorized to be used with your work email but zapier might already be provisioned so you can connect zapier to grokpot and it gets like a proxy access to your work email your google seats and all the apps that are already approved okay all right we got a link below go to it let's go to the next use case this one is i also have a better way for her to do this but i really like how clairvaux is using it to to help her cut subscriptions look at this Penny Pinscher does exactly what you...
Penny Pinscher is her agent that looks to save money. She does, which is she goes into my email and finds pennies to pinch. She finds things to cancel.
And so I just say, go through my receipts, find subscriptions I should cancel, things I'm paying for that I shouldn't do. And what was really interesting about using... penny pincher the grok bought is that it did find things that i should cancel like a apple game that i was paying for since my kids were really little and now they're big they don't need it anymore but it's also found things that i can negotiate that will renew in the future so a good example it found was our home insurance is going to renew in a couple months I want you to, I'm going to negotiate, find other rates and go back to our agent with market rates against the policy that we have so we can have a negotiated rate.
I like that. It also says that a trial for Apple Creator Studio is ending in November. It's going to remind me ahead of that renewal and I can decide whether or not I want to renew it or not.
And then it noticed I wasn't using Audible and it said, you know, you've been paying for Audible, but you haven't been actually redeeming your Audible. credits for a couple of months. And so again, I paused that for 90 days because I have aspirations to listen to Audible.
I'm not going to listen to Audible, but it's going to remind me in November when that pause is ending and help me decide whether or not I want to do it. Okay. I had a suggestion for her to improve, but now I think there's a merging between our two approaches.
The thing that just came out recently is the clad connection to your financial institutions. And I really like a direct connection to financial institutions instead of the way that a lot of these creators are doing it, which is going into their email to look for receipts.
Because you're going to have more billing come through those financial institutions than receipts that match them. And I would prefer to have a direct access into there. And that's what I thought she should do instead.
I do have that. And you could limit it. You can say, look, you get to view what's there, but you don't get to do anything with it.
But now that I'm hearing, again, her talk about Apple, I... realize that there are certain receipts that come in more granularly via email, like the Apple receipt, which tells you not just that Apple charged you, but what it charged you for, that I think it would be helpful to do this. So I do have an agent called Finance, and what it does is it connects to my finances through my banks.
Now I'm going to have it actually do the same thing with email. So there's my feedback on it. If you don't have anything, I think we should just move on to the next.
Yeah, I do. I do the same. And you can have a hybrid version.
You can have the banking and then you can also check your emails. I love this. I do it all day, every day.
Highly, highly, highly recommend. Next, delegate to a subagent. and the last one is my favorite you handed a small project ask it to spin up a sub agent and manage it and only report when it's done so this is the prompt i gave it i want to build a landing page for x spin up a builder bot handed the job and manage it until it is done show me the result on a live url only check in with me if you need a decision and it's doing that it's literally coding right now spinning up an agent and the builder is ready Let's see what it gives us.
It's crazy because it handed cursor, it opened the cursor tab and it handed cursor such a good prompt along with outcomes, the goal and success criteria, when to know that it's done. That is a great goal prompt, but let's see how it does. Let's see how fast it can do this.
Can I solve this Rubik's cube before it is done? with the landing page. All right.
That was one minute, seven seconds. And as I can see, it is still going on. We do have a preview already.
Oh, oh, there you go. All right. It's actually done.
Wow. Happening now X is the public conversation in real time. So it's made a landing page about X.
Okay. So I use this after watching him. I've got feedback on it.
And then I've got a question for you because you're more technical than I am. Um, I've discovered that what I could do is if I have a long project, I tell my agent, hand it to a sub agent. That way I have more direct conversation with the, with the agent.
Um, but what I've discovered also is that the agent will do it automatically. It will know, Hey, Andrew just asked me to do this deep dive research. I'm going to hand it to another agent so I could be available to him.
Um, so it's useful to ask, but you don't need to ask the question I have is he had the project go to cursor. Tell me about that. Do you know anything about that?
I also do that sometimes, but mostly when I actually want to create a website and an artifact like that. Otherwise, I try to keep it within channels inside Grogbot. And I would comment that this is on the negative side of Grogbot that anytime they...
chat between them because sometimes you really lose control of how many sub -agents are running at the same time. I see. So you're saying, look, this could actually get a little bit out of control.
