# OpenAI Just Changed the Game for AI Agencies (my honest thoughts)

## Метаданные

- **Канал:** Liam Ottley
- **YouTube:** https://www.youtube.com/watch?v=puo9rQ1xmGI
- **Дата:** 11.10.2025
- **Длительность:** 13:55
- **Просмотры:** 51,902

## Описание

📚 Join the #1 community for AI entrepreneurs and connect with 200,000+ members: https://bit.ly/46XADq4
📈 We help entrepreneurs, industry experts & developers build and scale their AI Agency: https://bit.ly/47kpUqY
🤝 Ready to transform your business with AI? Let's talk: https://bit.ly/43aiS5N
🎙️ Have a story worth telling? Be a guest on my podcast:: https://bit.ly/yt-podcast-application

🚀 Apply to Join My Team at Morningside AI: https://bit.ly/work-w-morningside
🚀 Apply to Join My Team at AAA Accelerator: https://bit.ly/work-w-accelerator

My Vlog/BTS Channel: https://bit.ly/LiamOttleyVlogs

If you run an AI agency or productized service, this breakdown shows exactly how to use OpenAI’s Agent Kit, Agent Builder, and Chat Kit to ship real client value now. You’ll see why platform shifts threaten SaaS but accelerate agencies—and how no/low‑code multi‑agent workflows plus white‑label chat interfaces let you deliver faster without adding headcount. I break down how the AgentKit production stack (versioning, observability, evals, cost optimization) can turn builds into sticky retainers while paying raw API rates instead of inflated platform credits. I also share how OpenAI’s “Apple‑like” ecosystem fits versus Zapier/Make/n8n and what text‑to‑workflow means for your roadmap in automation and multi‑agent orchestration.

⏱️ Timestamps:
00:00 - What We're Covering
01:30 — AgentKit, Agent Builder & ChatKit: what’s actually useful
03:21 — Multi-agent low-code (guardrails, MCP)
04:05 — Ship client-branded UIs with ChatKit
06:20 — Production toolkit: versioning, observability & evals

## Содержание

### [0:00](https://www.youtube.com/watch?v=puo9rQ1xmGI) What We're Covering

Hey guys, I hope you're doing well on this pretty eventful week in the AI automation space and I just wanted to jump on here and give you guys a bit more of a now that the dust has settled um what my thoughts are on this new open day update and really what it means for AI agencies as a whole. I've had a lot of questions around like oh does this mean our tech stacks changed now? Are these certain tools uh redundant? Does this mean there's going to be more people entering the space and it's going to be more competitive? There's a lot of anytime there's change, of course, there's going to be a lot of fear and uncertainty. And I've had a real think on what these different features mean, where they fit into the landscape, and um really what myself and the team at Morningside are going to be doing. So, I just want to break that down for you guys, hopefully give you a bit of clarity and kind of a mental model to understand what this platform is uh particularly like the agent kit and things like that, where that's going to fit into your stack and where it's good or bad for us as agencies. So, first things first, we had a lot to cover in there. I'm not going to cover it all in this video, only really the stuff that's super relevant to us as agencies. Obviously, you've had the chatbt update where you can now have these applications that can sort of be used within chat GBT. I won't go too deep into that because we've had these kind of is this an app store moment multiple times now. You've had chat GBT plugins. You've had GPTs. You thought there were these app store moments, but they've flopped. So, I'm very hesitant to make a bet on that. They promised us monetization and GPTs, which never really came. And so, overall, I think there could be an opportunity there. I think just stick a pin in it as if this does become an app store moment where you've got a new platform for building on similar to like the app store and iPhones and iPads and things like that gave birth to Instagram and Snapchat and these sort of applications. There could be that sort of opportunity brewing there. But I think

### [1:30](https://www.youtube.com/watch?v=puo9rQ1xmGI&t=90s) AgentKit, Agent Builder & ChatKit: what’s actually useful

for now we'll just focus on what we can use here and now today as agencies and that's primarily going to be the agent kit and off that the agent builder and the chat kit as well. So these are massive updates and they're really this is another classic example of why being an agency in the space is like such a good idea and cuz right here you can see I'm sitting here grinning not losing a wink asleep over the stuff whereas if you are in SAS and if you built a software you've just had the entire ground pulled out from under your feet and this is what we've been saying for a long time is that by being in the agency model you get to benefit from all the updates you get to bring them to your clients you get to take this awesome new stuff you get it to benefit essentially from these new dev tools that we get which is over time you just track It's getting easier and easier to deliver which is even better for us as service providers. So when I first started it was very difficult using text adventure 3 text completion models to like jerig them into chat bots and just from there it's gotten so much easier and this agent kit is really just the latest update in making the development and delivery of creating value with AI even easier for us as agencies who are really the ones who are out there doing a lot of the work. So out of this new agent kit agent builder and chat kit. What are the wins for us as agencies? as well. Firstly, we have that building multi- aent systems is now accessible via basically a no code platform. And this is great for all of us. You might have had to use lang chain before in their multi- aent framework to get the same kind of functionality or you would have had to sort of rig up multiple agents in something like NAN. But this is just a continuation of the trend that I spotted back in 2023. And I got clowned quite a lot for it back then and saying that you could build incredibly valuable AI systems for businesses on no code tools and we've sort of come full circle now that open AI has actually released their and I think the writing's on the wall now how big this no code AI automation thing has been over the past coming up 3 years now. It's just another step towards that and now not only do we have agents that you can build but you have multi- aent frameworks which is what agent kit allows you to do. So visual editor drag and drop with a bunch of handy features like guard rails and you've got MCP servers. It's just incredibly great news for anyone. Even if you are an agency like us at

### [3:21](https://www.youtube.com/watch?v=puo9rQ1xmGI&t=201s) Multi-agent low-code (guardrails, MCP)

Morningside who are doing a lot of custom development, we're looking at this and saying, "Hey, can we migrate some of our stuff over or can start to use agent kit and agent builder to deliver a lot of our agency services? " Like the insight that I had two and a half years ago now was that clients don't care how they get it. They just want the thing. Like they don't care if you custom code it. build it with like sticks and lollipop sticks and blu-tack. And we're having to look now as an agency, okay, should we maybe start to try to do prototype? can we approach this certain project? Maybe it's a chatting database like we're doing with one of our larger clients right now. Can we do that through the agent kit? So, I'm getting my developers to experiment heavily with that now. And I think there's a huge amount of potential there for agencies to start delivering even what we traditionally consider kind of more custom projects through something like this. The second big win for us here is the chatkit and being able to build interfaces. I've

### [4:05](https://www.youtube.com/watch?v=puo9rQ1xmGI&t=245s) Ship client-branded UIs with ChatKit

been screaming this from the rooftops to anyone that would listen that the issue right now for many agencies and for a lot of these agent building applications is it's difficult. But there's no one that really allowed you to easily create an agent on something and then integrate it into a custom front end. And for us as agencies giving say a link to like a relevance AI agent that has like the relevance branding and it cheapens the service, right? But you take that same agent that's powering that interface and you connect it onto a custom white labelled front end that your client with your client's branding and some sort of custom widgets and stuff like that really brings a chat app to life and you have a much much more valuable service. So this chat code essentially allows you to vibe code together a custom interface to connect your agents to which is incredible news for again people who are coming into this without as much technical skills but also for people like with extensive development teams who want to be able to more quickly create these kinds of interfaces and deploy them. So chatkit I think is a stroke of genius from OpenAI has been much needed for a long time and I really urge you guys to get in there and start seeing how you can connect these agents from the agent builder over to these interfaces as well. So massive one there for accessibility to building these because you typically would have to say you're a person with an AI automation skill set. You've learned entertain or make. com you got a really solid skill set there. But when it came to oh the client wants custom website or custom chat app but built you oh now I have to go and like get a developer to do it. I don't really know how to do all of this. So it's just a huge like value ad that you can offer to your services and I just think it's a great development for us as agencies. Another big win for us is of course just generally faster development and deployment but also the management and optimization of agents in production. I think if you look at what OpenAI is doing here, I see it more so as they see that Anthropic has really crushed the AI coding market and the amount of money that they're making through the coding tools and through the API usage there is astronomical and OpenAI has been very heavily dependent on Chachi BT as the way that they make a lot of their money. So I think this is a smart play for them to say, "Hey, we need to get we know there's issues with getting agents into production and these automations. Why don't we focus on making a really good toolkit so that we can be the platform that people come to put these into production? " And then when once you stack enough businesses using that, it's going to be incredible amounts of API usage going through there and that's how they're going to get a really solid and sticky base of customers um to drive their business revenue. So if you look closely at the agent builder, it's got things like versioning. So when you can publish it, then you can tweak around, and then publish again without affecting the live deployment. That's of course

