# Why Most Businesses Should Stop Buying AI (Right Now)

Channel: Liam Ottley
Video: https://www.youtube.com/watch?v=skpYjEGOdCs
Duration: 20 min
Language: English
Words: 4932
Transcript page: https://viewrankai.com/tools/youtube-transcript/skpYjEGOdCs

---

[0:00] Most businesses should stop spending on AI systems for at least a year. And I say that as someone who is selling the exact thing you'd be buying. So either I'm wrong or I'm about to cost myself a lot of money. And I've sat with it long enough to know which is which. And the truth is that most businesses right now are buying the wrong thing when it comes to AI. There are two groups of people who are watching this. There are people who sell AI systems and the people who are going to be buying them. So let's start with you. Let me guess where you are. You're on a claude plan. You got

[0:26] maybe a couple projects set up. you got some integrations, maybe a few skills, and that's to be fair, it is a good start, but nothing in your business runs any differently to how it did in 2023, or you paid for a custom setup. Maybe it's like a clawed code AI operating system or AIOS, and it fits your business like a glove because it's built custom, but it also feels like it's held together with duct tape. And it is because the people who have built that claude code operating system for you were improvising. And that's because nobody has actually figured out what an

[0:53] AI native company actually looks like. Not the labs, not the startups, and not even me. Like for example, I run an AI transformation company. We go onsite for seven days with my clients and we do a full AI makeover of their business. And over the past 4 months, we've changed our stack multiple times in terms of the workspace we use, how much custom code we use. But to understand what's actually missing and why nobody has built this perfect solution for an AI native organization yet, I want to talk about cars. Yep, we're going to talk about cars. So every major technology goes through this period that we're in

[1:23] right now. For example, Henry Ford. He had a a working car engine in 1896. But it still took him seven years and two failed companies and a pile of prototypes before a normal person could actually buy one of his cars. And that was the Model A produced 1,000 units. Then it was another 5 years after that that the first mass market car, the Ford Model T, was released. It turns out that

[1:44] the engine itself wasn't the hard part. It was the wheels, the suspension, the spark plugs, the carburetor, the exhaust, and the steering wheel. and turning it into not only a performant but reliable and economical package. For the past three and a half years or so, there's been at least three major competing ideas or camps on how businesses would actually adopt AI. We had the train your own model guys. We had the train a whole bunch of small models together and then we had the just give one really good model a whole set of tools and let it run. And out of those three camps, it was the third one

[2:15] that won. And that's what we now call a harness, which is one model and a whole set of tools and a space to use and access that model. For example, Claude Code is a harness. Crockpot is a harness. Codex is a harness. So to compare Henry Ford's cars to where we are today, the harness is the engine and the AI platform is the car body. And we have the engine, but it's only like 10% of the way to where we're actually trying to go, which is a performant, reliable, and economical car. So, every one of those car pieces is being invented and broken right now in just

[2:44] the same way as the fuel system and brakes were in 1898. And I call this phase that we're in the waiting room. And eventually, it will end just like the weight for a car eventually ended. And one car design eventually won and everyone copied it. And within a decade, every car on the road had the the same shape and the same parts roughly like a generalized model. So, what are we actually waiting for in this waiting room right now? So, the next step for the AI space is an all-in-one AI platform for business. It's the thing that's going to turn one agent harness doing tasks into a full real business

[3:13] that is running on and alongside AI where humans and agents are working in the same environment running off the same brain with a record of everything that is done both by humans and agents. It's like one AI platform or shared space for agents and harnesses and humans to work together. And right now the system breaks into seven separate pieces that I've identified through my

[3:32] own work of doing this for businesses. And in the current state of things, no startup or lab has shipped all seven pieces. So the first piece is the context layer and this is basically the documents that every agent runs off. So nothing drifts plus memory scoped down to what each person does. So you have organizational context and then role-based context. Secondly, you have the data layer and that's what is often called a second brain where every doc, every call recording, every source is all in one place that agents and humans can query. Thirdly, we have the agents themselves which are like AI employees with a role and tools and connections