I'm going to keep an eye on it. And it's true. I don't like when I see that they have a lot of conversations and anyway.
It's still helpful, and now I'm not so worried about giving it a beefy task when I know I'm going to need to keep asking it questions. I don't want to interrupt what it does. Yeah, and another quick comment here is that if you install Cursor or any other, let's say, Grok build within the virtual machine, you also delegate some of the token usage in that subscription instead of your main Grok bot.
Which I like. I like to do that, right? I want to pass it on to another tool.
Okay. Here's something that people don't use enough webhooks. Here's a great use case for it.
I used to be intimidated by webhooks. They're just now something that I love here. Webhooks.
Now I know webhooks sounds like a technical word, but it's really not. Just think of it like because in the routine section, you can see that they can be triggered based off a time, like a schedule every hour, 5 PM, or it can be off of an event. So these are event triggered routines.
like someone ringing your doorbell that could be on a slack message on a team message or a webhook and these are all basically the same thing this one is just like you can configure your own webhook call so let me show you an example i've got this bot down here called webhook and i created it just for this demonstration if i go into this app that i created what happens is i fill in this information so i will just put nate i will put nate herk 88 gmail .com and i will put hello I need a new dog.
And when I send this request, what this does is it actually sends a webhook. So it basically just shot a payload to this webhook address. And you can see now this bot has woken up.
It's going to take that payload. So basically the information I submitted in that form, and it's going to take action for me. So you can have agents automatically wake up when you get a new email or a new Slack message or a form submission or so many other possibilities, a new CRM record even.
so what happened is it said hey a form just came in asking for a new dog i emailed them that we got it and we're processing it and i dm'd you in click up so if i go to click up you can see that i got a message from up at ai if i can make this a little bigger you can see that form just came in and in my email you can see that we just got a message which says we got your form and you wrote hello i need a dog we'll follow up soon so obviously just a quick demonstration but the point being you can have grokbots take action when something happens andrew i'd love for you to explain a little bit about what web hooks are because Almost all the creators will use routines that are based on time, like we saw in that first Brock example.
And I think those just burn through tokens and it means you're waiting. So if something happened like two hours before your next search, you're waiting two hours for the action to happen. And often you want the instantaneous action.
That's where webhooks come in. But how do I explain what a webhook is? And anyway, what else do you have to add to that?
I love the example that he gave with a bell. Technically, it's a URL. It's an address that you say to the other.
for agent API, when you want, when you're ready, just hit this URL. And technically that URL responds to the Grog bot in here and it passes all the parameters that you sent in that URL. In this case, it's the form details, but it can be more complex and more rich.
And it's great because it doesn't have to pull for anything. It just waits and then it wakes up when it gets the message. I set up something similar.
way more advanced and more geeky, but I wanted to make a family chat within Telegram or iMessage with my wife, and there was no connector for that. So anytime a new message was sent, technically Telegram was pinging this Grog bot with a webhook, and the bot was responding back to us. I didn't know you could do that.
I could create a webhook for an iMessage that comes in? Yeah, but you need a third -party service to proxy that webhook. For Telegram, it was much easier.
You can technically ask your Gropo to set it up. Honestly, I would love a webhook for it. But that shows the limitation is there is no limitation.
You can use it however you want. It's up to your creativity. The tool that I really need one of these webhooks for is WhatsApp.
I hate going into WhatsApp, but a lot of the world uses it, and so I have an agent that just monitors my WhatsApp, responds to a lot of these messages, and now I've been learning to love WhatsApp for it. My one issue is it has to basically go check it on a scheduled time. Again, burns through tokens, makes mistakes inevitably because it anticipates stuff.
And it's not immediate enough. And if somebody responds to me on something urgent, like I'm at your door with a package, I need that immediacy. So maybe now that I said this publicly, somebody will have an idea for me.
All right, next. That's why I also retired it. It's very slow, especially with messaging.
But Grokbot can actually check your messages on WhatsApp versus Muse, which cannot check your messages on WhatsApp. And I do like that. You would have thought that Muse would do it.
I have a feeling it will in the near future. Okay, let's talk about customer issue. Customer wants a refund.
Here's how she set it up. And let's look at what it does. Holly Helpdesk, our helpdesk customer support.
I kind of like the way that she's naming it. Holly Helpdesk, Penny Pincher. I tend to give them real people names because they start to interact with the world.
