Transcription
You know, I hate to be that guy, but paperclipip and other agent frameworks where you're controlling an autonomous swarm of 500 CMOS just don't do anything. It sounds really futuristic and it sounds incredible, but at the end of the day, it's setup porn.
Have you seen these threads on X and and LinkedIn and now YouTube videos where people are like, "I just set up my paperclip Hermes agent Maximizer to run my Telegram topics on Veriz"? It's like, first of all, what the heck are any of these platforms? And second of all, what have you actually shipped this week? What have you actually built? How much money have you made in the last 30 days by jerking yourself off with a with an open claw? That's a good one.
The point that I'm trying to make is it's basically like organizing markdown files in a folder. And YouTubers like myself, people on X and LinkedIn and stuff like that are all incentivized to make every one of these new agent frameworks and developments much bigger than they actually are. I've tried removing names and stuff because obviously I don't want to put anybody on the spot, but check out this fell. This is the peak 19-year-old experience. He says, "You know, it's 2:10 a.m. I got three cloud code terminals on the MacBook, setting up Hermes and paperclipip on the monitor, and whisper flowing Hermes inside of my Telegram topics. What more could you ask for?" This reads like freaking ancient Egyptian hieroglyphs to anybody that's actually making money. "What more could I ask for?" How about some MR, little bro? Like, come on. What are you doing here that's actually economically valuable?
Or maybe uh this one here. You know, I really didn't want to manage a team, so XYZ name redacted found a solution. Paperclip, the open source project. What she showed me. Paperclip leads her agents using its project management setup. Humans on her team use it to assign tasks to agents. Agents delegate tasks to humans or other agents. Paperclip turns your goals into agent tasks. It turns an SEO audit into agent tasks. And finally, it organizes OpenCloud agents or even creates its own agents. Can I just get to the real meat and potatoes of whatever this person's business is? What are you actually doing? What is actually like external facing deliverable here that adds value? The only thing here out of these six tasks that is not completely masturbatory, even though it is still quite masturbatory, is it turns an SEO audit doc for example into agent tasks. Like this is just doing project management for agents. This is just doing project management for agents. This is just doing project management for agents. This is just doing project management for agents. This is just organizing some BS tool to do more project management for some other BS tools. Like the only thing a client or a customer would ever actually see is some sort of SEO audit doc here. And it's not even doing that. It's just turning that into a bunch of tasks for agents.
I don't think people realize, but like clients and customers and and people that are on the other receiving end of a business, they don't care how you do stuff. They only care that the stuff gets done. Likewise, when I see people make a mountain out of a molehill like this, it's like, what are you actually getting done? And then they're like, "Well, I actually orchestrated a team of whatever the hell." I'm like, "Okay, great. So, um, why didn't you just do that yourself? It probably would have taken a hundred times less actual man-hour than you trying to build this incredible wrapper around like an SEO audit doc. And by the way, what the hell are you doing with an SEO audit dock anyway, man? It's 2026.
Okay, check this one out. Zero human company install using paperclip and claude. You know, this looks really cool. It like, you know, went relatively viral on X and whatnot. You have like a little terminal with, I think, ghostty on the lefthand side. Then you have like the actual paperclipip dashboard. And I mean like I was like, "Wow, this is so cool." What are they actually doing? Have you can you read this? Hopefully you can. It says uh, "Community lead agent is defining the community content and launch strategy." You know, this one's researching and evaluating community platforms. This one here is improving the admin dashboard UX. And this one here is hardening the assessment pipeline reliability." Like three out of the four of these are just You're just shipping and building stuff that like helps you ship and build stuff. You're not actually shipping and building anything. For instance, "research and evaluate community platforms." I mean like that's just a Google search, man. Shouldn't you, as somebody that's about to like start a whole business or community on this thing, build with AI, return my time, whatever the heck, shouldn't you just make that decision on your end? "Improving the admin dashboard UX?" Oh, I'm sorry. Is paperclipip not good enough for you? So, you're going to make your paperclip dashboard better so that you can make your paperclipip dashboard better? How about "hardening your assessment pipeline reliability?" Guarantee you this is some assessment that you know other agents are going to use to assess the quality of the deliverables or something. Like one out of the four things here are actually probably moving a needle in some way. And it's something that realistically like this fella should not actually be having an agent do. You should just like sit down and think about it for five minutes. He'd do way more.
Okay, I got like another standout one here. "Day two out of 30 trying to make viral videos on TikTok with Paperclip." Paperclip started and it's already conducted research with Verlo to identify trends in the fitness industry. Then it researches some websites on Reddit and Hacker News. The main bottleneck though is agent- media. I'm not sure if it can create good enough videos yet. Sometimes the simple videos work better than complicated ones. Then everything is scheduled and post is. Sometimes I'll add the the Larry skill. Let me know if there's any other service that I need to connect to paperclip cuz that's obviously your bottleneck, bro." It's like, what is this thing doing? It's just doing research with Verlow, Reddit, Hacker News, and scheduling and posters. The only thing that matters when you make videos is are your videos good? Why are you like spending 99% of your time around stuff that is not about the main thing? Just like make the main thing the main thing. Just figure out if you can make good quality videos first. Let me tell you, I could have a quadrillion AI agents churning in the background, but I wouldn't be able to create the sorts of videos that I create today because at the end of the day, like I am the bottleneck and just like this fella is the bottleneck. He's currently not really realizing that. So instead, he's trying to do everything possible not to do the work that actually needs doing.