[4:04] and their own VPS often their own computer. Fourth, we have what I call a workspace. And this is where humans and agents actually work together. Fifth is the ledger. And this is probably my favorite because most work in businesses never gets tracked anywhere, but in mine and also my clients companies once I've finished working with them, we know what is being done across the or by humans and agents, what files got made, links to sessions if you want it to really dive deep into figuring out what's going on. I can ask without having to chat to any of my team. I can ask what is happening with this project. My agents

[4:32] will query the ledger and let me know what work has been done on certain projects. Sixth and something that is often overlooked is an admin panel which is a robust system for managing humans and agents and how they access data and tools monitoring token usage costs and work activity across the organization. So a proper admin panel and finally number seven is custom apps. So when there's some kind of workflow that the AI platform itself doesn't cover a cod and agent can build its own custom application and either host it somewhere externally or pin it within the AI platform itself. So the idea here with an AI platform is a self- sustaining AI

[5:04] operating system for the business where you have humans and agents working together in like a a warm cuddly software cocoon. And when you lay it out, it's clear that this is the natural next step for the AI space. We started off with models which ran from late 2022 when chatbt came out to early 2025. Harnesses ran from early 2025 when Claude code came out to this year where we are right now. And I believe that

[5:27] platforms are the next phase of this. It's like AI Russian dolls where each smaller one fits nicely into the next where you have models inside harnesses and harnesses inside platforms. If you run a business, you're already in the waiting room right now. So, you've got two options. You either wait patiently and time your move or spend tens of thousands of dollars on tech that's outdated by spring of 2027. But waiting

[5:48] isn't passive and it isn't free either. You're paying an opportunity cost because while you're waiting, there are two things happening. The first is the temptation to just go out and build it yourself or pay someone like me to do it. and I want to talk you out of it, which is strange, I know, because it's the main thing that I sell. But right now, people like me are building primitive versions of these AI platforms by hand for our own businesses and for

[6:09] our clients, and they absolutely work. Like, they get incredible results, and the founders who use them are absolutely wrapped with them. So, these custom AI platforms or or what I call AI operating systems do work, but long term, this ultimately isn't a problem that a fiveman dev team is going to solve. like building a multiplayer AI platform is not something that a fivep person dev team is going to solve. And the reason I say this is because in 2023 I half of the startups that I knew were building some kind of custom knowledge system on top of the chat GPT API. Those systems took months and months to build and then

[6:41] open AAI shipped GPTs and the assistance API and a whole wave of startups was gone in literally an afternoon. Like OpenAI dev day is literally starting in about an hour as of time of filming this and another wave of people who built ahead of the labs are going to get absolutely crushed as well. The thing is that the big AI labs, they just they always do this. If you build ahead of

[6:59] them, then they'll just run you over. But if you stay light and and pick your shots carefully and know what you're doing, then you can be able to ride their wave and benefit from all of the billions of dollars they're throwing at some of these problems. So here's the problem with a custom build right now, and it has nothing to do with the quality of the build. Like I said, they're fantastic. they get great results, but from my own experience, there's issues with access and permissions and scaling out to dozens, if not hundreds of members. It's not quite there yet on the custom stack. It is good for small businesses right now,

[7:26] though. A key issue with these custom builds is that it's held together by the person who built it. It really starts to feel like duct tape when the engineer who built it for you leaves or switches stacks. If you've built any custom software, it's going to need maintenance every time there's a model update, an API change, or a new platform feature that makes half of it redundant. And right now, at least one of those things is happening every month. Like the frequency of these model releases and new updates is is ramping up like ridiculously right now. And when it's finally time to relieve the waiting room, your custom build won't get

[7:55] upgraded. It'll just get replaced entirely. You'll have paid tens of thousands of dollars for a prototype the labs will charge like $20 a seat for. And I've had to make this call myself recently. For the AI makeovers I've done recently, I seriously considered building our own custom platform and putting every client on it. It would have been awesome. It would have been really, really, really cool. It would centralize all their data, allowed us to make a custom AI operating system or or our own custom AI platform, but we would have spent a year of our lives building something that the labs are about to ship for a fraction of the price and