But the idea that there's like this two -name thing is really good because then if I email somebody and CC one of my bots, I could use the first name. But internally, I know what they are. Anyway, I kind of like seeing these little details of how people do it.
All right, I'll take it away. I also like to give some funny names. Like my banking bot called Ledzer.
My tech bot is called Unix. Yeah. Stuff like that.
I do it too. Over to Grokbot. This, again, used to be an open club, which was pretty good.
I spent a lot of time giving it tools. access and controls and all that sort of stuff. But now it runs through Grokbot.
And why I say I've been so impressed is in the last week since running Holly help desk, I have gotten multiple people asking to leave reviews about our support being so good. And I've not heard that before. I like that.
Here's how she does it. So what she does is every hour she sweeps the intercom and email inbox. She uses our support playbook and tools and she manages refunds, technical support fixes.
She's empowered if there is a bug to fix it using Cloud, Cursor Cloud agents. she is empowered to request refunds from me and i really love this request tool use approval flow you can see here every time a refund needs to be processed it gives me a button that says, should I refund or not?
I'm going to go ahead and refund Ruben's $19. It should push to me a Stripe request. That Stripe request, I can open in Stripe.
It's called an agent action. It allows me to explicitly approve that action for the agent to do on my behalf in Stripe. And then the refund is issued and Holly emails back.
The other thing Holly does is on a weekly basis, looks back seven days and sees if there are any items we need to add either to our support playbook. or to our docs to make sure that most of our support issues can be self -service. I absolutely love that she's doing that too, that first of all, she's giving refunds in an easy way.
The second thing is I really like how at the end of the week, it goes in and says, here's how we can improve our docs so that we can answer these questions faster. Let me see if I can actually, let me see if there, I brought up the play button. I want you to see something that is again, something that's been coming up.
You see how she also has a check one minute ago, then a 1032, then today at. 939 and at 833. Again, if somebody's waiting for a refund, this is a computer.
They shouldn't have to wait another half hour for the computer to respond. If she was using webhooks, she would be able to give the customer immediate answer. Imagine you send a request for a refund and immediately comes back and it says, I see your request.
I understand it. Here's the solution. And I'm also checking in to see if we can give you a refund.
That's where webhooks would be useful. Not to make this into a big webhooks thing, but that's what I'm taking away from it. And I will become a little more grumpy in this case, just because I can imagine that this whole thing can become really messy if you have too many messages.
And that's, I would say, a problem with Crockpot, that everything lives within a chat, one thread. And imagine if you have to approve one or three or five or ten refunds in the same thread. You can lose track of them.
So I would love to have a way to have separate chats or threads that... may also be disposable. You know, I've actually been struggling with this myself.
I have a lot of things that I need. One of these like multiple choice buttons, like she has at the top of her screen. And what I do is I have a chat message and then The multiple choice.
Then another chat message and multiple choice. And it does get a little disconnected. I wish that there was a better way.
For now, I don't have anything that's out of control, so I live with it. But I see your point. Let's go on to the next, which is selling on Marketplace.
Everyone talks about using news to sell on Facebook Marketplace. These are two people, Peng and Lauren, who are lead designers and engineer for Grokbot at SpaceX. Check out how they sell on Marketplace.
I have opinions on this, too. Let's say I want to sell my DJI mic on Facebook Marketplace. I just send it a photo and tell it do a couple of things.
So like sell something, list something on Facebook Marketplace sounds easy, but actually there are a lot of steps. Like you need to check the latest price for the model on the official website. I want it to find out what's the lowest price offered in my area, like who I'm competing with.
I want it to learn like how I write listing descriptions in the past. and also post this new listing and then if nobody's interested in it automatically lower the price by five dollar every week so you see like it's like a series of tasks that i dedicated in one single prompt and it will go execute the task and it's very interesting that i know like when it should write the description it consult my writer bot so it messaged the writer bot it asked like how i wrote marketplace listing description in the past and then we're able to do the research it'll be find out like how other people are selling it and then eventually put up my listing so in the past we may do like all those five steps manually by ourselves But now, not only can you delegate individual tasks to it, but also ask it to train the tasks for you.
Okay, so I've done a bunch of selling on Facebook Marketplace, like my driving lawnmower. And the thing that I like about this that you still can't get from Muse, which is Facebook's own tool, is it can respond to people proactively without checking with you. Muse will keep coming back and saying, hey, Andrew, can I send this message?
