Transcription
Uh, we are dedicating this entire month to learning how to build AI agents. Very specifically, what you need to know to build AI agents as a product manager. Uh, there's no other like way or shape or form about it. You have to learn these skills, and you have to learn how to build. Uh, it's as if, like, all of us here have to really learn how to speak French tomorrow. We just, like, there's no working around it. Okay.
Uh, tonight's, tonight's uh uh webinar is actually attracted a lot of us to come here today. It's going to be replacing our marketing team with cloud co-working teams, right? Now, they are getting replaced right now, effectively speaking. I'm going to let you guys know this. This is not me trying to be an alarmist. Everybody is going to be an agentic product manager. Customer success managers are going to be agentic product managers. Uh, software engineers are going to be agentic product managers. 80% of what it is that we do, investment banking analysts, management consultants, lawyers, they're all going to be agentic product managers. Close to 80%, minimum 65 to 80% of what it is that we do every single day is repetitive manual work. And we're spending so much time trying to memorize like, oh, what did we do again? But we've all done it before, right?
So, it's going to start off with Claude Co-work. I'm also going to be adding the RSVP links for all of this. Uh, April 8th, which is actually tomorrow night, we're going to be going through a live build of the job application AI agent. This is actually going to be led by former uh, director of product over there at Mastercard. Uh, using AI to go ahead and ace the product manager interviews. Uh, Hueie is actually here tonight. Compliance product manager, putting together a compliance product manager dashboard. So if you guys have thought about applying to like gaming companies, any like uh companies like Harvey AI, compliance dashboard, absolutely, absolutely huge. Um, because legal tech right now is blown off the shelf. Uh, and then market research and prediction AI agent for product managers. Uh, we're going way beyond right now from just the normal like, we're going to go ahead and chat back and forth with it, to actually building AI agents.
So, if you guys actually don't have your Claude Co-work open already, now actually might be a mighty fine time for you guys to do it and actually build with us. Um, no more freaking decks. Like, you know, remember like way back in the days, you'd watch Martha Stewart, get your chopping board out, wash your hands. We're gonna be making this pot roast here with James live. Okay. And then uh uh my pal here uh and also co-founder of uh Tech Careers, James Juan um, with once product manager at Capital One. He will be walking you through the live build. Let's get as much interaction as possible. Please everybody pull up your uh please go out there and pull up your Co-work and then uh I'm expecting to see a lot of questions coming back. Uh, and then also letting us know in the updates that you guys are also at this stage right now because you're doing market research and like everything you can do has marketing team, right? So, James, I'm going to let you take it from here.
>> Yeah, sure. And I see a question here already. Um, I'm just going to answer it really quickly. I think Rich dropped a newsletter opt-in and so we we typically share our webinar recordings uh only if you're subscribed there and so um if you missed that link, we are dropping it in here once again. Go ahead and opt in and we will be blasting this recording out um to those folks that went ahead and um signed up. And to Rich's point, this is a live build and so just to sort of preface and set the stage a bit, what we're trying to accomplish here, or what I was um trying to build here, was an end-to-end workflow uh where we identify the top themes, trends, and topics, right? Um, over the last seven days so that Claude and Co-work can go ahead and write content, marketing content that is for us that we use across multiple platforms um and build an end-to-end automated workflow that does that for us on a whatever cadence we want, right? Now, right now, we're looking at weekly, right? Based on some of the recommendations that we're getting. But um, without further ado, I will go ahead and pull up my screen here. Um, and I'm also going to start by just letting you guys know how you guys can get safely set up because I know that was a a big concern for Richard and myself where we were hearing horror stories about, you know, co-working, Claude, code going off and doing just like nightmarish things on your machine or or your or your or your instance that is and so and session like deleting all of your emails and such. Um, and I believed like the head of AI safety at Meta lost her job because she went and gave Claude a bit too much um, in terms of permissions. Now, Claude and Anthropic has has gone ahead and, you know, put some more effective guardrails around all this stuff. And so, first things first, let me pull up my screen here.