[8:22] ultimately is going to be far more robust and reliable and require a lot less effort on my end to maintain and I can deliver the value for my client much faster. Now, the second thing is that you're about to get sold harder than any software company has ever sold to you because this market is going to behave very differently to any kind of other software market you've probably bought before or been exposed to. So, here's what no one's really talking about with these kinds of AI installs and moved onto these AI platforms. Because if you've ever moved a business onto HubSpot or Salesforce, then you know the

[8:50] drill, which is the data cleanup and the field mapping and the 3 months of we're just going to use the old one for now. You've got weeks and weeks of work and thousands of dollars in consulting fees just to set up like one department's tool, your CRM or your Salesforce or HubSpot. And the thing with an AI platform is that it's not just one department's tool. In order to get anything out of it, it needs all of your data, all of your context and all of your work in one place that the agents can see. So, it's kind of all or nothing by design. So, half of a business on the

[9:17] platform is a business with two different sources of truth and an agent that's going to be confidently wrong half the time. So the install of getting a company set up on one of these AI platforms is the real work. Mapping the processes, connecting the data, writing all the context docs, building the agents, setting the permissions, and training the team on how to use this new thing. But once a business is installed, it's likely not going to be going

[9:38] anywhere. You've got sunk cost now. You've got training costs. You've got switching costs. So whoever installs you onto their platform will likely keep you for years on a deal worth thousands to hundreds of thousands of dollars per month, depending on the size of your business. So every lab and every funded startup can do that maths. So the next two years are going to be free onboarding, free engineers, discounted first years, and a lot of very convincing demos. And that's what you're waiting in. It's not a lull. It's it's going to really be a land grab for these startups and labs over the next 12 to 24

[10:09] months. So for the business owners, when you come into this, don't just dip a toe. When you move and when you go onto a platform, you need to go all in. Like a half migration onto a platform is the worst of both. You're getting the maintenance of a custom build and the lockin of a platform and you're basically getting the benefits of neither. But for the AI agency guys, the value is moving off building the platform for your client and building all of this custom code to installing them on a platform really well. And that's what you've been doing by hand, badly paid and on a stack that keeps

[10:35] changing under you for the past 3 and 1/2 years. The agencies that will win are going to be the ones who pick a platform or a set of platforms, go very deep on it. And all of this leaves a big question, which is what should you go all in on right now? Especially when every platform on the market is selling you essentially an engine with a seat bolted onto it and calling it a car. So based on my own testing, here's where the market is actually sitting right now. Scored out of seven based on how many parts of the car or how many pieces of the AI platform each of these labs

[11:04] and startups are delivering at a high level. Now, I've covered most of the labs. I've put in a couple of the startups I've noticed that are doing these kinds of AI platforms or moving towards it. So, Anthropics Co-work is probably the most polished workspace of the lot. If you're solo or a very small team, you're going to love it. But there's no shared context. There's no data layer. The agents are essentially just scheduled tasks or routines, not employees that your team can actually use. Based on my rubric and the testing I've done, it's about a three out of seven. In Google's case, they've already got your data. The admin controls on

[11:34] their side are solid, and it's only seven bucks a seat. But then you find the automations that you can build on there only runs on spreadsheets and nobody else can see the agents you build. So Google's at a four out of seven. Open AAI is honestly the best of the labs right now. You have shared agents company knowledge that they can query. You have credits at plain API price which is great. There's no markup on it. So OpenAI scores a five out of seven on my rubric. Microsoft is on paper the most complete but it's not really built for small businesses. And I take it you're either a small business

[12:03] or you're someone selling to small businesses. So that's why I'm making sort of that disclaimer. It's really a mid-market and enterprise play. And on Rubric, it's scoring a 5 1/2 out of 7. A newcomer on the scene is XAI's Grockbot, and it's honestly a pretty exciting one. It's dead simple to onboard, but nowhere near an all-in-one AI platform yet. It