It's too late. I don't want to deal with these people. I just know here's the price I'm willing to sell.
Here's where it is. Have an agent go back and forth with them. is what this can do really well.
It does it in the browser. It's dependable. It's effective.
You could even say something like, look, if by two days it doesn't sell, I want you to lower the price and go back to everyone who's made a suggestion before and tell them about the lower price or whatever it is. Really helpful in Facebook Marketplace. I agree.
Okay. Now, since he has these two people on, and by the way, why won't any of the SpaceX people talk to me publicly? I would love it.
I don't know how Peter Yang gets these two people and so many people. Yeah. And also a quick tip here that we can steal is how they prompt Grogbot here.
One, two, three, four, five with steps what follow. Oh, interesting. Right.
Instead of here's what I'm trying to do. This is the end result. I want you to take these steps.
Okay. I also like how they were using it for email and calendar. My bot is replacing my email.
So like I would do things like, hey, what are the action items in my inbox today? My email is still out of manage. and like for things i identified i can do it for me for example i need to pay my friend money for dinner i would just ask it to do so and if it doesn't have access um it will prompt this sign in page for me and i can take over and sign in on its computer and it would do it for me uh it's replacing my calendar so if i'm having a coffee chat with somebody not like basically any edits any event creation find booking like book conference rooms anything i just put a screenshot and then let it do it.
Let it like add the events to the calendar, modify it, stuff like that. Okay. I'm going to disagree with one thing.
It is not replacing his calendar. He still need that visual of the calendar. Let's stop pretending that everything needs to happen via chat, but.
For asking questions, for getting information, it is fan -freaking -tastic. There was one video that I ended up, there are tons of videos that I ended up removing from this, but one that I really liked and was painful was there was a creator who said, look, I'm going to respond to my email now using voice. And it's not here in this video, but since this video was done.
voice conversations are now what we're calling phone calls, I guess, with your agents have been added into Grokbot. You can now have a conversation with your agent, pick its voice, and you can say, tell me what my messages are. Tell me what's going on and respond and tell it what to do.
And it's really, really helpful to do that. Wow. Yeah, that's really useful.
And also about the email, I have a board that is called Mailman. And every day, because I love Inbox Zero, it reads through my emails and I ask it to Mark has read emails that look newsletters or maybe spam, stuff like that, and then come back with unsubscribe opportunities.
And then he tells me, I have this. Do you want to unsubscribe? I say thumbs up to that or the other.
And sometimes it even spin ups the browser and tries to click the unsubscribe link for us. So helpful. So helpful.
All right. Let's go on to the next one, which is actually speaking of email. Here's a prompt.
I was thinking maybe we leave it out. Maybe we keep it in. I have a problem with this, but let's talk about this.
Here's his prompt for getting through his email. The thing you can set up is an email. triage or an inbox triage.
I'm going to send it to Ledger, which happens to be our email bot. And I'm going to send him. He uses the same name as yours, Ledger.
By the way, if you all like this approach to understanding new tech, especially AI tech, you should subscribe because we do this whenever a big thing comes out. I want to study what everyone else is doing, show you the best, and then bring someone like Andrew, who has a deep experience to come and give us opinions on it along with me.
So you know how to use it better than even the creators themselves use it the best. with the best feedback. Let's look at it.
This, read my inbox first, batch every email that is safe to archive without me reading it and show me one list with an archive button. Then summarize the ones I need to know but do not need to reply to. Then for anything that needs a reply, reply the full thread and draft a reply in my voice, ready to send.
Learn my preferences over time, never send without my approval. I just realized I didn't specify I wanted it from the last 24 hours. So I said only the last 24 hours, please, because it's been like five minutes.
And I'm like, what the hell? And this is very interesting because just by telling it this, not by telling it that I want anything else fixed, it figured that it should fix its other scheduled jobs to run the emails in the last 24 hours, unless I say otherwise. That is fascinating.
And just like that, it came out with everything that I just asked for. The fourth thing is a research. He didn't show it, and then I said, okay, I'm going to use it, and I'm going to show it, and I realized why he didn't show it.
It really does get you such personal stuff. But I do like it, and I think what I did was I screenshotted the prompt, and then I gave it to my agent, and I said, do this for me, and I forgot that he had done it for 24 hours. So I got it for everything.
