What ChatGPT Agent Means for AI Agencies…

So not too long ago, we had OpenAI release their chat GPT agent feature. Now, while this may seem like a handy consumerf facing feature within ChatGpt, it actually has significant ramifications for the AI automation agency model long-term because we're not just seeing a new consumer feature. We're not seeing a manis competitor come out from OpenAI. We're actually seeing a fundamentally new platform of automations be created. I knew this was coming eventually and I actually predicted this on Twitter back in 2024. So in this video I'm going to walk you through how to understand this change, new AI automation landscape that we're going towards and probably more importantly what we are doing at Morningside AI to prepare for the shift so that we can be the first ones to take advantage of it.

So the reason CHP agent is so significant is because we've actually finally got a digital worker like someone who can operate a computer and one for one replace a human in a task or even potentially in a role down the line. This is so important because we now have a software that can use computers like humans do. And this form of agent is very different to what we're used to in building agents maybe programmatically on something like NAN or building something like a co-pilot where someone is chatting back and forth with it. We have more of a conversational agent. The best way to understand the kind of agents that we've been building up to this point is you can see them as dishwashers, right? They are machines built for a specific purpose. But what this new platform of chatbt agent and everything kind of downstream of it represents is essentially like a general purpose robot that can be programmed to do anything that a human can. So you may know of Optimus and kind of the humanoid robots that are taking off right now and that I mean they're still early stage but they expected to enter the economy over the next decade or two. And once you have a platform like an Optimus robot or some of the incredible ones coming out of China as well. Once it can do all the things physically then when given the right instructions it can feasibly do anything in the real world that a human can do. So it can replace one for one maybe someone who comes over to your house and assembles furniture for you. And so these new computer operating agents like CHP agent represent essentially the software version of those humanoid robots that can go and use a computer just like we can. And when given a task, it can move through the digital environment and do things that we tell it to just like a humanoid robot will be able to do. Now, going back to what I was saying about the dishwasher, we have for a long time been building these specialized agents and automations that do specific tasks. And these systems are largely relying on hooking in via API to everything. So, it's all highly engineered to work via API and over the internet. This is the very opposite of the spectrum of how agents can interact with the world. So, this has been a bit of a debate for a while and that's that goes back to the tweet that I put out back in 2024. This wasn't super clear how it would play out. There's always been two routes that agents could go down and the question was which one would get there first. It was between are our agents going to be all purely run programmatically and via APIs and everything is just going to be this massive web of APIs that connects together or is it going to be we're going to use vision models to navigate through the browsers and navigate through the digital world. And I was in the camp of saying hey I think it's going to be much faster for us to be able to get agents that can use the digital world as we do without having to fundamentally change it all than for everything to suddenly be hooked up by APIs. And we've seen this play out now and chat agent has kind of validated this hypothesis. But that's not to say that it isn't having the same issues that basically all agentic systems are having right now which is a massive fall off in accuracy and performance as a task get longer and long again which is a different conversation to have. And so with all that context out of the way, what does this actually mean for businesses? How is this going to change how businesses operate? How is it going to change the labor market? And I think the easiest way to explain this is by thinking of a headcount graph. How many people are working at a company at a given time? It might be say running flat at 50 people then it goes up to 52 then goes down to 48 and over time it's a fairly kind of step-based graph right you're hiring or you're firing but moving forward in the AI age we're going to see a mix of both these computer operated agents and the more specialized kind of dishwasher agents that we've been creating for a long time and say you go into a meeting and you are having some a strategy discussion with your leadership team and each of the five members of the leadership team then go and take action on everything that you agreed upon and they spin up multiple instances of either like computer-based browser based agents and they're also interacting with their agents and they're spinning it up. And then if you looked at the real-time graph of the headcount of how many people or like human equivalents are working, it would start to look like a crypto chart basically like it's spiking in massive surges and then falling back down to baseline and then spiking back up. So we have this ondemand labor and this ties all the way into the whole data centers and having making sure we have enough power and the enormous capex spend we're seeing into GPUs for these data centers. And then you can start to see how it all fits together, right? We are converting power and algorithms and these data centers into ondemand human intelligence and labor by these two different types of