### [6:20](https://www.youtube.com/watch?v=puo9rQ1xmGI&t=380s) Production toolkit: versioning, observability & evals

great for us in producing these production grade agents. We also have stuff like the observability page for breaking down what happens in each run of the agent. And I mean I just think that's incredible because you can start to see I mean it's pulling in a lot of these kind of lang chain features and other platforms that have been doing all the LM ops and it's bringing it all into one place for us and which is incredible for us as agencies. But now you can see, hey, if I wanted to, I think there's multiple stages to development of these AI agent use cases for companies. And yeah, sure, it's just getting a proof of concept on MVP up and testing it around. But once it's in production, I believe a lot of the AI agency value that's going to be created over the coming sort of few years is as development becomes a lot easier, they're going to come to the specialists who know how to get more out of it. So it's like, can you break down that observability? Why is it taking so long? What models are being used? Can I use a cheaper one? Um how is this like how can I bring the timelines down and make this run better? And so that optimization and the management of production AI agent systems particularly these multi- aent systems I think is a really great way for agencies to start to get ongoing sort of retainer based offers and also just overall provide a lot more value than just the initial development and deployment. So the observability love that I think it's an excellent tool for us and it's also got those eval so you can very easily type in hey did this agent do X and you can evaluate all of the threads all of the traces I believe they're calling them now for certain criteria. Is it did this happen or does it take this tone? And again, it's just an incredible tool for us to be able to debug and sort of dig into how these agents are operating in production. So great production grade toolkit for us as agencies. And that sets us up for great retainer offers with our clients. And finally, another big win for us here is that this is essentially cutting out the middlemate. If you think about it, when you have use any kind of AI platform right now, they're using the underlying models. They're using OpenAI, Anthropic, all these other Google Gemini. They're using these in their products, but they need to make money too. So they're putting their margin on top. So you're paying the in some way you're still paying for the open AI usage costs for the for say GBD5 and then this platform is usually slacking on top their margin on top and then us as the agency or for our end customer is going to end up having to pay more in order to make sure everyone's still making money. So you can see this as a way of kind of cutting out the middleman and by using this open AI ecosystem you're going to be getting essentially the raw API costs which is very very cheap compared to a lot of these platforms and then ridiculous credit costs on some of these automation platforms. So, I think this is great overall for us as agencies if you're running it as a retainer offer and you're the one paying the costs, but also if we're developing it for our clients, we can say, "Hey, look, this is only going to cost a like sort of 1/5if of what it would have cost on these other platforms because we're able to use this OpenAI ecosystem and cut out the middleman. " So, I really love that aspect of it as well. A question that I've got a lot recently around this is with this becoming more accessible now, is this mean that the model's kind of toast? Businesses are just going to do it themselves. What does this mean for AI development moving forward? And I think it goes back to if you just look at other industries like I mean social media marketing is done via the Facebook ads manager and there's still an enormous industry of specialists who can get the most out of that. So I'm definitely in the camp of when there are core business systems like agents like particularly on the acquisition side if it's bringing in more business if it's on customer support they want to use experts. They don't want to go in there and just you're not going to risk getting the one employee who's maybe spent a couple months tinkering around with AI to build your entire customer support system out on this. Right? these things the value occurs to us as the agencies who are delivering on it's dev tools it's a developer dev day right they're tools for us it's not like business day so just keep in mind that these things are helping us they help us to deliver more quickly and sure the odd business might go in there and try a tinker around but at the end of the day it still seems big it still seems scary and the specialists who can get the most out of it who can do that breakdown of okay what where's the money being spent okay how long has it been taking how can I evaluate this to make sure it's working correctly how can I optimize it over time that is really where a lot of our value as agencies go when I mean the development becomes so easy with these things that we just go in there set it up and we work with them back and forth to get really the best results out of it and you can just start to offer way more systems you can do more for your clients for the same amount of time so it's a win for us and them another question is of course what does this mean for platforms like NAN and Zapia and make and where does it fit into your stack when should you be using this those so I think the easiest way for me to understand maybe might be helpful for you guys as well is I see this play as kind of like the apple of AI automation where something like N8 you could see as maybe Microsoft or Android where it's more customizable um it's not as kind of a closed circuit or closed uh closed environment and closed ecosystem. So the Apple model with their devices and things like that was that we're going to have this family of devices. They're all going to kind of talk together and we're going to have our own operating system that is very closed and we have to sort of it's very managed. We manage the whole user experience across all devices. And you could look at agent kit and the agent builder. That's kind of the start of this open AI ecosystem of AI automation where it's the user experience is going to be so seamless where they may not have the absolute best model in every single kind of uh use case, but if they've got a really good model, they've got a couple different versions of it. you got GPT 5 Pro, you've got the normal one, have a small like faster performing one that's a bit cheaper, you can go down to GPT40. As long as you've got a good mix of these models that serve different purposes, then within the agent kit, it's able to assemble those in the right way, and you're not necessarily like, man, I really wish I could pull in a clawed model here. They're essentially saying, "We've got pretty much everything you need here, and we're going to make it super seamless for you to set them up and connect them together. We're going to handle all the difficult stuff of making sure they're talking together. " And when you add on to that something like a text to workflow or a viable automation where you can type in and say, "Hey, I want to build a workflow or an agent system that does this. " And then they're able to just piece it all together with their own ecosystem. And literally within seconds, you can have a workflow that's ready to go. Take a few minutes to build a custom interface and deploy it. It's a pretty compelling direction that OpenAI is going in terms of making it easy for developers to do like everything. And the question is, are you going to start to use that more or do you want the flexibility and customization of something like NA10? So I think they definitely got their work cut out. It seems right now that the agent builder doesn't have all of the triggers that you probably want in something like a make. com. And it also doesn't have a lot of the nonAI parts that you need for building out kind of any workflow. So, it's going to be interesting to see if they pursue that and say, "Look, we are actually coming for their lunch. We want to build out essentially a make. com or an NA10 equivalent, but within our own ecosystem, or they're saying, no, we want to do more of this kind of multi- aent thing only and not necessarily taking over every automation use case. " So, the jury is still out on that one. I think a good way to maybe understand what this is in its current form would be like a a voice flow but without the control that voice flow gives you in terms of being able to write specific messages. So like it's very nondeterministic whereas something like voice flow is a good mix of like the agentic capabilities with also being able to really make a deterministic system of this happens and sends this message. So that's sort of how I I'm thinking of it right now. So yeah, those are my thoughts on this dev day update. I think it's a massive win for agencies. Anytime we get new dev tools, it's a massive win for us. And that's why we do this stuff, right? It allows us to just surf along all of these massive updates. We have had updates like this previously like the assistance API and GBT's release when that came out just like this has. It nuked so many startups because they were taking they were kind of building ahead of open AI in many ways and they're adding custom loge bases and tooling to the chat completions endpoint and then they dropped the assistance API and just sort of absorbed all of those different use cases. And so more than ever, it is a dangerous time to be in SAS, but a incredible agencies where we get to benefit from this and development gets easier. We can deliver much faster. We can deliver a broader range of services without having technical capabilities. It's just a massive win for us. So, I hope you guys are all as excited as I am to jump in and start building this and delivering it uh some client projects with this stuff. So, yeah, the sky is not falling. Everything is all good. And uh if you guys want to check out more on this kind of vibe automation idea, which I think is very relevant um as this development becomes easier and easier, I've got a video up here breaking down what Vibe automation is um and how this text to workflow revolution is really coming for agencies and businesses wanting to adopt AI automation. So that's all guys. I hope this been helpful and I'll see you in the next

---
*Источник: https://ekstraktznaniy.ru/video/11772*