It is very helpful. It goes, here's 63 emails that are safe to archive. So I got promos, like, from Spotify, from Paper, which I used for one of my past things, for something called IHG Points.
and then it gives me like a radio button or a button that says do you want me to auto archive those then it says automated notices like acceptance notices from Micah my guitar teacher who accepted meeting with me from Chili Piper for whatever reason I could I could archive those it gave me cold pitches that it really identified well 17 of those came in and then your team already replied to 17 messages so you probably want to archive those very very very surprisingly helpful and then there was a bunch where there was no reply needed so you can just tell me over here like Shonack Patel sent you $15 ,000 to Andrew Warner Texas LLC on September 30th Lauren pulled money out of the company T -Mobile says the payment is late wait I thought I cancelled T -Mobile I know I did I gave them that back their device anyway really helpful for me to see all these things I like that prompt I was going to pass up on this but when I see a good prompt even if I can't show the result I think it's worthwhile any thoughts before we go to the next tip
No, big fan of it as well. Very, very well done. Okay, I would probably adjust and maneuver it, but I'm going to start to integrate this into my system.
Okay, I like this a lot. Here is my buddy, Eric Hsu. Here's what he does.
Okay, so yesterday I saw that someone wrote a post on Amplitude. And this is specifically on X, formerly known as Twitter. He uses it a lot.
And there's a lot of insightful stuff on X and on YouTube and on Reddit and other platforms. Whatever he's doing for this, you can do for all those platforms. It has a lot of problem pages that they create ahead of a search term getting more traffic, right?
So they've done a really good job with that. I'm like, okay, well, hey, how can we use this? Are we going to make a net new bot?
Like, what should we do exactly? This is all I do, right? If I see something nice on X, I should see someone share a cool post or I should see them share a cool bot.
I just say, hey, is this an update for us or is this a net new bot? And I just shared that post. Okay, pulling that post now.
Okay, this is upgrade, not a new bot. Okay, so you see how short I am here? I'm just like, hey, just do this, right?
Just do this and tell me what you're doing next, right? Okay, let's test the upgrade. And then it's like, okay, to convert, is this going to...
to create more problem pages for us or give me a list of 10. And it's like, okay, here's the top 10 over here. And then I'm like, hey, how can you verify that these are trending upwards?
So this is how I go about building the bot. And this is how I pull from different resources. Andrew, I get a lot of private messages from people and sometimes even comments on my YouTube videos from people who are a little sheepish when they tell me that they hand a link to this YouTube video or to my YouTube videos and they tell their agent, how can we use this?
And I think they should not be sheepish about it. This is exciting and it's effective. And I want to make sure to highlight that.
And I personally have another bot that is very similar and I call it Stashy from Stash. I want to stash stuff. So every time I find a useful post like that, I just share it with that bot.
It saves it, it stashes it. And then any time I want to recall it or if it's related, let's say, to this one, I say... can we i say can we use that for this job now is it relevant or what was that thing that we talked about and it's very useful and it can be used with many ways not just this way that he just showed us all right final one boy and there's so many that i would have wanted to show you but this one is recurring appointments that you get to book on sites that are pains in the neck This next spot I call Booker.
He's in charge of making appointments for me that need to happen on a regular schedule. For instance, I need to get my haircut about every three weeks. Otherwise it gets way too crazy.
So Booker makes the appointments for me and adds them to my calendar before I even think about it. And all this happens because he's able to check my calendar and he has access to a browser. So if we click on the top right here on this computer icon, this will show me Booker's screen.
And I can open this up. You see that he's looking at the Great Clips website. Let me just point out a few things over here.
Notice, first of all, all the privacy notifications on the bottom that make it hard to use. Notice how it's got all these different, it's not iframes, but it's iframe -like experiences. My barber in Austin uses...
MyTime as the booking system, but it's an iframe on their website. So when I go to their website, if I'm looking through the password manager for my login information for the password, which is a pain in the neck anyway, it doesn't find it because the password's associated with the MyTime, which is an iframe within the site.
All these people, they got hoodwinked in Web 2 .0 with terrible sites, and we now have to navigate them. And I put beyond barbers. I would also put into it, we have a place here which is a gated community that I was living in for a bit.
It had some... some service for communicating with me. It was the worst app, kept getting logged out.