What you guys will need to do is download Claude um, Claude for desktop. Um, I will also note that in order to use the Co-work and effectively integrate with computer, cloud computer, that is, you will need the Pro version of this. And so as you can see, you have chat, which is, you know, free out of the box to some limit and a set number of tokens on a daily basis. Um, it is the same case for the Pro, but you obviously have access to more features of their stack like both Co-work and Code. Uh, so first things first, you want to go ahead and get the desktop version. Secondly, as you can see on my browser here, I have this wonderful sphincter of a logo and Anthropic um, Claude's Chrome extension added. And so that's also what you will need if you want to do more exotic tasks that require browser integration or um interfacing. I'll pause it for a quick second. Um, and so what you see on the page here is you have you see you're seeing X, you're seeing Ampify, you're seeing our website techcareers.io. Um, and you'll notice here that um, this tab is actually grouped already. And it's because um, I've built, I would say, the first half of the workflow where we are going to numerous sources to identify what the hottest trends and themes are and topics are that we want to write about and um, craft content around for marketing purposes, as well as becoming a thought leader in the AI and agentic space um, especially as it as it refers to um, a career in product management, right? And so this is our website for those of you that weren't aware. Um, feel free to jump in there, read our articles, right? Um, you will discover today that going forward, a lot of our content uh will be in part be contributed to by not just Richard and myself, but by Claude's stack as well, right? Um, we are going to be doing some level of filtering from the perspective of topics and and sort of the the tone and voice that we're we're trying to portray. But um, what we are looking to do is set this on an automated workflow and build different agents that are able to help us do away with repeatable tasks. F, as I mentioned, first things first, you need to set up Claude uh desktop. Go ahead and add the Chrome extension. I believe they just updated both and so you'll need the latest there. Um, secondly, when you open up Claude desktop, you want to restrict the access that desktop has to your machine. And so you go ahead and actually go to chat here and you'll be able to add to Pro. You could you could add different projects which uh builds context, or you could give it access to a very specific folder. Where the hell did I go? I'm already in that folder, I think, which is what it is. This looks like, in all honesty, a bug. Um, if anybody has this up, you will notice that, hey, I have Claude in Chrome turned on. You will also need to log in to the extension uh on the Chrome browser. You should also see um connectors where you could connect like Google Drive and other sort of ecosystem that you want to integrate with and interface with. And before I believe, where in the world did that go? You should see an add folder, which it may be that I'm already in there. Um, but you should set up a folder in not your desktop, not your root of your machine, but you're going to go to whatever your machine is named here, right? And you are going to create a brand new folder and path specific to the directory that you want Claude to have access to so that it isn't able to go and mess around with any other documents, important files that you have saved locally on your machine. I've gone ahead and named our sort of project directory Tech Careers. Um, I have, as you can see, numerous folders here. Um, and and I'll explain what each of these folders represent shortly.
Now, first things first, what I learned as I started to build these agents were was oftentimes um, if you don't if you aren't explicit about the simplicity of a task, code is the foundation for what co-work is, right? Co-work is just a friendlier interface for Claude Code. What is Claude Code? Right? Claude Code is a tool built originally for engineers to automate engineering tasks, like PR reviews, like bug fixes, right? Um, like even code updates, right? Um, based on skills that you build within that ecosystem. Highly um, highly uh technical, built specifically for engineers. Now, what they've done is extend that capability and provided a much friendlier interface through Co-work, right? Where you are effectively interacting with it very similarly as you would any of the other LLMs, right? Um, and it also has a sort of diluted version of Claude skills built in. That's what I was showing you here before. Um, I have two skills that I've built um, and those are specific to the front end of the workflow that we're trying to automate. That is, hey, first things first, what's going on in our ecosystem and domain, right? What is trending in the intersection of product and AI and all things agentic, right? And so what I did was I started working in chat. I I told it like, hey, what these are the things that I'm trying to accomplish, right? And my understanding is that Claude is driven um, their agent workflows, that is, off this concept of skills, right? And so I went down this rabbit hole of learning what skills meant in the code ecosystem versus what it could mean for a co-work, right? Um, what I secondly did was then I started to break up the process, right? We have trend discovery. We have selection of topic that we are going to use for the week. We also have the writing of the actual content. Fourth, we also provided it additional context, right? We said, "Hey, go take a look at our website, read through all of our articles, get a feel for tone, voice, style, right?" So that we are um, putting sort of a um, a unified face forward, right? Um, and then I asked it to go ahead and write these three articles. I'm just going to scroll through this just to give you guys a sense of u what, you know, I was working on earlier today. It's basically um, the individual steps. And so as I mentioned, the trend analysis, which I broke up into two parts, it's using sources like Reddit um, I think I think it was Medium and some other website um, to go and fetch themes. I also realized based on my conversation with Rich earlier that Grok is a great uh resource for identifying data that can be um, that that lives currently on X, right? And X is a real-time, for lack of a better term, social news platform. And I felt it'd be very relevant um, for topics that are trending real-time. And so the two skills that I ultimately ended up building were this trend discovery skill and the Grok trend analysis skill. Um, but what I started with doing was asking Co-work, hey, this is what I want you to accomplish. Go ahead and write me the steps um, and effectively ultimately and and ultimately the prompt that I will use to create a skill file so that I don't need to repeat myself every time I want to do this. Um, and what we're going to be building from this point on is the back end of this process, right? Um, it's going to be adding a skill file for the writing of the blogs. It is going to be creating the the assets that we need, like a CSV file in our case, to upload this content to our website, right? And then it's going to ultimately be creating a sort of an umbrella skill that ties all of these different steps together to make it easy for me to prompt with a single skill or call to a single skill to perform all of these tasks in one, in a singular workflow, so that I don't have to prompt it multiple times.