agents. So how do these new computer use agents fit into the whole AI automation landscape? So let's just sketch it out quickly. Uh these are the categories that I've defined and I find fit the best. You have your firstly automations. So automations are things like your make. com scenarios, your Zapia if you're using Zappier, your NA10, these kind of automations that not necessarily using AI agents within them. um but they can also feature sort of automated agents within them but you know what I mean by building a workflow and we can call those automations. Then we also have uh what I would call AI tools and so AI tools are kind of discrete human operated uh tools that you can create. Say you go into relevance AI or you boot it on whatever you want. It takes in some kind of input data they press run and then it's going to do some sort of operation in the background using LLMs or doing deep research. It's going to spit out some kind of output. Maybe it's generating content whatever. So the difference between AI tools and AI automations is that these are human operated and automations are typically run on a schedule or on a trigger. And so that leaves us up here with AI agents. And so these two are technically linked because AI tools often are integrated into AI agents because AI agents take the place of the human using it. So there is quite a link between those two. But what we get here from AI agents is three distinct groups now. So we've had the uh kind of human operated. So human operated agents are like a co-pilot. Say you built an agent for a sales rep and you built it specially for them. Thinking back to the kind of dishwasher thing, you've built it specially for them to have a bunch of tools like updating the CRM or generating a pre-call report and summary, taking certain actions after it. It's essentially a human operated agent that is assisting a person in a role. You can see those as human operated. And then we have uh the NA10 has taken this by storm, but we have the automated agents. And so that's kind of combining uh AI agents with automations where you have them embedded into a workflow. And the AI agent itself is a core part of that automation and it's taking action depending on the input but it's essentially being run on some kind of trigger or a schedule rather than being directly operated by a human. It might be indirectly But what JBT agent now creates are these uh computer operating agents. Now you may be thinking are these two not kind of the same thing? Uh because if I'm using a computer operating agent I will be operating it like it's going to be human operated. But the difference here is that these are we can call these specialized and these are general. So the human operated again is like a dishwasher. We've engineered it specifically for a person to use in their role. Whereas a computer operating agents acts as more of a general assistant which I think is really the important thing to uh to draw the line of distinction between these is that the agents that we've making for a long time have been these specialized human operated agents or we've been baking them into workflows like NAN but now we have this new category of these general and computer operating agents that somehow need to fit in like when do you use the general agent and a specialized agent which is the next thing we're going to talk about in this video. So

as a bit of an exercise to help see where you would use the general computer using agent versus the specialized kind of dishwasher heavily engineered for a specific role, I've got a list here of pretty common tasks that either you starting your agency or your clients would be wanting to do. So, we can run through these and see which one performs best in which area. So, a task like competitor research. In my opinion, the specialist is going to win here. A dedicated research agent. I mean even using things like clawed research which I'm a big fan of um open AAI's the new GBT 5 Pro using deep research with that I think the specialist agent that is dedicated to research is going to perform better in a competitive research or any kind of research task it's going to be much faster and much more structured if you use say chbd agent to research it takes a long time it's going through it's not going to get anywhere near as as much scale as in many sources something like Claude's research does he get ridiculous amount of citations and sources so in this case we see that the specialist agent is winning the dishwasher and the specialized thing is going to be better. Uh for simple lead generation, again using AI, building an AI agent that goes out and actively searches the web, I can scrape information, can dig deep on the web to find their information, their LinkedIn profiles, etc. A specialist agent is still going to be much better um for doing this kind of lead generation. But then when it comes to things like making slideshows where we don't necessarily have specialist agents that are able to do this stuff, uh that's where these computer using and generalist agents seem to shine. And I know currently the slideshows that CHP agent is putting out are not that great, but that's obviously going to improve, but something like CHP agent is especially good here because it can do the research before you. It can analyze the data. It can then put it all into a slideshow in one motion. Um, so compared to something like a Gen Spark, which is very good at creating good-looking slideshows, it's able to do the full stack of slideshow preparation and creation. Whereas GenSpark is more of a really good slide builder. Um, then we get something that's a lot more specific and kind of a admin task, which is managing a cold email campaign. In this case, we start to see the generalist and computer use agent start to shine because it's more of a an annoying task where you're going through multiple apps. You might be clicking through here and clicking here. You might be checking a spreadsheet going back and when you're doing all the API calls between those, it can be a lot more difficult to get it to be reliable. So, so the fact that you can sign into certain websites and it can access the browser on behalf of you becomes very valuable because it can you can say in the morning, hey, can you get a report for me across all of our email campaigns that can go in there, it can check the numbers and it can come back to you after performing that task. And you can even get it to manage the cold email campaign. Maybe you want to run some split test and you say, "Hey, here's the split test I want to run. Can you go and check the performance of these campaigns? " And it can look on the screen and say, "Well, this one appears to have won. I'm going to put that one in and I'm going to set another split test up. " So these kind of admin tasks where you're interacting with an existing software or navigating the web seem to be where these generalist agents can really shine. And then another use case maybe something like LinkedIn outreach and in this case the generalist in the computer using agent also shines because just like with the cold email example instead of maybe signing into instantly it can be signed into your LinkedIn and it can just like a human automate some of these tasks that have traditionally been quite difficult to automate. So for the solarpreneur or the person like you probably starting an AI agency, you can start to use these to assist you in these kind of admin tasks that would typically take maybe hiring a virtual assistant or a VA and that would of course mean an increase in expenses for you. The key takeaway here is that we are going to see an entirely new platform or section of the AI automation landscape open up. And it's your job as the AI automation agency owner to know when you're working with your clients which tasks require the specialist automation or an entire like dishwasher to be made to really just do one task really well and when to use the humanoid robot or the generalist computer using agent that can do a certain task for them without having to set up a big fancy system or overengineering it essentially. Or maybe you're auditing the daily tasks of an employee and you see that they're going through spending an hour or two every day going through all of these different softwares just to get a kind of morning report. Well, in that case, the right automation to recommend is to have a computer operating agent to do that kind

of task. So to sort of to bring this home and kind of wrap this thing up, how can you prepare for this as an AR agency in order to be some of the first to monetize this new technology and this new platform? And the first thing to realize here is that this is just the consumerf facing application that we're seeing chatbd agent but there will feasibly be a business platform or kind of the API access that will appear and open AAI I'm sure is going to release this how it actually take shape whether it's like the assistance API with computer use or maybe they release an entirely different kind of AI employee backend where you can come in and set things up. You could very easily see a future where you could go in and be setting up these agents for companies where they have custom system prompts and they have custom knowledge and you've equipped them with custom tools just like you have with the assistance API right now. And then those customized agents can be called and sent on different tasks by various people across the organization. And if this is the case, what myself and the team at warning are looking at very closely is being of course the bridge between businesses and this new technology. How can we help them to make their teams way more efficient by using chatb agent correctly? immediately in the short term you can teach them how to use the consumer version of catchbd agent. I think the big opportunity there will be setting up personalized agents for each staff member. So say uh Lindy in her role as a marketer she has a certain set of things she needs to do. She can go in and log into all the accounts on her personal agent. We've prompted it and giving it full context of her. This is one thing that I'm sure you guys are running into when using any kind of generative AI for copyrightiting or marketing or doing anything really. the context of who you are and what you do is so essential for it to be as seamless as possible. So going into a company and saying hey look our service is to come into your company audit the workflows and tasks of each of your employees and then from there we will create personalized versions of this chatb agent using the business platform provided where we will provide it all the context of who it is. We will give it access to the correct knowledge bases it needs to perform and then we will train all the staff members on how to use it. So essentially, you are setting up a personal AI assistant for every single person in the company. That's going to allow them to spend way more time on the high impact and high leverage skills that are actually going to move the company forward rather than all of the admin stuff. So this, just so you know, is where we're looking and where we think the puck is going. I keep saying it, but you need to be skating to where the puck is going, not where it is right now. In short, right now you may need to be up to your gills, grinding away development projects and really building that technical expertise you need as an agency long term. But looking up ahead and seeing that, hey, this is where things could be headed is an essential part to do as an agency owner. And so I thought I'd just come on here and share what I've been thinking through what we're looking at Morningside because we've been seeing this coming for a while. And we think this is going to be an enormous market for AI agency and service providers to get into when the business layer from OpenAI finally comes out. If I'm honest, there's a bunch more that I could keep going into on this topic. gets really deep when you start digging into just how this is going to affect labor and the fact that is this going to actually replace jobs or is it going to create more work like when everyone has all of their basic tasks done okay they can start to take cracks at tasks that previously were not even feasible because it didn't have enough time like according to Aaron Levy the founder of Box he thinks that AI is actually going to create way more work because there's so much stuff in companies that we aren't doing because it's just not economically viable and agents can essentially either make those viable or it can free us up so that we can have the time to work on them. I think it's really interesting space um and I really get buzzed up about it and so I'll wrap it there but I'll I hope this has been helpful for you guys to really see this chat agent platform of computer operating agents and how they're distinct from the ones that we've been building for so long these dishwashers these specialist agents and how you need to be aware that there is a new frontier emerging for AI automation that you need to be aware of. So that's all for this video guys. Uh, thank you so much for watching and I will see you in the next