I said, never, no more of this. I created my own agent that was interacting with the site and I didn't have to touch it at all and it did everything for me. This is the thing that I just am so frustrated by and I want to highlight that because I thought, am I really highlighting how Paul Lipsky gets his haircuts?
That doesn't seem big enough. No, no. The bigger message is there are systems that need to be booked on websites that suck.
This is where I get my haircut. And Gray Clips allows you to book your appointments online. So Booker will work as long as the place you want to book an appointment.
has online bookings, which Great Clips has. My dentist also has that. So Booker makes my dentist appointments for me every six months as well.
Now, he doesn't just pick a random time for me. Instead, what he does is when he sees that I'm due for a haircut, he will check my calendar and look for a time that works. And then he checks in with me.
He says, hey, you're due for a haircut. I see that there's an opening on your schedule, in this case, 4 .30. Do you want to lock in that date and time?
And I'll say yes. And then around that time, he will go to the Great Clips website and put me on the wait list so that by exactly 4 .30, I am the first one in line and I'm ready to get my haircut. And then the first time that this actually worked, it honestly kind of blew my mind.
So he booked the appointment, he put it on my calendar, and then I showed up and sure enough, I was on the wait list. I was the first one in line and I just went in. Got my haircut and didn't have to think about it at all.
It was really cool. I'm also going to set up Booker to do the same thing for oil changes and tire rotations for me. Super helpful.
Thoughts on this? And an amazing feature that Grogbot has, and I wish he would show us, is teach a task. If you go back when he was at the browser, there is a button that says teach a task.
So if the website is complicated, you can do the task manually. It records a video of you. And then next time it creates a skill that can reproduce it every time correctly.
And this is what I love with this. What are you using for passwords? Because I use one password and I share it using agent mode within passwords.
And lately Grokbot is... kind of spitting that back out at me because it has its own direct one password integration, which I don't love. I don't want to give it full one password access.
I don't know. How have you been feeling about it? I haven't tried it to be honest.
How do you give it passwords? Yes. I just log in.
It keeps the session. But it's not very reliable. Sometimes you have to re -login occasionally.
I don't like that at all. I would highly recommend that you look into agent connections for one password. And then whenever you have a password that you want your agent to use, you move it into that collection and your agent will have access.
If it gets logged out, it can log itself back in. It's super useful. Honestly, it is fantastic.
And what I did in the past was I... bought another user account for uh one password for my agent and you don't need that one password will have it i don't love one password it's slow it's got clunky it's clunky it's got issues having said that it does have good agent mode and so i like it for that all right if you all like this video i've got another one for you that will show you how it's not just you who's going to work with your agent it's your team and even if that team is just you and a friend or you and a spouse there's a way to bring all these agents now into slack to allow them to communicate with each other and more importantly communicate with other humans on your team i've got a video for you on that here and we'll see you there.
The Hook
The bait, then the rug-pull.
Two hosts spend forty minutes testing other creators' publicly posted GrokBot setups against each other, agreeing on what to copy and arguing about what each one got wrong.
Named ideas worth stealing.
09:06concept
Zapier MCP (Model Context Protocol)
- 9,000+ apps / 40,000+ actions exposed as tools
- Works with Claude, ChatGPT, Cursor, Windsurf and more
- Two steps: grab a key at mcp.zapier.com, point your client at the server URL
- Secure by default: auth, encryption, rate limiting handled by Zapier
A sponsor slide shown mid-episode explaining how Zapier's MCP server plugs any MCP-compatible AI client into thousands of apps as callable tools.
Steal forany agent build that needs access to a niche app GrokBot doesn't natively connect to, without hand-rolling the integration
17:42concept
Routines vs. Webhooks
- Routine: a scheduled task that polls on a timer, e.g. every 3 hours
- Webhook: an event-triggered wake-up the instant something actually happens
The recurring distinction the hosts return to across nearly every segment: timer-based checking burns tokens and adds delay, while a webhook wakes the agent only when there's real work to do.
Steal fordeciding whether any given automation should poll on a schedule or be wired to fire on the actual triggering event
How they asked for the click.
VERBAL ASK
30:01subscribe
“If you all like this approach to understanding new tech, especially AI tech, you should subscribe because we do this whenever a big thing comes out.”
Delivered mid-video inside the Inbox Triage segment as a direct aside right after demonstrating concrete value, not held for the very end.
MENTIONED ON CAMERA
Visual structure at a glance.