I'm going to pause really quick there. Um, hopefully there aren't too many questions, but um, I'll keep it going if I don't see anything here. Let me see. Awesome. And so, as I mentioned, we had gotten to the point of writing these three articles, right? Um, and as Rich had mentioned, um, we we'll probably revisit the themes, but like the PM role is literally evolving as we speak, right? This one here, I think, you know, actually um, triggered some anxiety for myself, right? Like you really need to be building, right? And you really need to start separating yourself from the pack because the gap is starting to grow, right? Because because AI technology and its capabilities are attracting product managers by the by the groves. Like you really need, if you are an aspiring product manager or a current product manager, you absolutely need to be well-versed in all of the capabilities in the AI sphere, right? Because that is what you are going to be leveraging um, to make yourself relevant um, and make yourself, allow yourself to deliver enough value to join the team, right? So, as I mentioned, I had written all these uh I had I had Claude Co-work build all of these or write all of these articles. Now I am going to ask it to go and write a skill file that I could leverage so that I don't have to prompt this prompt Claude Co-work to do this all over again. Um, Claude actually has a pretty neat shortcut to voice prompt it. Um, I've noticed that if you hit caps lock, you'll notice that it brings up my microphone. Um, a limitation there, I I realized was every time that you hit this hotkey, it starts a brand new chat, which was really annoying. And so I decided to go ahead and just use the Co-work specific little feature over here. Um, where it's actually able to transcribe what you're saying without this like limitation of creating a brand new thread every time. And so I'm going to voice prompt this to go ahead and create a skill. Hey Co-work, I'm really loving the output and and the articles you wrote. Um, what I need you to do now is write me a skill.mmd file that puts um that outlines the steps to write these articles really clearly, making sure that we're including the links to the the sources. Um, yeah, go for it. And so while that's thinking, any questions so far? And I'm seeing some stuff come in here. Now it is going to create a content creation skill file that I could call um any time to go ahead and write articles based on the topics that we fetched. Right? While that's doing that, um, what I'm going to do next is ask Claude Co-work to then take all of the the work that we've done today. um, put together a skill file that I could call singularly to go and stitch all of these different skills or these different steps of the process um into a single skill call.
Not on the Pro plan. Yeah. Sorry. This is a this is a Pro feature. It's 20 bucks a month. Like, if you're not going to invest 20 bucks in yourself, I don't know what to say. There you go. Went ahead and actually created a file. James, I actually think it went and created it. Like I didn't even need to tell it to go and put it in the right place. It went and did it for me because I've been working with this thing all day. Actually, frankly, over the last few days, right? And now it knows what I'm trying to do without it getting additional context, which has been really incredible. Um, how is this how is doing on this under Co-work different from doing this under chat? Um, Co-work allows you to Co-work is the um interface that is able to call the browser interface, right? It is the one that's able to perform tasks and have access to your local machine effectively. Hopefully that answers your question.
And so it went ahead and created this file and now I'm going to ask it a awesome Co-work. So now I've given you all of I've asked you to create all of these skills which effectively represent each of the steps um in what I am trying to build as a singular automated workflow. And so if you go um take a look at the skills folder um locally on my machine and read through all of the files um in detail and then create a fourth or final skill file that I can call um singularly so that I could actually prompt the entire end-to-end workflow that I built through the three separate skill files that I created today. There you go. Calling it an orchestrator. That's what I should have called it. Oh man, this is so sick. Now I just I'm as this is writing, like I'm actually being reminded that I forgot the CSV portion, which is a pretty big pain in my ass that I have to create every time we want to upload these files um to our website. And so after this is done, I'm going to prompt it to to to make those updates as well. Rich, I remember you mentioned before the call, like you weren't a huge fan of the titles of the actual articles, right?
>> Okay, let's let's let's let's prompt it to do that first. Actually. >> Huge fan of the titles of the articles that we've written. >> No, the ones that Claude wrote earlier. >> Um, >> Let's see here. Beautiful. >> It wrote a lot. Nice. Already did did that for me. Hey, Co-work, I actually forgot about um that last step where we're trying to take all of the content created or written and then upload them into the CSV template file that I had given you previously. Um, I might have given it to you in the web app and so let me know if you need access to that as well. I'm going to stop sharing just in case I have something weird in my folders while I look for this. Just making sure here.
>> Oh yeah, that folder. Glad I did. Wait a sec here. Let's see. Career upload. With Neil. I've actually given this access to a part of our um, our Google Drive domain as it is. And so hopefully this here. Cool.
>> If you guys don't know, uh, we've announced a ceasefire with Iran. So, apparently we weren't we weren't going to go in heavy at 8:00. >> Oh god. Trolls, man.
>> All righty. So, I gave it the CSV file and so I want to Where is that wonderful little like Rich has been like huge on the the voice prompting. I just started doing it um a lot more recently and I I understand why I'm never going to go back to typing prompts again. So that's I don't know where the hell that went. >> Worst idea ever, man. >> Where why can't I use it now? There we go. Do I just need to make this [ __ ] No. Okay, I guess I'll type this one up here. Here's the CSV temp upload template. We're going to add each of just pay mine to the actual content column and dump and drop the content you wrote. Drop your articles you wrote into that column. That could take care of the rest. James.
>> Yeah. What's up, Armor? >> Click the file. Uh, on the side, it should show you where it went. Um, when you click the file, it'll be like, oh, it's in this folder or whatever. >> Oh, nice. Where am I looking here? >> Wherever the file was initially, I think it's that uh TC. >> Oh, yeah. No, no, it's because I'm like switching between the web app and the the desktop app, and I realized that like they're they're they're a bit independent. I think it's >> They're still in beta mode, right, with with desktops. Yeah, it's it's all on your right hand side though, the progress. It'll tell you where things are going. So, just keep keep a watch there. >> Awesome. Appreciate that, brother.
>> Yeah. Um, and so it went ahead and created and one thing I also noticed, Ammer, like this view isn't accurate, right? Often um, I thought it was broken, but then once you download it, you'll find like it's in the right place. Um, I don't know why because of the way like the comma separation happens. Um, it doesn't render well here. Just so just in case people y'all were wondering when I download this, it's going to look a lot better. Show folder. And you'll see here, right, the three articles added. It actually didn't heed my warning of or heeded my like um note of like, you don't need to fill in the other columns. It actually went ahead and did it for us and I did not remember to ask it to to format this in markdown format and it went ahead and did that for me as well. And so like once you start working with this like like these like different um AI stacks, like you'll the more you work with it, the better of a feel it gets for what your what your desired output is and it it starts to fill in the blanks, right? Which has really been super neat. And so now I have Let's see here. I have a pipeline. I have trend analysis. And I actually have um the blog writing. Um, let me see if it went ahead and view roles.
>> Katherine, name your question. >> Yeah, I was just hoping you could spend more time, excuse me, going through the initial creating that skill the first time. just in terms of like I'm guessing sort of the world is your limit, but I just wanted to see like if it gives you ideas of skills or you just start basically like just trying to understand the better use cases I guess of when you ask that initial task and I know that you're doing this in Claude desktop with the Pro version and you would do it in the Co-work tab, but can you also like talk about task or do task in chat or would that only be in Co-work? >> I actually have been jumping back and forth and so the Pro version uh with desktop. Um, even chat can do some like pretty basic tasks as long as you have the Chrome extension. And so that's actually a really great question, Katherine, because what we had previously were all these articles, right, that we add like we created Google Docs around. What I needed to do was take that all that content and format it in markdown, which is what's required for our our current um website implementation and the database. um Co-work went and tried to like write scripts, Python scripts on the fly and it it it um it went through all of my tokens for the day. Chat on the other hand gave me a much sl simpler process after I described in more detail what I was trying to do. And so I I definitely would start in chat, right? And again, you could actually start with like use case exploration, go really broad and say, "Hey, like I'm a cafe owner. I am trying to automate like I'm sick of like creating design um assets like I'm thinking of how I could leverage AI, right? Like can you can you tell me how other other folks are using AI today to generate media assets on a regular basis, right?" And then continue to narrow narrow narrow down and then you ask it like, hey, I ultimately want this to run um on some sort of automated cadence or make it really simple for me to prompt a um a repeated long workflow, right? And so that's where I that's where I ended up in Co-work, but I did start in chat. I hope that kind of answers your question there, Cat. But basically like I I started with this like really long prompt in Co-work and it went and tried to do it way too exotically and so I went back to chat to break and then asked it to provide recommendations on how can I simplify this process. That's when I ended up with this like step-by-step sort of approach um building individual skills and then I realized that's really the best way to to break down these like um longer more complex processes. Right? This being not a very complex process, but this also being like the third time I've tried to use Co-work to automate a workflow um I found I found going back and forth was really helpful. >> Thanks James. Super helpful. Um, just to confirm one thing that it sounds like, if I heard you right, you're using the chat to refine the prompt, but then you can go into Co-work after like I'm not sure if the tokens are a big deal if you upgrade to Pro or you still face that limitation quite frequently. >> You face that limitation when it starts to go off and doing like do things really exotically like writing Python scripts in Google App Scripts, right? I don't know how familiar you are with like the Google Workspace, right? But they have like you could write custom scripts right within the workspace. Um, and so that's what it tried like instead of going and reading the article and just marking within the document, right? Um, the formatting that I needed. It like read the document, cached it, tried to write run a script against it to output a markdown formatted document versus doing it a lot simpler, which is what I ended up doing, what what I was actually able to accomplish with um, chat actually. Um, but I still wanted it to be automated, right? Because I'm going to be doing this on a regular basis. And so um, I accomplished it through chat, realized how I needed to prompt Co-work or how I needed to essentially design the process and and communicate that those steps to Co-work eventually so that it could provide the output. Now, the actual skill file specifically, both chat and Co-work can produce for you. And so feel free to to kind of like jump between the two and see what makes steps depending on where in the process you are. Um, I found myself just jumping back and forth. I don't have candidly like a really clear like delineation or answer for like when to use what. Uh, but that that tends to become a bit um obvious as you go through um real time. Um, Omar, I see your hand up. I don't know if that's intentional. Yeah, I was just going to say uh with chat, you have a a Chrome extension that can allow you to see your Google browser. Um, and that's its limitation there, right? However, with Co-work, it has access to everything in your computer. And so that's why it works so well for James. Once he creates those skill files to do a repeated automated task, it just looks up into his local computer. Um, and that's I think mainly the big difference that I have seen. And then if you go into code, if you have something like Windsurf or any type of um terminal, you can connect this there. And that's kind of like the full stack for Claude right now. Um, but Co-work is literally made for like it sits on your local PC um more than chat ever does. >> Yep. Thanks. That's actually a really great way to to explain. And like as I mentioned, like I created my own sort of like domain or like workspace, if you will, for Claude to operate. And so and again, I don't know why I I did do an update. I think this might be a bug or it might be because I'm in there already. I'm not sure, Armor, but like I don't see like work in project, right? Which is what I was seeing earlier when I prior to the update. And so it could be a small bug. Could be that I'm already in the folder. Um, but you basically get to set where and what kind of access that Claude has every time it even initiates or instantiates the the extension. Every website you visit, you need to give it permission, right? You need to give it manual permission and no and no at no point am I giving it like um universal permissions or access, right? Um, I'm always setting it to like this time only so that I can control the access that desktop has to my machine and and to my browser connection or to my internet connection.
Awesome. Um, just wanted to make sure that nine coms. Awesome. Hey Co-work, able to actually update the necessary skill files so that we are ultimately producing a CSV file and also that the the end-to-end pipeline file um skill file that is is also updated to include the CSV output. No, there you are. So, there you go, Cat. Um, like I'm on Pro. Um, I am absolutely going to be upgrading to Max, which is 100 bucks a month. Um, I think Rich and I were discussing whether or not we want to like move to a team version, which allows us to then like share project spaces because I know Rich has been going ham on Claude and his Claude is a lot smarter than mine is right now um in terms of the marketing content creation and so I'm actually pretty excited to see what we could leverage in terms of what Rich has built up in terms of context um and and our business, if you will. And so that that's probably the next step for us. But unfortunately, it looks like I just hit my limit. Um, but I hope like that gives you guys a view of like A, like how to get set up. B, like I don't think there's a ton of concern around access and like security right now because you could configure it very explicitly. And I would absolutely be mindful of when you give Claude any sort of permissions um on your machine, especially your local machine, that is um, and um, it's it's also taking, you know, the the precautions of requesting access whenever you have not explicitly provided it. See, I'm going to stop sharing there and we could hold for some questions.
What is the skill? Where can and I'm not sure why folks like are posting questions anonymously, but did can somebody elaborate on what is the skill? I >> think what is a skill to start with? >> Oh, what? Yeah. Um, yeah. So skill is a it's similar to what you have seen here in code, right? It's essentially a script, right, is is I think the best way to describe it that that you could um um prepopulate within your Claude ecosystem. So that let's say you have like a like a service that you manage um and there's a repository, right, of code um you could write a skill that is like your bug fixer, right? Um, and and the more it reads through your codebase, the better context that it has, right? Um, and it will be able to run like a scan of all the bugs that are identified through like a linter. Um, and then go and address those bugs for you on a regular basis. Now, skill as it refers to within the co-work ecosystem. Um, I just realized I'm not sharing. So I'll go ahead and do that. I'll give you guys a view of what this looks like actually. Skills as it pertains to or in in in the context of co-work is literally a process document, right? You are explaining u what is important, you are explaining guardrails, you are explaining rules, you're explaining steps, right? And so the more descriptive, explicit you are um, the more refined and narrow your output is going to be. Now, in in in um, in order to move quickly, I just went ahead and started using what Claude outputted. I did make maybe like a couple of minor edits as we went, but I certainly would am going to revisit all of these files, um, add additional context where I feel needed, review the outputs in more detail, right, and refine further and further, but they are essentially process documents that Claude refers to. Yeah, let me see here. Anonymous is not going to do very well for you. I'm sorry I have to say this, but like probably the most the most trending uh thing that you're going to find in terms of like getting a job in 2026 is something called build in public. So if nobody knows you who you are like if you guys don't if you guys are not building your portfolios in public, BIP is an acronym you guys need to get familiar with because if building in public, BIP is an acronym that you are not comfortable with, you are going to get intimately familiar with with another acronym called NGMI. Otherwise, otherwise, no one is not going to make it. All right. NGMI. Okay.
What incremental value does the browser extension provide? Like, does Claude desktop access uh does Claude desktop access everything on your local machine while the extension accesses everything into your browser and they work together? >> Yes. So Cat um, it's literally opening up tabs on your browser and going to websites, right? Um, it's running Claude computer, right? Um, through the Chrome extension and so if I had not run out of credits and why I had Appy up here was I was going to ask Co-work to go and identify the best TikTok scraper, right? This is one of the popular ones. I I've had mixed sort of experiences with it, but I was going to ask it to then add another source of trending topics, right? By leveraging one of these endpoints, go and scrape the most relevant, I don't know, posts on hashtags or whatever it might be, and come back with maybe like a more real-time like rolling 3-day analysis of trending topics in the TikTok ecosystem as it pertains to AI and product management. Um, and so what it would have done is like obviously if you're logged into some of these platforms, it makes it a bit more seamless. And so I'm logged in here. I was going to literally ask it to create a call, right, to this endpoint, right? Put that in a skill and add that as our TikTok trend discovery to skill. And so it would have gone in here, gone into here, identified what my API keys and all of the other relevant sort of like configurations um and um data points are or information um in order to make that call. Um, it it might have had to stand up some sort of server. I'm not sure exactly, right? But what I've seen is like it can run an MCP directly within its platform. And so all you all I would need to do is like go and create and then this is the part that I might have forgotten to mention, but there are a ton of different connectors right um available out of the box, but you could also add a custom connector, right? Like you're literally just creating a virtual connection with different platforms and endpoints and that allows you to directly integrate with platforms like Amplify, for example. And so that was going to be sort of like the the like the the cherry on top of the demo. But fortunately, I ran out of credits, which is and I think Eileen just posted some like pretty annoying article about these guys being cheap as [ __ ] about this [ __ ] But um [ __ ] um easier on the fingers. Yeah, voice prompting is incredible. Um, for those of you that want to experiment with Pro, what I've also learned is, as you guys can see, like the limits are running out pretty quickly. Um, for those of you that might be a bit more technical, understand that like certain tasks that require like more computer vision tend to be a lot more resource-heavy. And so if it's like, yo, go to a website and just find me something, it's going to be much more resource intensive than you might think versus like, hey, I really like these types of design elements on X website. Go and identify that for me, right? Or if you're a bit more like narrow about the the task or more explicit about a task, especially on how to accomplish it more simply or recess or less intensive re from a resource perspective, then you're going to have better results in terms of res um like your token use.
Let's see. So the browsers only need a really if you need to. Yep. Um, but I'm not sure about you, Cat. Like if if you're like I I I I have nothing on my local machine. Everything's in the cloud for me now. Everything's on like different like Google Drives. Um, cuz God forbid your machine gets, you know, stolen or broken, right? Like that's the last thing I'm trying to deal with. Um, and so it's super buggy, mayor. Um, and it is known to be a credit hog, right? But um, like tech is super deflationary, right? Like Claude is leading the the pack right now and so their usage rates are probably through the roof. They just dropped that cybersecurity stack today, I think announced it today or yesterday, right? Um, and so I'm assuming like we're just getting a swarm of people coming in. Um, and so they're throttling everybody. And notice like they no longer tell I might have missed it. Omar, maybe you have the answer to this. Do they have like a token count on on Claude? >> I have never seen one. It's literally it's just all of a sudden you can't do anything. >> Dude, these guys Yeah, that these guys are getting smarter about it, right? Because like they can't >> tell you they can't be like transparent about like they're going from 10,000 to,000 credits, right? Nor are tokens going to ultimately matter in the long run, right? Um, but um, the the lack of transparency there tells me like that's highly strategic, right? Because they're going to go and throttle based on traffic, based on volume, based on their capacity, right? Effectively Co-work, I I don't think it's considered beta. The desktop um is still on version 1 something, I believe. And so in my mind, like anybody trying to survive in this in like anybody just trying to survive period are are they're building and shipping at a rate that we've never seen, right? Like I don't think QA is happening at this point based on what I've seen across different platforms. Um, and so yeah, while it might be beta um or like a really early version, they're they're building and shipping all day, every day. And so >> Dispatch Dispatch is beta. Co-work is in preview. There you go. >> There you go. Give me a second. >> Oh, wait. Um, James, I just found it. If you go deep into your settings and you go to usage, then you can finally see how much you're using, but other than that, it's not like it's not super straightforward. >> I I I see usage here. Yeah, but like it's a percentage bar. >> Yeah, it's not like [ __ ] Yeah. Gosh, I'm just gonna pull that up really quick so y'all can see. So, we just went into Pro settings, usage. Claim my credit for claim it. Come on. That's another thing. Some people have that for $200. Some people have for $20. >> These bastards. Um, like my weekly limit. I mean, clearly like they're trying to Actually, you said bastards. Like, I'm sorry, Claude. I'm I'm sorry, Dario. Like, I didn't mean it. >> Um, no. I'm sure I I'll have an opportunity to meet with their team face to face. you know, SPC's >> right after the Pentagon. >> Yeah. >> Um, but yeah, like and and as you guys can see, like we're really just getting started. It's it's skill like a PR. I wouldn't call it a PRD. Um, MD stands for markdown. It's just like a really simple like text file format. Um, I wouldn't go as far as a PRD, but if you think of it that way, you're probably going to get fairly decent outputs because you're going to be very thorough, right? So, you're you're going to be very thorough about what you ultimately want the technology to accomplish your output and or output, but it's a that's a cute way to like make it relevant to to like the product ecosystem. Um, SC mentioned unstable. Yeah. I mean, again, if if you're one of those guys that like are super organized and like all of your folders are labeled like really well and like um you you don't want like sensitive information on the cloud. Um, you absolutely can run it directly on your local directory. Um, and you're probably going to be use much more efficient with your your token use. SC. So, um, I I I'm just not trying to go and download 50 files and then dump it on my machine. Um, to to to do to go that far. I fully expect like all of the the cost of like this compute to to continue to go down. And so, um, and I'll just upgrade to the max because we're using this like like I I don't know if some of you guys might have heard Rich and I say this, like by the end of April, we are trying to automate our entire operation end-to-end. Like we do not want to work unless it's working with the people, right? And that's it, right? We want to spend 100% of our time or as close to that on our sales efforts and on our coaching effort, like where we're just taking on way too much, right? And so I think we're in a place where we're super close to being able to fully automate our back end. So it is they do, but it's not us. We'll give you a lot. Yeah. So the token used shown somewhere near real-time. So users don't have Yeah, I think I just pulled it up. SC um, but if you go to Claude desktop, let me show you to you one more time here. Go to here. Settings, usage. You'll see your current session resets in. I'm not sure why it's 2 hours and 7 minutes. That's interesting cuz it did it told me 8:00 a.m. I believe, but oh, 11 p.m. Okay, interesting. I'm not sure what that like cycle looks like specifically, but uh, let's see here. So you can edit existing files or fine-tune skills and then are the skills aware of other skills and can skills be autonomous ongoing task versus doing do I right now can be a yes you can schedule um, the skills are independent files, but Co-work is able to reference between the two, right? And so I asked it earlier like, hey, like I want to be I want to like persist just the source of information when we're doing our trend analysis. And so it was able to create another file, like a cache file. It was able to hand that context off between tasks or skills, that is, because Co-work itself had that context. The skills are just static files. Hopefully that answers your question, Cat. And sorry if I'm calling you Cat when you don't like it. Um, because I know some people it can be.
>> Okay. Uh Chance, we got three minutes left. Uh anybody have any additional questions before we go ahead and have some closing thoughts on it. Uh okay. So, uh, please let me go ahead and do a final count on the RSVP. Give me a second here. If you guys are Wow, we actually have a good Wow, we actually had some some some more people actually show up towards the tail end. If this is a if getting a product manager job, uh either you are trying to level up in your next product manager role or you are currently actively in the job hunt right now, I'm going to go ahead and tell y'all that the secret to the success of our program in being able to mobilize people into placement has been building in public, having a very public portfolio, leveraging this uh public platform because this is actually what companies want. You know that like that time of you thinking that 2021's going to come back. It's really not. So, if y'all ready to get it, if y'all are ready to get it, then by all means, please go ahead and schedule a call. Uh, if y'all are not ready to get.
It, that's okay because by the end of the year, it's not really going to matter. There is, I don't mean to sound alarmist because there is something that you can do about it right now versus doing nothing about it right now.
Everybody that's working in technology will become an agentic product manager because Claude has basically shown time and again that every plugin it drops, it wipes out the stocks that are in that domain. Like when it dropped its legal plugin, when it dropped its cyber security plugin, because companies, uh, Oracle right now just laid off 30,000 people, and it's not because they're not profitable. I mean, the guys like the families just bought like TikTok and Paramount. They're flowing in cash. They're taking that money and they're investing in data centers so they can go out there and automate more of the business.
So companies are either going to see you mobilize into an agentic product manager role, uh, where you can actually be the one to build out these AI agents and actually help them get it set up, or trust me when I say from a recruiting standpoint, they're already talking to third-party agencies. Uh, so like if you're going to automate recruiting with an AI recruiter, the best person to do that is a recruiter that's already working there right now at the company. If you're going to automate the customer success team with a third-party agency or service provider, the best person to automate the customer success management team is going to be a customer success manager right now. Uh, automating like QAS, the technology effectively exists to do it, but we still need somebody to see it through end to end. It should be a QA to go ahead and actually do it.
So for starters, you're either going to become an agentic product manager by the end of the year. You're either going to be a growth product manager where you're telling stories, building in public, uh, and helping them acquire users, or you're going to become a cyber security product manager. Or unfortunately, an acronym that's been thrown a lot around in the VC world is NGMI. Not going to make it. Uh, I hate to sound alarmist, but like that's exactly where the business is going.
Did you have something to add to NGMI, James? Absolutely. I, I think like I think it was this topic that like had me a bit nervous because it, I'm, I'm clearly playing catchup here, right? Like a single agent workflow is like is like yesterday's news, right? Especially if you are a product person, right, that is tasked with, um, owning product life cycles, like the odds are you're going to need to understand how to build multi-agent, more complex workflows ultimately. And as Rich mentioned, you are going to be the orchestrator, the owner, the deployer of this capability and tech stack. And if you are not wildly familiar, comfortable, and effective at leveraging this technology to force multiply, like companies are now looking at each individual employee as not the the the candidate themselves, but how much can they force multiply themselves, right?
Um, I think Jensen Wong on an interview recently said like if you're if you're a half a million dollar engineer at his company and if you're not spending at a bare minimum 250k a year in tokens, you're out the door, right? Because you're not doing enough to to go and lever up yourself and force multiply yourself using technology that exists today. Um, and so, um, and that is leaking very quickly into product management because we are at still, I would still say that we are, you know, the glue that, you know, brings engineering and the business together. Uh, but you're going to need to know how to navigate.
Yeah. Yeah. I'll basically show you guys this last slide. So if you guys haven't booked the call, please go ahead and actually do so, uh, to inquire about the program. But by the end of, I mean, 2025, they talked about, um, your, uh, 2025, they talked about your resume being your YouTube channel being your resume. Um, I'll tell you what's going to be your resume and your interview pitch by Fourth of July this year. It's going to be, bring your own personal AI agent product management team. Like you're gonna demo for them, like this is my product marketing AI agent. This is my UX research AI agent. They're gonna want to see, uh, and it's already happening. Two of my friends, two of my friends that are in recruiting, they're already seeing people bring that in and they're like, "Yeah, you want to see our operating, you want to see our like this is my this is my product management OS." And now you are.
Because every product manager interview goes from where you don't want to be is where you're trying to prove yourself. It goes from, I everybody mark down the sentence. You want to go from proving to confirming. Like when you go on the interview, you're confirming you have these skills. If you got to go ahead and prove, they're always going to think, well, you know, there's a chance that like this person's just not who they say they are. But when you actually bring your, uh, AI product management, like your whole team and say, "Yeah, this is my whole this is actually my whole team right here. Let me let me demo for you. Uh, let me go ahead and demo for James. Let me go ahead and demo for you. Um, you know, my team in action right here, right now, it is going to be a very different vibe because, you know, think about it. Somebody brought like if you were to go ahead and interview at like an actual Michelin star restaurant anywhere, right? You brought your resume versus somebody brought their knives and was ready to go out there and put on their apron and start doing like start cooking live. I think the other person is going to go ahead and take the job, right? That's exactly what it's going to be. Uh, they're going to be interviewing product managers the same way they're going to be interviewing chefs, uh, very soon. I'm talking about Q, end of Q2, Q3 tops because the more people that we bring in to set that standard, that is going to be the standard. And even though we're even if we're not going to go ahead and do it, my friends that are in recruiting is already seeing people do it. So, you might as well go ahead and start building that portfolio up right now. Uh, schedule the call. Let's go ahead and actually chat. Uh, yeah.
Anyways, I'm familiarizing with these topics. Can you help me understand the statement? Gartner predicts what 40% of agent more than 40% will be cancelled by 2027? Yeah. I mean, again, if you are building like an agent building platform and Claude comes out with like a major update, like I'm, I'm, I'm pretty close to like the VC ecosystem here. I know that cohorts of the last like 18 months, 95% of the projects are pivoting because they were literally wiped out by like a Google update or a cloud update or an OpenAI update, right? Like, um, building AI capabilities in this environment where we're not talking like ones or tens of billions of dollars, we're talking hundreds of billions of dollars that are being thrown around. Like it's just going to be borderline impossible to stay ahead of the curve, right? And so, um, I actually make the prediction that I think like 80 to 90% of the the funded startups over the last 24 months are not going to be around in short order. Um, and and it sounds like we're already seeing that happening, right? So, um, I, I think that's what it's referring to. Um, NAN is one that I'm keeping a really close eye on because I think they built a really neat tool, like a really good balance of technical and, um, like like layman layman friendly, but, um, I can't imagine them being able to remain relevant with the direction that these stacks are going personally. So.
Steve Blank, your startup is probably dead on arrival. Yep. And so that is basically, um, my, my, and I agree, right? Like I'm just seeing like.
Wow. We've been in, I've been in tech for like, Rich has been a recruiter for, you know, was a recruiter more than 10 years ago. I've been in tech for close to a dozen years now. Like I have never seen this speed. The speed of like building and shipping is just frightening, right?
I have not seen a website that looked like this since I was in high school. Jesus. Is it the speed?
2026? I thought I thought I thought you were going to give me like a no, like a like a professor Jean like this. Look at me like 1996 or something. Holy. Oh, wow. Okay. All right. Yeah, I 100% agree. Um, anyways, James and I are gonna go ahead and hop. We're eight minutes over for. I saw a lot of people actually schedule calls. Uh, for those of you that also sent me DMs, also on, uh, LinkedIn, I put James' handle here. We're, we're here to help everybody. Uh, the reality is like 95% of your competition has still not gotten on these tools yet. And based upon some of the questions that I've actually seen, uh, there is a actually a lot of catch-up work, but I think we can still actually, we're still very much within the realm of being able to catch up. Uh, when you're talking about that halfway point in the year, Fourth of July, uh, fireworks aren't just going to be the only things going up in the air. Fireworks. It's it's really not. It's like everything is like it happens like gradually and then like all at once, right?
Yeah. Yo, it it's real. Yo, guys, NGMI is very, very, very, very, very, very, like, like we're talking about like entire enterprises and industries that are going to get wiped, right? Like it will be gradual until it isn't. Like I actually think like you don't need to go and buy a Mac Mini and install OpenClaw because like Claude is about to make that effectively obsolete with like their next two updates, right? Like I, I, I, I literally call it, I look at it the same way as riding a wave. If you're all the way at the front of the wave, you're going to get crushed. Like there's that sweet spot on the wave where you're coasting and look like a badass, right? Like, like that's what product managers are here to do, beyond that surfboard, cruising that AI wave and just being a boss because you know just enough, you know how to orchestrate, you know how to lead, and you know how to deploy, right? And design, as Ammer mentioned, because I think understanding how systems work will be really valuable, um, as we start to see like agentic capabilities flourish and and and yeah, evolve. All right. Thank you all.
Awesome everybody on, uh, see y'all actually tomorrow night where we actually, uh, Yeah, that's right. Yeah, I'm excited for you. You guys how to build an AI agent, uh, with Claude, automate the whole process. Shir actually sent me the thing. U, anyhood, thanks. Yeah, thank you. Talk to you all very soon. Have a good night everybody.