[12:19] only scores a 1 and a half out of seven. But I wouldn't doubt Elon though. So I think they're showing a lot of promise. Definitely angling the right way for small sort of one person or small teams once they add in this multiplayer layer. Now, for the startups, who are actually ahead of the big labs right now by a reasonable amount. Firstly, you have Kylon, which I've been making a bunch of videos about lately as I've been testing it in my makeovers. Overall, they're the most complete architecture that I've tested. It's kind of like Slack plus NA10 plus Lovable. In terms of a small business platform or a soloreneur, I

[12:48] think it really has a broad range of of tools and capabilities that I haven't seen any other platform match this kind of all-in-one aspect to it. It's definitely not suitable for hundreds and hundreds of employees yet, and I haven't seen anyone use it for that. But in terms of a small business platform, it really has pretty much everything you need at this point. On my rubric, it's a 6 and 1 half out of seven. Next, from startups, we have Victor AI, which is as close as I've seen to a real multiplayer platform. It's all based in Slack, though, so a bit different to Kylon. And when you score it, it ranks 5 and 1 half

[13:18] out of seven, so pretty good. Then we have Lindy, which has been around for a while, but it's kind of the more boring like compliant one. It's got slot two and and a BAA, which matters if you're a clinic. So, they scored a five out of seven. Then we have Open Claw, which is the harness itself. It's free. It's single player. It's sitting on a permission that Anthropic has revoked twice this year, which scores it at a 2 and 1/2 out of 7. So, the lay of the land is that the labs are on their way, but the startups have actually nearly finished building a complete car, at

[13:45] least for small businesses. So my estimate is that it's 6 to 12 months before a startup is genuinely ready for most small businesses kind of like the middle of the bell curve to try this stuff out and maybe 12 to 18 months for the labs to catch up and give us something suitable. We are so far ahead in terms of the user interface and how AI is being adopted in businesses and that's the startups because the labs are so focused on the model race and

[14:06] research and training and all of that. They don't have the time or the resources to devote. These startups are putting 100% of their effort into the user experience, into onboarding, into how businesses actually use it, the missions and things like that, automatic skill generation. Whereas the labs are probably putting 5 to 10% of their effort into these kinds of platforms where the rest of their resources and time is going into research, new models, and being at the frontier. So when these AI platforms do arrive, they'll come from one of two directions. And neither bet is safe. that we have a generalized platform with billions of dollars behind

[14:36] it or a specialized platform with millions and there's an optimal one for you and your business and it comes down to what you can't afford to lose. So firstly we'll do the lab bet. So the case for the labs is that they have hundreds of billions of dollars and the best engineers on the planet right now. Compared to startups they are closer to the raw cost of intelligence. So you're not there's no middleman markup between you and their models. And the case against the labs is that you're likely

[14:58] going to be locked onto their models. For example, on the OpenAI platform, you'll be using OpenAI models. And if they fall behind in models, then you're going to be stuck with them. And switching to another platform means reinstalling your whole business on another platform. That switching cost will be a huge time and money drain if your platform does fall behind on model development. And in the case of these big labs or companies like Google, they're building a platform for millions of different businesses and it's

[15:19] generalized and it might serve no one. It's too general. On the other hand, we have the startup bit and it's for the more model agnostic play where you can swap to the best models as the market moves, which is a huge deal. You're not stuck on one model provider and you can always be using the best intelligence across video generation, image generation, coding models, reasoning models, and pick and mix the best models. Another interesting piece about startups is that you're likely going to have AI platform that is say for e-commerce businesses or marketing agencies that comes out of the box with a lot of the things your business model

[15:47] is going to need rather than this general platform can be a massive advantage. And another key factor in the startup route is that it's going to be something that you could influence and shape their road map. You can give them feedback and they'll adjust it to you. And that's not something that you're going to get out of the labs. Now, the case against is longevity. Like, are these guys even going to exist in 3 years or are they going to go bankrupt right after you onboard your entire business. Another thing is that the price you sign at isn't necessarily the price it might be in year three. It