You know what? All this reminds me of this just generational meme where you have a simpleton on the lefthand side and what tool does he use to get his stuff done? Apple notes. You have this sort of like median average fell over here and what tool does does he think he needs? He thinks he needs Readwise and an Quizlet and Adobe Acrobat and Notion and a bajillion other tools. But then what does like really brilliant like person that actually moves the needle need? Just you know Apple notes. The point that I'm trying to make is it's never the tool. It's just how you use the tool. And so in the case of paperclip, which is just so fundamentally misaligned in terms of ag and capability and then like what human beings can do that it's not even really worth talking about, although I will talk about in a second, the only thing that you actually need in order to get something done is is really just the space between your two ears. And this is somebody this is from somebody that like I I work with AI on a daily basis. I sell AI on a daily basis. I I make money teaching people AI. It is not in my best interest to talk about models and frameworks like this. But the reality is you do not need models and frameworks in order to do things. You just need like a healthy dose of gumption and high agency to go and actually do the thing yourself, not delegate it off to a quadrillion agents.
The last point I'll make on this is I don't know if anybody remembers this hilarious exchange between Parag Agrawal. He was I think the former CEO of Twitter or something like that and then Elon Musk right before he purchased it. Parag clearly is in rage mode over here. Sends like a 10-line message and you know if you go back further there's a lot of these sorts of exchanges between the two and Elon just cuts straight to the meat and potatoes of it, which is just, "What did you get done this week?" So anytime I have anybody be like, "Nick, why aren't you using paperclip or Opus Maximizer? Why aren't you using Ventro to manage your Telegram topics?" It's like, "What did he actually get done this week? How much money have you actually made?" If you look around at the people that are actually out there billing the largest amounts, uh people that are actually like moving the needle in whatever substantive way, shape, or form, of course, they use these tools, but the tools aren't the main thing. They're tools, right? You use the tool. You don't let the tool use you. And unfortunately, I think some of our tools, like social media, for instance, are now at the point where they're very clearly using and abusing probably some of the less capable among us for nothing more than just like internet hype points and clicks.
There's so many problems with stuff like this, it's not even funny. I mean, why would you organize an AI company the same way that you would organize a human company? You're borrowing an organizational hierarchy chart or framework from a 100 years ago. We've been organizing companies in this way because like it or not, that just tends to be the way that human beings do good. We have some sort of centralized power, right? CEO that delegates to a few other centralized powers, CMOs, CTO's, COOs. Those in turn delegate and so on and so forth. But whether it's like a consequence of um, you know, our internal Dunbar numbers or just like how much management load every individual person could take on, you know, human companies tend to look very, very similar as you get bigger. There's no fundamental reason why an AI company has to look like that. There's no reason why you have to have an AI CEO delegating to an AI COO and an AI CTO who delegates to their own sub agents. You're just attempting to fit this concept of an AI company in a shape that we are all very familiar with, which is obviously that of a human company. An actual extremely efficiently organized AI company would look nothing like this.
AI intelligences are really spiky. They are vastly more parallelizable and they can also do specific tightly scoped tasks thousands of times faster than human beings can because of these very, very big differences. There's no reason why you'd want like one, you know, senior developer agent doing a task. Realistically, you'd want to spin up 50 of them. Have all 50 of them produce the same deliverable with minor differences. Calculate the mode, the median, figure out the average things, the outliers, collapse those into some, you know, formal description of what the thing is that they just made, and then like, you know, create it with all of this information in mind. Human beings, on the other hand, are fantastic at like zero-shot tasks. Uh, have you guys seen ARC AGI 3? That's the the new like AGI benchmark essentially. AI freaking blows at the ability to do things long-term in a reliable manner. Human beings are kind of like sniper rifles right now. It's almost like self-adjusting sniper rifles where the bullets can change trajectory. We're capable of like learning on the fly and feeding back much more effectively at these AI agents, which instead tend to just sort of be like in a beeline trajectory just because of the nature of our intelligences and our various task domains and specializations. Like the best most efficient company in the future is not going to look like the way that human companies are run today. And anybody trying to pretend like that is the best way to do things or it's some revolutionary thing. They're just trying to garner engagement and clicks and a bunch of other silly stuff that I really want to talk about right now. So hopefully what I'm saying here makes sense and I'm open to, you know, feedback and some sort of like debate or argument in the comments. So if you guys have some interesting points that I might not have considered, just let me know down below.
My take on all of this stuff is agent frameworks like this may eventually get to a point where they're good, but this paperclip thing and this theme of organizing, you know, agent teams the exact same way you'd organize human teams. This is just like people looking for some form of engagement. We saw the same thing with like NADN AI agents uh, I don't know, two and a half years ago or so when it's like, "My NADN agent controls my whole life and pleases my wife." It's like, "No man, it probably doesn't. It's just a lot of motion but not a lot of movement, if you catch my drift." All right, have a lovely rest of the day. PSO out.