[16:10] could be five times that. This is because they might use pricing to get you in and then ultimately they need to increase your prices to actually have some kind of profit margin. And most of these startups don't have the kind of compliance paperwork a serious buyer or larger business or certain industries are going to require. But one thing that's for sure on both sides is that neither camp is going to make leaving their platform easy. So you can assume that whoever installs you first is who you're going to be staying with. So picking carefully is key. Which means that you need some kind of scoreboard of

[16:34] your own so that you are ready before anyone comes to try to sell to you. Carefully choosing and timing your action in this waiting room is essential. So first you need to ask yourself three questions. Number one, how big are you now? and how big are you going to be in in 2 years. Most platforms are going to be built for some solo operator or a small 10 20 30 40 50 person company. Almost none at the moment are built for both. So these platforms tend to be very specific like Grockbot for example very good for a very small company whereas the Microsoft platform is going to be better for

[17:00] mid-market and enterprise. Number two, what can you not afford to lose? Is it the platform disappearing or being locked into one labs model? Number three, what's the number you'd never go above in terms of software spend for this category per month? You need to write that down because the first year's price is going to be a land grab. Then when the platforms come knocking, you can ask them four questions. Firstly, will you actually install this thing for me or are you just going to hand me a login and then a help center to do it myself? Secondly, what does a real month cost? Not seat price, but a company of

[17:27] my size, how much are they paying in usage per month? They should have numbers and if they don't, then you need to run. Thirdly, what models do you offer and can I swap them out? Are you actually model agnostic? Do you support the full range of models so that I can change when I need to when certain companies take the lead? And finally, will you exist in 3 years? If you don't, what do I get to walk away with? Like, is everything just going to disappear in a cloud of smoke, or do I have something I can take to another platform? So, when you get into conversations with the

[17:50] sales reps of these platforms, and they answer those questions for your specific needs, and they are tailored to a business of your size, they're at a price point that you've you're comfortable with, and they're going to offer a forward deployed engineer to help you get set up, then that's a really, really, really good option, and you should consider that and going all in onto that platform. But until then, business owners or agencies, the answer is the same. Don't try to build ahead of the labs, guys. Your job isn't to build the features that these labs are going

[18:14] to ship in the next six to 12 months. It's to watch and wait in this waiting room patiently and be ready for the right opportunity for your business at the right time. And having that checklist is going to allow you to know what's really a good fit for you and your company. Timing your entry and picking the right platform and being aware of what the state of the market is. It's all you really need to do. I would not be diving into things unless you are either comfortable with that switching cost or comfortable with the fact that what you're building custom or the platform you're getting onto may

[18:40] need to be scrapped in 2 3 months when a major lab releases a big update and you kind of need to go through that process again. Better to sit and wait until the right one comes along. Apply the checklist and then go all in and make sure that the platform gives you access to FTEES and people who can set the stuff up for you so that you don't need to be doing all the work. They want your business. Nobody selling AI is going to tell you it's okay to stop except me in this case. But right now there are thousands of very very very smart people

[19:02] with billions of dollars staying up at night trying to solve like the carburetor of the AI native business. Just like took a lot of experimentation to figure out the components of a car. We're trying to figure out the components of an AI native multiplayer AI workspace for businesses and we haven't figured that out yet. Now since I filmed this, it's already played out exactly like I predicted. Just way faster than I expected. Like remember I said OpenAI dev day was starting in about an hour? Well, it happened and they shipped a bunch of new releases into the chat GPT business suite. Then Google dropped their new Gemini

[19:31] platform. I did live streams on both. That's two of the biggest labs bolting new pieces onto the car in a matter of weeks. And that 12 to 18 months I gave the labs, honestly, I think it's going to be a lot shorter. So, my message is to kick back a bit, hang out in the waiting room with some patience and confidence that this is going to be resolved soon, and then pick your shot when the time comes. If you want to stay in the loop on all of this and the platforms that I'm personally testing in my own business, in my client makeovers when I do them, make sure that you are

[19:56] subscribed to the channel, follow me on all my other socials which are in the description below, and also check out the video up here, which is a walkthrough on one of these platforms that you can have a look at right now. Guys, that's all for the video. Thank you so much for watching. Be patient in the waiting room and I will see you in the next

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt