WEBVTT
00:00:02.000 --> 00:00:04.160
How are you using AI right now?
00:00:04.639 --> 00:00:07.200
If you're like most people, it's intermittently.
00:00:07.599 --> 00:00:12.240
Maybe to start a project or in the middle to end or maybe the whole thing.
00:00:12.480 --> 00:00:19.280
But chances are you go out of your way and you make a choice to go and use AI.
00:00:20.640 --> 00:00:21.839
It's not like that anymore.
00:00:22.000 --> 00:00:25.839
I like to talk about AI like the internet.
00:00:26.800 --> 00:00:29.519
You can't work without the internet, right?
00:00:29.679 --> 00:00:38.320
And I'm not saying going to different websites, your email, your CRM, your project management system, your ERP, everything is connected to the internet.
00:00:38.640 --> 00:00:42.960
And it's not like we now make a choice to use the internet.
00:00:43.200 --> 00:00:44.799
It's just there and we need it.
00:00:45.359 --> 00:00:47.439
That's the parallel I'm trying to draw here.
00:00:47.600 --> 00:00:52.640
I think most people for the last couple of years have had to make a choice to use AI.
00:00:53.200 --> 00:00:56.079
In 2026 and beyond, it's not like that anymore.
00:00:56.240 --> 00:01:03.359
I think AI is, for better or worse, going to infiltrate every single corner of where and how we work.
00:01:03.840 --> 00:01:21.280
And so on today's show, the Start Here series, volume three, we're going to be talking about how I think you should be using AI as an operating system, not as an afterthought or not intermittently, but the foundation of how you and your team work.
00:01:21.519 --> 00:01:21.920
All right.
00:01:30.000 --> 00:01:40.239
So if you are new, our Start Here series is everyday AI's essential beginner and advanced AI overviews to launch your year strong.
00:01:40.480 --> 00:01:45.680
So if you haven't heard, well, you should probably go to starthearseries.com.
00:01:45.840 --> 00:01:54.079
That is going to get you a free entry into our inner circle community and all of the Start Here Series podcasts.
00:01:54.239 --> 00:01:56.480
You don't got to go fishing for them on our website.
00:01:56.640 --> 00:02:05.680
They're all going to be right there in that space, as well as you get access to our free Prime Prompt Polish Chat GPT prompt engineering course.
00:02:05.840 --> 00:02:07.359
So make sure you go check that out.
00:02:07.519 --> 00:02:10.400
And if you're very, very new here, well, this is everyday AI.
00:02:10.479 --> 00:02:13.840
And we've done now almost 700 podcast episodes.
00:02:13.919 --> 00:02:16.240
And one of the most common questions I get is where do I start?
00:02:16.319 --> 00:02:19.360
And that's why we're starting the start here series.
00:02:20.080 --> 00:02:27.039
So uh if you missed our last uh start here um episode, it was episode 693.
00:02:27.120 --> 00:02:30.240
So you can go click back a couple of times on the podcast player there.
00:02:30.400 --> 00:02:32.479
But we talked about AI without the jargon.
00:02:32.560 --> 00:02:38.319
So we kicked this whole thing off by talking about what is generative AI, what are large language models, volume two.
00:02:38.400 --> 00:02:44.000
Uh we dove into just the language and the lingo, because if you don't understand that, you can't compete.
00:02:44.080 --> 00:02:53.120
Uh and now today we're going to be talking about how I think you should be looking at AI as an operating system, not as a tool that you selectively go to use.
00:02:54.319 --> 00:02:56.479
So why?
00:02:56.879 --> 00:03:02.879
Well, I think AI is no longer an app you visit, it's becoming how you work.
00:03:03.039 --> 00:03:16.879
You know, if we think back to the early days of Chat GPT, which I think is most people's introduction to generative AI, but even to last year, I think still most people using AI went and sought it out.
00:03:17.120 --> 00:03:19.919
It's not going to be like that for very much longer.
00:03:20.159 --> 00:03:20.639
Why?
00:03:20.879 --> 00:03:26.800
Because wherever you work, AI has already infiltrated it and you can't really get around it.
00:03:26.879 --> 00:03:31.120
Just like you can't get around doing knowledge work without using the internet.
00:03:31.599 --> 00:03:32.639
Think about it.
00:03:33.120 --> 00:03:38.479
If you're in Microsoft Excel, Microsoft Copilot's there.
00:03:38.719 --> 00:03:43.280
If you're opening up a blank uh Google Doc, Gemini's there.
00:03:43.439 --> 00:03:46.400
If you're in Salesforce, Agent Force is there.
00:03:46.560 --> 00:03:54.319
It's very hard now, regardless of where you work, for there not to be AI already embedded there.
00:03:54.719 --> 00:03:58.000
Yet you still have to kind of go out of your way to use it.
00:03:58.240 --> 00:04:11.280
But we're going to talk about how I think that's not only changing, but how you as business leaders need to shift your mindset and understand that what I think is gonna happen, it's definitely gonna happen.
00:04:11.360 --> 00:04:15.840
Uh, front-end large language models are going to be the future interface of work.
00:04:15.919 --> 00:04:18.560
And yeah, we got we got receipts, y'all.
00:04:18.720 --> 00:04:19.680
Don't worry.
00:04:20.000 --> 00:04:30.480
So, uh, in today's show, we're gonna talk about how the internet is becoming a background service for your AI assistants and probably nothing more, and how that changes knowledge work.
00:04:30.639 --> 00:04:53.519
We're gonna be talking about the protocols and platforms that are making this shift to an AI operating system happen, such as ChatGPT apps, uh, the new Claude Co-work, uh, MCP from uh Anthropic, their model uh contacts protocol, and that how that helps agents kind of not need us for the things that we've maybe been doing a lot over the last year or two.
00:04:53.680 --> 00:04:59.519
And then last, but how business leaders need to change.
00:05:00.079 --> 00:05:01.519
That's the reality.
00:05:01.920 --> 00:05:07.519
A lot of people think of AI as a tech implementation, it's not, it's people management.
00:05:07.680 --> 00:05:13.519
And you need to know the adoption stats, the risks, and how to choose your AI operating system.
00:05:13.600 --> 00:05:15.439
So we're gonna be diving into all that.
00:05:15.519 --> 00:05:17.040
But let's start at the top.
00:05:17.360 --> 00:05:20.160
Why do you need an AI operating system?
00:05:20.319 --> 00:05:22.000
And what the heck is it?
00:05:22.240 --> 00:05:24.399
Um, it's not an official term.
00:05:24.560 --> 00:05:31.120
I don't know if I made it up, but I've been calling, I've been talking about AI operating systems now for more than two years.
00:05:31.279 --> 00:05:39.040
Uh, but the reality is right now, there's almost 2 billion people that use AI almost weekly.
00:05:39.120 --> 00:05:46.879
And 90% of knowledge workers who use AI report higher productivity than those who don't, obviously, right?
00:05:47.199 --> 00:05:54.560
Um, and not just how we use AI, but AI's capabilities have obviously greatly changed.
00:05:54.720 --> 00:06:00.959
I think, especially um in the year 2025, more so than the two years prior.
00:06:01.360 --> 00:06:08.160
And the main reason for that is large language models mainly did there's two big advancements.
00:06:08.399 --> 00:06:11.839
Number one, bringing in your company's dynamic data.
00:06:11.920 --> 00:06:13.279
And that's big, right?
00:06:13.360 --> 00:06:17.759
That's what cuts down on hallucinations, but that's also what increases collaboration.
00:06:17.920 --> 00:06:20.079
And that is the second one, right?
00:06:20.160 --> 00:06:26.480
Uh large language models and front-end AI chatbots are no longer, oh, I'm just chatting with a helpful assistant.
00:06:26.639 --> 00:06:32.079
No, now your business's data is there and you can collaborate with team members, right?
00:06:32.319 --> 00:06:39.360
Work in the same chat, like you're sharing a uh Google Doc and you're all collaborating there live in in real time.
00:06:39.600 --> 00:06:46.399
And that's because right now integrated AI systems are a baseline requirement, right?
00:06:46.800 --> 00:06:55.199
Um, I like to think most companies at some point in the uh in the 90s made a decision, right?
00:06:55.279 --> 00:06:57.279
And early 2000s as well.
00:06:57.519 --> 00:07:02.879
They said, we're either a Windows organization, we're a Mac organization, we're a Linux organization.
00:07:03.120 --> 00:07:16.720
You chose your operating system, and then you kind of your front-end processes that happened before the computer, during the computer, and obviously what happened after, were largely based around what operating system that you chose.
00:07:17.040 --> 00:07:23.519
And I think the same can be true for the choices that we have now ahead as everything becomes AI native.
00:07:23.600 --> 00:07:32.399
Uh, that's because a single person with good AI skills, using the right agents, the right co-pilot, can accomplish the work of a larger team.
00:07:32.639 --> 00:07:40.879
So naturally, we have an extremely strong orchestration layer that's already in place.
00:07:42.079 --> 00:07:44.240
And it's just begging, right?
00:07:44.399 --> 00:07:52.240
The smartest organizations I talk to, I've, you know, lucky enough to get to talk to hundreds of very smart people in AI every single year.
00:07:52.480 --> 00:07:55.839
The smartest organizations have already done this.
00:07:56.079 --> 00:08:09.839
They've already brought their entire team, their entire processes into, you know, ChatGPT uh enterprise or you know, Google Gemini Enterprise, uh, Claude Enterprise, obviously Microsoft 365 Copilot.
00:08:10.079 --> 00:08:12.160
And they've changed their processes.
00:08:12.319 --> 00:08:20.399
Not just, oh, we're gonna use uh, you know, we have a 10-step process, we're gonna use AI at step two, four, and seven.
00:08:20.800 --> 00:08:27.040
No, they said that 10-step process process is now a three-step process, right?
00:08:27.279 --> 00:08:33.200
Uh human orchestrates something, agents go and do something, human checks it on the back end.
00:08:33.360 --> 00:08:39.360
And yeah, we went from 10 steps to three steps, and from 10 hours to 30 minutes.
00:08:39.679 --> 00:08:44.240
Those are the moves that the business leaders right now are taking.
00:08:44.559 --> 00:08:47.120
And that's why you need an AI operating system.
00:08:47.360 --> 00:08:49.919
All right, and we're not going to dive into this too much, right?
00:08:50.080 --> 00:08:50.960
Oh, what do you choose?
00:08:51.039 --> 00:08:51.600
What do you choose?
00:08:51.679 --> 00:08:51.919
Right.
00:08:52.000 --> 00:08:58.639
We'll maybe do another uh show later in our start here series on this, but I think you essentially have four choices.
00:08:58.879 --> 00:09:03.759
And well, if you're a Microsoft co-pilot organization, you might be able to choose two.
00:09:04.080 --> 00:09:05.200
So let's start there.
00:09:05.360 --> 00:09:08.799
Microsoft 365 Copilot, hats off to them.
00:09:09.120 --> 00:09:14.799
They started technically the enterprise AI craze, right?
00:09:14.960 --> 00:09:19.519
This is back when large language models essentially you had front-end chatbots, and that was it.
00:09:19.679 --> 00:09:28.159
Microsoft was the organization that brought enterprise AI into the mainstream by launching Microsoft Copilot.
00:09:28.559 --> 00:09:30.720
It is extremely robust.
00:09:31.200 --> 00:09:33.519
It is wildly flexible.
00:09:33.840 --> 00:09:38.320
Yet the learning curve is bonkers, right?
00:09:38.480 --> 00:09:45.600
That's one of the reasons why I think Microsoft Copilot hasn't just run away with it and they haven't been running away with it for years.
00:09:46.080 --> 00:09:51.120
Yes, there's a lot of um technical intricacies that I don't even understand.
00:09:51.200 --> 00:09:53.919
And the majority of people, unless you work in IT, don't, right?
00:09:54.000 --> 00:09:57.759
But there's mainly, I think, permissions, right?
00:09:57.919 --> 00:10:06.799
Anyone I talk to that's a Microsoft 365 copilot organization, they see all these great uh you know features that Copilot has.
00:10:07.039 --> 00:10:11.600
And the overwhelming majority of enterprises don't know how to get it set up, right?
00:10:11.759 --> 00:10:14.639
It's essentially you have gatekeepers and bottlenecks, right?
00:10:14.799 --> 00:10:18.240
Usually, you know, IT is looking for approval.
00:10:18.320 --> 00:10:19.759
People are like, why don't I have this?
00:10:19.919 --> 00:10:24.879
Someone else has access uh to this, you know, this in you know, copilot studio.
00:10:25.039 --> 00:10:25.519
I want that.
00:10:25.679 --> 00:10:26.639
No, you can't have that.
00:10:26.720 --> 00:10:27.600
How do we get that right?
00:10:27.759 --> 00:10:29.120
That's what a lot of it is.
00:10:29.360 --> 00:10:38.159
For the organizations that have robust uh, you know, Microsoft training already in place are the ones that have obviously done really well here.
00:10:38.320 --> 00:10:43.039
But uh Microsoft 365 Copilot obviously works at the OS level, right?
00:10:43.120 --> 00:10:47.279
Uh, it's the only uh AI that's baked into an operating system.
00:10:47.440 --> 00:10:51.600
Obviously, Apple is not there and they're very far from ever being there anytime soon.
00:10:52.000 --> 00:10:53.919
Next, Google Gemini.
00:10:54.320 --> 00:10:59.039
They won 2025 and they're innovating faster than anyone, uh, right.
00:10:59.440 --> 00:11:06.159
And similarly, like Microsoft 365 Copilot, you can get Google Gemini in a lot of different ways, right?
00:11:06.240 --> 00:11:07.120
People don't even know this.
00:11:07.200 --> 00:11:14.080
As an example, probably the Gemini that you use and have heard of is not even the best Gemini for Teams, right?
00:11:14.240 --> 00:11:23.840
Uh, about four months ago, Google actually came out with a business version of Gemini called Gemini Business and Gemini Enterprise.
00:11:23.919 --> 00:11:24.720
It is separate, right?
00:11:24.799 --> 00:11:35.200
So I have a paid uh Gemini account for my personal Gmail, I have a paid Gemini account um for my work email, and then I have a paid uh Gemini business account.
00:11:35.279 --> 00:11:37.120
All different work a little differently.
00:11:37.200 --> 00:11:40.480
Um, and that is also one of the pros and cons of Google Gemini as well.
00:11:40.559 --> 00:11:43.279
They all the versions work a little differently as well.
00:11:43.440 --> 00:11:50.240
There's pros and cons to them, and that's not even the tip of the iceberg of where you can use Google Gemini, right?
00:11:50.320 --> 00:11:54.799
Uh in AI Studio, uh, they have so many different, you know, vibe coding platforms.
00:11:54.960 --> 00:11:58.480
Uh, but it's uh Google Gemini is great, it's all over the place.
00:11:59.679 --> 00:12:02.399
Then you have Anthropic Claude.
00:12:02.559 --> 00:12:10.480
Uh so I will say this I am much more bullish on uh Claude than I was probably two months ago.
00:12:10.639 --> 00:12:17.600
And a lot of that has to do with, well, mainly two big uh pieces that I've seen from Anthropic.
00:12:17.919 --> 00:12:25.279
One is they were the first to really popularize file creation, which might sound crazy, right?
00:12:25.360 --> 00:12:29.519
Like, why haven't large language models just been doing that uh at the forefront?
00:12:29.759 --> 00:12:33.759
Um, well, Anthropic popularized it, right?
00:12:33.840 --> 00:12:44.799
They were the first one to get um Excel uh sheets right, right, which is kind of crazy, even before, in some instances, Microsoft 365 Copilot.
00:12:44.879 --> 00:12:46.320
The same thing with PowerPoint.
00:12:46.720 --> 00:12:48.720
Uh Claude really pushed the boundary there.
00:12:48.799 --> 00:12:56.559
And then they also uh, you know, just came out with their uh Claude co-work, a great um, a great tool, FYI.
00:12:56.639 --> 00:12:58.320
We're gonna have a show on that tomorrow.
00:12:58.480 --> 00:13:05.759
But I think that in 2024, in the early part of 2025, they were overly focused on coding and software engineering.
00:13:05.840 --> 00:13:13.759
Uh, and I do think that they've made a pivot over the last quarter or so uh to really go after the non-technical business leader.
00:13:13.919 --> 00:13:17.759
So uh they also have a great enterprise team product as well.
00:13:17.919 --> 00:13:20.480
And then last but not least, ChatGPT, right?
00:13:20.559 --> 00:13:23.039
In terms of users, no one's coming close.
00:13:23.200 --> 00:13:25.440
In terms of retention, no one's coming close.
00:13:25.600 --> 00:13:27.360
Stickiness, no one's coming close.
00:13:27.519 --> 00:13:30.960
Um, but it seems like they're really focused on consumers.
00:13:31.120 --> 00:13:34.080
Yes, they have a great uh Chat GPT business tier.
00:13:34.159 --> 00:13:42.000
They have a great, you know, Chat GPT enterprise tier where, you know, reportedly, I think 92% of Fortune 500 companies uh use one of their products.
00:13:42.320 --> 00:13:49.279
Uh, but I do think uh that right now, OpenAI is maybe a little more focused on the consumer side.
00:13:49.360 --> 00:13:52.559
Uh, you know, recently announced, you know, ads, right?
00:13:52.720 --> 00:13:55.039
A lot of things, ChatGPT health.
00:13:55.120 --> 00:13:56.720
Uh, but maybe it makes sense.
00:13:56.799 --> 00:14:05.840
Maybe that's their strategy to bring teams into the fold is to, you know, become so sticky in their personal life that it's hard for people to use anything else.
00:14:06.000 --> 00:14:17.600
So when you talk about bringing your team, and like I said, all four of those obviously have great options for, you know, individuals, small teams, large teams, uh, enterprise organizations, right?
00:14:17.759 --> 00:14:19.679
You have to choose which one is right for you.
00:14:19.759 --> 00:14:24.320
We're not gonna go too far into that, but in my opinion, those are your choices, right?
00:14:24.399 --> 00:14:25.759
Especially if you're in the US.
00:14:25.840 --> 00:14:28.960
Uh, yeah, you can look at open source solutions, right?
00:14:29.200 --> 00:14:30.879
Uh, I'm not gonna get into that.
00:14:31.039 --> 00:14:39.039
I think you have to choose, just like you know, you choose uh Windows, you choose Mac, you choose Linux.
00:14:41.519 --> 00:14:44.720
And this is actually a very telling sign.
00:14:44.799 --> 00:15:01.919
Uh, so we covered this a couple of months ago when this was first reported, but there was a report that Microsoft's CEO Sadia Nadella uh essentially warned his employees that Microsoft and Windows, their office suite, could become obsolete.
00:15:02.320 --> 00:15:07.440
Yeah, he said that they had to reboot and focus on AI or the company.
00:15:08.000 --> 00:15:10.399
The company could become obsolete.
00:15:12.559 --> 00:15:15.759
Now, if that sounds kind of like a wild statement, well, it kind of is.
00:15:15.840 --> 00:15:27.440
I mean, Microsoft has, you know, for the past, you know, three to four years traditionally been uh top one, two, or three company in the world when it comes to market cap.
00:15:28.159 --> 00:15:39.840
Yet, Nadella warned that without a fundamental overhaul and a focus on AI, that Microsoft's most successful businesses could become irrelevant as AI agents replace traditional software.
00:15:40.080 --> 00:15:40.320
Right?
00:15:40.480 --> 00:15:42.720
Yeah, saying no one's gonna use Office anymore.
00:15:42.799 --> 00:15:45.600
No one's gonna use, you know, all of these different tools.
00:15:45.759 --> 00:15:51.840
And that's why Microsoft reportedly is transitioning from a software factory to an intelligence engine, right?
00:15:52.080 --> 00:16:09.279
Sadia Nadella, one of, I think, the smartest tech people probably to ever live, is saying that, hey, even we as the largest company in the world, that we power the enterprise when it comes to an operating system, we're not gonna make it.
00:16:09.600 --> 00:16:13.120
Because the business world is transitioning away from that.
00:16:13.279 --> 00:16:15.519
And it's all about your intelligence engine.
00:16:15.600 --> 00:16:20.639
It is about how you put your knowledge to work using AI, right?
00:16:20.799 --> 00:16:30.720
And he said that Windows uh becoming an AI first operating system, uh where the keyboard and the task bar become secondary.
00:16:30.960 --> 00:16:36.879
So his goal was for every Office application to function as a development environment for AI agents, right?
00:16:37.919 --> 00:16:38.799
Think about that.
00:16:40.080 --> 00:16:52.480
That's right, like billions of people use Microsoft in his goal that all that's going to be is a base for agents to work.
00:16:53.600 --> 00:17:02.480
That's the shift that I think a lot of people, uh, right, you should probably be listening and paying attention to what he's saying, uh, because that's the truth.
00:17:04.400 --> 00:17:09.920
And it's also because, like I said, large language models are much more than that, right?
00:17:10.319 --> 00:17:12.960
They are now agentic.
00:17:13.839 --> 00:17:15.279
They're now agents, right?
00:17:15.359 --> 00:17:16.000
That's the thing.
00:17:16.160 --> 00:17:19.119
You go into Chat GPT, people don't know this.
00:17:19.200 --> 00:17:28.400
You can schedule an agent to go, you can give it access to you know your uh connected and synced data.
00:17:28.559 --> 00:17:36.400
Uh, you can log into certain websites and you can say, hey, every day at you know, 7:30, go do these five tasks for me, right?
00:17:37.119 --> 00:17:47.680
People are still looking at uh large language models as you know, fun little toys to help you write better emails, and they're not that.
00:17:47.839 --> 00:17:56.000
Uh, the orchestration, the tools, uh, you know, with the browsers, code executors, app connectors, uh, bringing in your company's dynamic data.
00:17:56.079 --> 00:17:57.279
Uh it's it's huge.
00:17:57.440 --> 00:18:07.759
And now as memory and context improves, so done so does an agent's ability to orchestrate, to plan and execute taps, right?
00:18:07.839 --> 00:18:16.799
As the memory, the context window, uh the personalization and the ability to properly use data uh with the harnessing of these models.
00:18:17.039 --> 00:18:19.039
It's it's it's nuts, right?
00:18:19.279 --> 00:18:25.359
And and this is the cognitive layer that Saudi and Adela, the intelligence layer that he's talking about.
00:18:25.599 --> 00:18:32.160
As the models get exponentially smarter and they get more and more access to our data and they get better and better at using tools.
00:18:32.400 --> 00:18:33.440
Hello, people.
00:18:34.000 --> 00:18:35.680
This is what humans do, right?
00:18:35.759 --> 00:18:39.519
Uh, I've talked one or two times uh through this start here series.
00:18:39.599 --> 00:18:46.559
Um, and I'm gonna talk about it probably a little bit tomorrow when we talk about Claude Cowork, very much uh related uh to this show.
00:18:47.039 --> 00:18:58.559
I think a lot of the work that humans in companies that are ahead in AI is acting as duct tape, right?
00:18:58.880 --> 00:19:02.240
We're duct taping and connecting different agentic flows.
00:19:02.319 --> 00:19:04.319
That's a lot of what I do all day, right?
00:19:04.720 --> 00:19:09.839
I have a bunch of agents going to do things, I orchestrate them, you know, I taste make.
00:19:09.920 --> 00:19:11.440
I'm like, good job, agent one.
00:19:11.519 --> 00:19:14.240
Uh hey, agent two, that's not what we wanted, right?
00:19:14.400 --> 00:19:15.519
I'm duct taping.
00:19:16.000 --> 00:19:19.680
I mean, I think even though that's weird, I think that's where we're headed.
00:19:20.000 --> 00:19:28.640
Because large language models are not chatbots, they're literally that human intelligence layer that Sonia Nadella was talking about.
00:19:30.720 --> 00:19:33.279
So let's get into the web, right?
00:19:33.519 --> 00:19:36.960
And how we can't work without the web.
00:19:38.240 --> 00:19:43.440
Right now, if you're listening to this, if you're watching this, it's because of the web, right?
00:19:43.519 --> 00:19:44.240
It's because of the internet.
00:19:44.720 --> 00:19:45.680
Everything we do is the internet.
00:19:45.839 --> 00:19:51.519
If you're a knowledge worker, it is because of the power of the internet.
00:19:52.319 --> 00:19:52.640
Okay?
00:19:53.119 --> 00:20:01.839
I like to tell people you need to replace the word internet with AI, because the two are gonna become one.
00:20:02.319 --> 00:20:02.559
Right?
00:20:02.640 --> 00:20:06.240
You think of anything, think of a Google search, right?
00:20:06.400 --> 00:20:09.759
Um, eventually it's gonna be just an AI mode search, right?
00:20:09.920 --> 00:20:14.240
Um they've already said Google has said that that is going to be the end goal.
00:20:14.319 --> 00:20:16.240
Everything's going to go to AI mode.
00:20:16.720 --> 00:20:19.839
So are you gonna browse the web in the future?
00:20:20.000 --> 00:20:20.799
I'd say probably not.
00:20:20.960 --> 00:20:23.759
Browsing the web is an absolutely terrible experience.
00:20:24.400 --> 00:20:25.119
The same thing.
00:20:25.279 --> 00:20:30.960
I think for the most part, the web will be made for agents, right?
00:20:31.200 --> 00:20:39.279
And in the same way that we have these protocols that allow agents to talk to each other, they're gonna be doing the web browsing as well.
00:20:39.759 --> 00:20:40.319
Right.
00:20:40.480 --> 00:20:42.319
Um, I mean, look where we are now.
00:20:42.400 --> 00:20:47.200
Look at some of the I think the best uh advancements of 2025 were agenc browsers.
00:20:47.279 --> 00:20:51.440
I think uh if you had to give categories a grade, right?
00:20:51.599 --> 00:20:54.720
I think niche agents got an A.
00:20:55.039 --> 00:20:58.400
I think general agents that we thought were gonna take over got like a C.
00:20:58.880 --> 00:21:01.839
And I think that agenc browsers got an A as well.
00:21:01.920 --> 00:21:02.400
Uh, right.
00:21:02.480 --> 00:21:09.279
I think agentic browsers performed much better or at higher uh aptitude than agents did.
00:21:09.519 --> 00:21:18.880
So working in uh autonomous capabilities into your browser and then pairing it with memory, which I think OpenAI's Atlas does really, really well, right?
00:21:19.200 --> 00:21:28.720
That gives your browser the ability to number one just perform tasks for you, uh, right, but to bring your context and your history over as well.
00:21:29.119 --> 00:21:29.440
Right.
00:21:29.599 --> 00:21:33.519
Think about, I don't know, an example, filling forms.
00:21:34.319 --> 00:21:36.079
Why should we do that anymore?
00:21:36.400 --> 00:21:37.519
It's stupid.
00:21:38.160 --> 00:21:38.720
Right?
00:21:38.960 --> 00:21:48.480
If if if an agentic browser has all of that in its memory, has all that information and can navigate these, we should never be filling forms, right?
00:21:48.559 --> 00:21:53.440
That's why I think so much of what's happening on the web is just going to become for agents.
00:21:53.599 --> 00:21:59.599
And I don't think that we're gonna be browsing the web much longer, which does change the future of work because.
00:22:00.640 --> 00:22:01.599
That's what we do.
00:22:01.759 --> 00:22:02.960
That's what we do as humans.
00:22:03.119 --> 00:22:04.880
We go do work on the web.
00:22:06.319 --> 00:22:08.079
Asians are better.
00:22:08.240 --> 00:22:11.599
If you know what you're doing, if you're using the right model, right?
00:22:11.920 --> 00:22:16.960
The the the frontier state of the art, they're smarter than us all, right?
00:22:17.039 --> 00:22:27.200
You look at offline IQ tests, they're nearly at genius level, but smarter than probably 99.99% of anyone listening, a thousand times smarter than me.
00:22:27.680 --> 00:22:34.559
Um so what used to take many searches, clicks, and tabs is now just delivered as one concise answer.
00:22:37.519 --> 00:22:46.240
So let's get back a little bit to how using AI as an operating system can solve that.
00:22:46.400 --> 00:22:51.119
Because one of the things that I think we waste so much time for is context switching, right?
00:22:51.599 --> 00:22:54.400
Every extra login, every copy and paste.
00:22:54.559 --> 00:23:01.119
Oh my gosh, this formatting's wrong from Copilot, uh, Chat GPT, I, you know, I need this in Markdown, it's not working, right?
00:23:01.359 --> 00:23:08.160
We spend so much time reformatting, copying and pasting, you know, switching uh, you know, tabs.
00:23:08.240 --> 00:23:12.160
Uh you know, analysts call this the app hop tax, right?
00:23:12.640 --> 00:23:18.079
Going from all these different apps, all these different softwares, you lose so much time, right?
00:23:18.400 --> 00:23:29.519
And to be honest, aside from summarizing information and synthesizing things and personalizing information, that's what knowledge workers do, right?
00:23:30.000 --> 00:23:37.839
We use our brain to go and summarize, synthesize, and rewrite information.
00:23:38.079 --> 00:23:40.400
And then we context switch, right?
00:23:40.559 --> 00:23:42.640
Okay, I'm gonna read this industry report, okay?
00:23:42.960 --> 00:23:46.880
I'm gonna personalize that context, take it by to my team, okay.
00:23:47.039 --> 00:23:50.720
Uh, whatever the team says, then I'm gonna go put it in a blog post, okay?
00:23:50.960 --> 00:23:52.640
Then I'm gonna go put it on the web, okay.
00:23:52.720 --> 00:23:56.319
Our customers are gonna read it, I'm gonna sign up calls right with the customers, okay, right?
00:23:56.480 --> 00:24:01.839
All you're doing is you're just slightly switching that context over and over and over and over again.
00:24:03.920 --> 00:24:05.359
It's gotta be agentic, right?
00:24:05.440 --> 00:24:15.599
Right now, that's friction because agents can handle most of that right now, right?
00:24:16.000 --> 00:24:26.880
Um in a in an agent native flow or an LLM native flow, an agent can execute everything inside of a single thread.
00:24:28.559 --> 00:24:28.880
Right.
00:24:28.960 --> 00:24:34.559
Think of those multi-step tasks that you do every single day, right?
00:24:34.640 --> 00:24:38.240
I'll even give you a very quick personal example of how it's changed for me.
00:24:38.480 --> 00:24:38.880
Right?
00:24:39.119 --> 00:24:43.759
When I first started this show, there wasn't a lot of AI out there.
00:24:44.079 --> 00:24:44.400
Right?
00:24:44.559 --> 00:24:50.160
There wasn't, you know, there was no Google didn't have AI, Anthropic didn't exist, there was no Metallama, right?
00:24:50.240 --> 00:24:52.559
There was Chat GPT, it was a nice little chat bot.
00:24:52.640 --> 00:24:57.440
So, how this used to work, I'd have spent a lot more time reading and synthesizing information.
00:24:57.599 --> 00:25:02.960
The quality of my shows wasn't able to be as good because it took way too much time.
00:25:03.359 --> 00:25:04.160
Um, right.
00:25:04.319 --> 00:25:14.799
I wasn't able to go as deeply into topics, you know, it was a lot more interviews, a lot more topical things without being able to dive deep and I think have a much higher quality.
00:25:15.119 --> 00:25:27.119
Now I have multiple agents going out, looking at my context every single day, looking at my podcast stats every single day, looking at what people are clicking on in the newsletter every single day.
00:25:27.279 --> 00:25:32.480
And then based on all of that, right, based on my context, my point of view, right?
00:25:32.640 --> 00:25:42.160
What stats, you know, what podcasts are performing well, what people are clicking on, they go in real time, base all that, do reports autonomously.
00:25:42.240 --> 00:25:52.079
I have an agent that looks at multiple agents' reports and they get back to me and they say, Hey, based on what you cover, based on what your audience cares about, here's what you should be covering.
00:25:52.160 --> 00:25:53.599
And I look through that, right?
00:25:53.759 --> 00:25:59.440
As the orchestrator on the front end, the taste makes tastemaker on the back end, and I say, okay, great.
00:25:59.680 --> 00:26:00.000
Right?
00:26:00.079 --> 00:26:01.519
I still take the time to read it.
00:26:01.680 --> 00:26:08.960
I always, you know, go through and read, read the chain of thought, uh, you know, the observability and traceability 101.
00:26:09.599 --> 00:26:16.559
But it's no longer a step that myself and my team have to go do over and over and over.
00:26:17.920 --> 00:26:21.200
And that's the promise of an AI operating system.
00:26:24.000 --> 00:26:28.160
And it's a lot of the recent technology that's actually going to make that work, right?
00:26:28.240 --> 00:26:33.200
And I think one of the things you have to call out by name is Anthropic's model context protocol.
00:26:33.359 --> 00:26:36.559
And actually, a pretty uh important recent update to that.
00:26:36.640 --> 00:26:55.279
So in December, Anthropic actually donated the MCP, made it open source to the Agentic AI Foundation under the Linux Foundation, with also uh the that same foundation, the Agentic AI Foundation, also supported by OpenAI, Google, Microsoft, and AWS.
00:26:55.440 --> 00:27:02.720
So, right now, MCP, the easiest way to think of it, it's like uh, you know, people call it the USB C of AI, right?
00:27:02.799 --> 00:27:07.039
It's just everything plugs into it and you know, it just converts and it's good, right?
00:27:07.119 --> 00:27:14.799
It's it's it's like a US, you know, one of those dongle converters where you can, you know, put anything on one end, put it anything on the other end, and it just works and it's great, right?
00:27:14.960 --> 00:27:19.920
This is kind of like the same thing with uh, you know, how websites have APIs and can talk to each other.
00:27:20.079 --> 00:27:24.160
Websites built on all kinds of different frameworks, it doesn't matter, right?
00:27:24.240 --> 00:27:30.319
If if one thing's on, you know, React, one thing's on PHP, one thing's, I don't know, JavaScript, HTML, CSS, right?
00:27:30.559 --> 00:27:35.920
All those websites can talk to each other, even though they're built in different languages because they have this API.
00:27:36.079 --> 00:27:42.799
It's a common way to transmit languages, and that's what you have now with the model context protocol.
00:27:42.880 --> 00:27:49.200
It allows AI agents, AI models, AI tools to all talk to each other, right?
00:27:50.160 --> 00:27:52.160
Part of it's great, right?
00:27:52.240 --> 00:27:53.519
That's amazing, right?
00:27:53.680 --> 00:28:05.519
And it's also great that OpenAI, Google, and Microsoft, uh Anthropic's technically three biggest competitors, have all adopted it and have put it uh you know, put it to use in their products for their users.
00:28:06.319 --> 00:28:17.359
But at the same time, you also have to think wait, doesn't that just in theory take away a lot of the work of what a human would be doing?
00:28:17.680 --> 00:28:19.039
Absolutely, it does.
00:28:19.200 --> 00:28:21.680
I don't think that's a bad thing, right?
00:28:22.000 --> 00:28:24.720
But that's a lot of whether you know it or not, right?
00:28:24.880 --> 00:28:40.240
If you took time to really go deep in model context protocol and look what's capable as an example, you know, in uh Claude Co-work, what's exam uh like what's available with uh you know OpenAI's uh like agent SDK, right?
00:28:40.480 --> 00:28:50.079
And being able to use in in co-pilot studio model context protocol, you probably wouldn't know it, but probably just about any different website, any different app that you use.
00:28:50.160 --> 00:28:56.160
And normally right now, maybe a human is interacting with that data via AI and they're going back, right?
00:28:56.400 --> 00:29:05.680
Human one is is working with you know, uh tool one with AI, human two working with tool two with AI, human three working with tool three with AI.
00:29:06.400 --> 00:29:11.200
You don't need that because now you don't have to have this vertical integration, right?
00:29:11.359 --> 00:29:16.000
Humans working vertically, going just up and down with a single tool, right?
00:29:16.240 --> 00:29:25.119
Now those AI tools and softwares can all uh work horizontally across the plane as well because of the model context protocol.
00:29:25.200 --> 00:29:26.400
They can talk to each other.
00:29:26.559 --> 00:29:35.519
You don't need a human, you know, multiple humans uh to use different AI processes to talk to these 20 different uh you know pieces of enterprise software that you use.
00:29:35.680 --> 00:29:39.279
You can have an agent that changed them all together using the model context protocol.
00:29:39.440 --> 00:29:43.440
That's where you know expert-driven loops uh you know come into play.
00:29:43.839 --> 00:29:49.519
Uh and right now, some estimates suggest that 90% of organizations are already using MCP servers.
00:29:49.759 --> 00:29:50.160
Yeah.
00:29:50.640 --> 00:30:00.319
You're probably, especially if you're uh a claw user, you're probably using NP like M uh sorry, MCP model context protocol servers without even knowing it, uh right.
00:30:00.559 --> 00:30:04.319
Especially if you're using their desktop uh application, which is really good.
00:30:04.400 --> 00:30:07.680
That means you're using some of their built-in MCP servers.
00:30:10.000 --> 00:30:19.759
So you want more proof that AI operating system is the future of work, that you need to be bringing in all your processes, all right.
00:30:19.920 --> 00:30:23.519
I'm actually gonna uh start with Microsoft Copilot Studio.
00:30:23.599 --> 00:30:27.279
Uh, they just launched their computer use in public public preview.
00:30:27.519 --> 00:30:32.240
What that means, they want everyone to be using this, right?
00:30:32.319 --> 00:30:34.160
It gives them more data, right?
00:30:34.240 --> 00:30:36.799
Uh, but they have a great computer use agent.
00:30:36.960 --> 00:30:44.079
So all the things that us humans do, we've been training models for the last probably two years on a lot of this data.
00:30:44.640 --> 00:30:47.920
And they're gonna get even better at using computers, which is wait, what?
00:30:48.000 --> 00:30:49.680
That's what we do, right?
00:30:49.920 --> 00:30:57.599
The biggest companies in the world are creating uh specific models to go and do the work that we do.
00:30:58.000 --> 00:31:11.599
Navigate between, you know, applications on the desktop, files uh on your local machine, uh, you know, terminal on your local machine, uh, AI models, the web, right?
00:31:12.640 --> 00:31:13.279
That's there.
00:31:13.440 --> 00:31:18.640
That's what Cloud Cowork uh does that was just released uh last week, right?
00:31:19.039 --> 00:31:36.160
It's a non-technical version of Cloud Code that allows you to navigate and orchestrate files on your local uh machine, a virtual machine, virtual terminal, um as well as big websites, right?
00:31:36.319 --> 00:31:37.839
It brings that all together.
00:31:38.480 --> 00:31:46.319
But I will say, I think the biggest nod to the AI operating system is what ChatGPT is doing.
00:31:47.279 --> 00:31:58.400
I don't think a lot of people are talking about it because they're comparing it to the iOS App Store, but their ChatGPT apps are all the proof you need that the internet is going to die.
00:31:58.640 --> 00:32:05.519
And the line between AI and the internet is going to blend, right?
00:32:06.079 --> 00:32:09.839
So what tools do you use on an ongoing basis?
00:32:09.920 --> 00:32:11.440
I'm gonna I'm gonna name a couple here.
00:32:11.680 --> 00:32:12.160
Ready?
00:32:14.079 --> 00:32:27.759
Photoshop, Canva, Agent Force, Asana, Box, Clay, Dropbox, GitHub, Google Drive, Gmail, HubSpot, Monday.com, Lovable, Ramp, Pipe Drive, Replit, Notion, SharePoint, Teams, Slack, Stripe, Zoom.
00:32:28.079 --> 00:32:33.759
Chances are those are pretty much a lot of people's tech stacks, right?
00:32:33.839 --> 00:32:38.240
I probably covered at least half of your tech stack, maybe two-thirds, maybe that's all the tools you use.
00:32:38.319 --> 00:32:39.680
And that's not everything.
00:32:40.079 --> 00:32:40.960
What are those?
00:32:41.119 --> 00:32:45.599
Those are apps that are available today in ChatGPT.
00:32:46.400 --> 00:32:47.920
And let me tell you why.
00:32:48.160 --> 00:33:02.400
This is a huge nod and a big reason why you need to get your team working in an AI operating system and bringing all of your day-to-day processes into, as an example, something like ChatGPT business or ChatGPT Enterprise.
00:33:02.640 --> 00:33:05.039
Those apps, okay.
00:33:05.519 --> 00:33:15.759
Yes, you can technically use ChatGPT to um interact with a graphical interface, but it also brings in all of your data, right?
00:33:15.920 --> 00:33:31.599
So as an example, uh, all of your data if you use Dropbox or Box or Google Drive, uh right, all of your company's CRM data, if you're using HubSpot or PipeDrive for your CRM or Monday.com, there's also click up on there.
00:33:31.839 --> 00:33:38.000
All of your, you know, your Zoom meetings, your Slack messages, your Teams messages.
00:33:39.599 --> 00:33:50.319
All of that is now can be remembered and can be shared in the same context window by multiple people of your organization, by an agent.
00:33:52.000 --> 00:33:53.519
You get where we're going here?
00:33:53.759 --> 00:33:57.680
You don't have to go into those 10 different tools anymore, right?
00:33:58.160 --> 00:34:01.200
And context switch or 10 different people.
00:34:01.359 --> 00:34:06.960
You know, there you have all these broken lines of communication, whether it's yourself, oh my gosh, I forgot what I did yesterday.
00:34:07.039 --> 00:34:11.599
Yeah, I was using an AI tool inside, you know, Dropbox or inside Google Drive.
00:34:11.760 --> 00:34:13.840
And well, I forgot now, right?
00:34:14.159 --> 00:34:17.920
It's because when you're going on all of these separate runs, right?
00:34:18.079 --> 00:34:24.400
When you're using, you know, 10 different tabs, five different apps, you're losing all of the context.
00:34:24.800 --> 00:34:29.039
When you're bringing it inside of ChatGPT or something like that, it has it all.
00:34:29.519 --> 00:34:30.159
Right.
00:34:30.480 --> 00:34:41.360
And the the ability to triangulate all of the different uh opportunities within there, that is the real promise of an AI operating system.
00:34:41.440 --> 00:34:46.800
It it sees the things that you see and it can compound your impact.
00:34:47.199 --> 00:34:55.039
But the biggest thing is it's going to see what you can't see because it has access to all of that data in real time, dynamic data.
00:34:55.119 --> 00:34:58.880
And an agent can just go be working for you around the clock.
00:35:00.239 --> 00:35:08.960
And even if there's a certain piece of software and you're like, yeah, our team, you know, half of our team, this is where we work, right?
00:35:09.119 --> 00:35:10.719
There's never gonna be AI in this.
00:35:10.800 --> 00:35:11.679
Yeah, there is, right?
00:35:11.760 --> 00:35:18.800
A Gartner study said that 40% of enterprise apps are gonna have agents by this year.
00:35:19.039 --> 00:35:23.519
So essentially, all enterprise software is gonna be agent-led, anyways, right?
00:35:23.760 --> 00:35:34.000
So don't just think, oh, well, you know, right now there's not an AI present, you know, our you know, piece of software that we use, you know, you know, whatever it is, you know, no, it's not gonna be touched.
00:35:34.079 --> 00:35:35.039
Yes, it is.
00:35:35.360 --> 00:35:37.519
It's everything's going agency.
00:35:37.679 --> 00:35:47.679
Not only that, but with the MCP protocol and some of these other protocols, at that point, that really takes away the traditional human work, right?
00:35:47.840 --> 00:35:57.039
And that turns us more into orchestrator, uh, orchestrators, tastemakers, and experts who should be driving these loops.
00:35:58.480 --> 00:36:11.440
So choosing your AI platform is now probably more strategic than choosing your actual operating system or saying we're a Microsoft organization or we're a Google organization.
00:36:11.840 --> 00:36:19.039
It is huge, but you have to make sure that you understand the data side as well, the governance, right?
00:36:19.199 --> 00:36:22.320
You have to make sure you understand what controls you have.
00:36:22.400 --> 00:36:41.360
Uh, you know, human approval for sensitive actions, uh, audit logs, how you can uh tackle prompt injections, um, you know, making sure that your data is integrated in the right way and still making sure that whatever systems you're using, that you still adhere to the uh the permissions that you've set in those tools.
00:36:41.519 --> 00:36:43.920
And also portability, right?
00:36:44.079 --> 00:36:46.639
Are you gonna rely on just MCP?
00:36:46.719 --> 00:36:48.960
Are you gonna build modularly, right?
00:36:49.039 --> 00:37:00.480
Because if all of your workflows depend on one single AI and you haven't been testing them and scoping them in any others, if that model has a pretty big change, right?
00:37:00.639 --> 00:37:07.599
As an example, going from a uh GPT-4.0 to a GPT-5, and maybe things don't work like they used to, right?
00:37:07.760 --> 00:37:12.800
Especially when we talk about uh agentic models that are doing a lot of the work by themselves under the hood.
00:37:12.960 --> 00:37:15.119
What if their tooling really changes?
00:37:16.079 --> 00:37:19.039
Then switching becomes nearly impossible.
00:37:19.360 --> 00:37:26.719
So not only do you have to be smart in how you set it up, but you also should always be running backup and building modularly.
00:37:26.800 --> 00:37:37.760
And not saying you need to be working in multiple AI operating systems because I don't think that's the best way, but you should have at least a small set of your team always looking at what's our backup plan, right?
00:37:37.920 --> 00:37:41.199
Because people are always like, okay, well, with all this extra time, what should humans be doing?
00:37:41.280 --> 00:37:46.239
Well, hopefully focusing on higher value work, more creative work, more strategic work.
00:37:46.400 --> 00:37:50.639
But you need people like me on your team that are keeping up with everything every single day.
00:37:50.800 --> 00:37:54.639
You need people who are creating backups and redundancies, right?
00:37:54.719 --> 00:37:56.159
And looking at the redundancies.
00:37:56.320 --> 00:37:58.000
That's what you should be doing.
00:37:58.320 --> 00:38:02.880
Because, like I said, human work is changing.
00:38:03.519 --> 00:38:11.519
So, to quickly recap, the line between large language models and the internet is going out the window.
00:38:11.679 --> 00:38:22.880
So, in the same way that all of our work now happens on the internet, and we don't necessarily make that choice to use the internet, it's just, well, when you turn your computer on, there's your outlook.
00:38:23.039 --> 00:38:24.239
It's on the internet, right?
00:38:24.400 --> 00:38:28.079
Your teams, it's like everything is powered by the internet.
00:38:28.320 --> 00:38:31.039
And that's where we're headed with AI.
00:38:31.280 --> 00:38:37.519
We're no longer, I'll say if you know, listen to this in January 2027, and I think you'll agree.
00:38:38.000 --> 00:38:43.599
We're no longer going to be making an active choice to go use AI like we have for the past two or three years.
00:38:43.760 --> 00:38:56.559
It is going to be there, it is going to be in our face, and more and more smart teams are going to be pushing all of their day-to-day operations into front-end large language models, the team versions, or what I like to call AI operating systems.
00:38:56.719 --> 00:38:59.360
So you need to be thinking about it now.
00:39:00.400 --> 00:39:02.800
What is your plan, right?
00:39:02.960 --> 00:39:06.159
When your competitors beat you to this, right?
00:39:06.320 --> 00:39:12.000
And they can run leaner, they can run faster, and they start gobbling up your slice of the pie.
00:39:12.159 --> 00:39:14.320
You have to start thinking about it now.
00:39:14.880 --> 00:39:25.679
The technology is there, the capabilities are there, the results and the outputs are there, but you just have to meet them there.
00:39:27.199 --> 00:39:29.519
What's keeping you from doing that?
00:39:29.679 --> 00:39:44.159
We've talked about in the first two episodes of the Start Here series, but it is making sure that your team seems your team stays up to date, educated, and you have to be able to carefully sprint as fast as you can.
00:39:44.320 --> 00:39:44.559
All right.
00:39:44.639 --> 00:39:48.320
I hope this show was helpful talking about AI as an operating system.
00:39:48.480 --> 00:40:06.480
If it was, make sure you go to thestartherseries.com and that will give you free access to our inner circle community where you can go uh catch up with all the different Start Here series, all the different uh episodes in that volume, as well as get free access to our prime prompt polish course.
00:40:06.559 --> 00:40:13.519
And you know, you'll get to network with hundreds of other uh just trailblazers in the AI space.
00:40:13.599 --> 00:40:26.559
I mean, we have leaders from Fortune 500 companies, small business, you know, CEOs, uh, you know, marketers, advertisers, anything you can think of, a bunch of great people who are pushing uh just the pace of AI in the inner circle community.
00:40:26.719 --> 00:40:27.840
So thank you for tuning in.
00:40:28.000 --> 00:40:31.119
Hope to see you back tomorrow and every day for more everyday AI.
00:40:31.199 --> 00:40:32.079
Thanks, y'all.
00:00:02.000 --> 00:00:04.160
How are you using AI right now?
00:00:04.639 --> 00:00:07.200
If you're like most people, it's intermittently.
00:00:07.599 --> 00:00:12.240
Maybe to start a project or in the middle to end or maybe the whole thing.
00:00:12.480 --> 00:00:19.280
But chances are you go out of your way and you make a choice to go and use AI.
00:00:20.640 --> 00:00:21.839
It's not like that anymore.
00:00:22.000 --> 00:00:25.839
I like to talk about AI like the internet.
00:00:26.800 --> 00:00:29.519
You can't work without the internet, right?
00:00:29.679 --> 00:00:38.320
And I'm not saying going to different websites, your email, your CRM, your project management system, your ERP, everything is connected to the internet.
00:00:38.640 --> 00:00:42.960
And it's not like we now make a choice to use the internet.
00:00:43.200 --> 00:00:44.799
It's just there and we need it.
00:00:45.359 --> 00:00:47.439
That's the parallel I'm trying to draw here.
00:00:47.600 --> 00:00:52.640
I think most people for the last couple of years have had to make a choice to use AI.
00:00:53.200 --> 00:00:56.079
In 2026 and beyond, it's not like that anymore.
00:00:56.240 --> 00:01:03.359
I think AI is, for better or worse, going to infiltrate every single corner of where and how we work.
00:01:03.840 --> 00:01:21.280
And so on today's show, the Start Here series, volume three, we're going to be talking about how I think you should be using AI as an operating system, not as an afterthought or not intermittently, but the foundation of how you and your team work.
00:01:21.519 --> 00:01:21.920
All right.
00:01:30.000 --> 00:01:40.239
So if you are new, our Start Here series is everyday AI's essential beginner and advanced AI overviews to launch your year strong.
00:01:40.480 --> 00:01:45.680
So if you haven't heard, well, you should probably go to starthearseries.com.
00:01:45.840 --> 00:01:54.079
That is going to get you a free entry into our inner circle community and all of the Start Here Series podcasts.
00:01:54.239 --> 00:01:56.480
You don't got to go fishing for them on our website.
00:01:56.640 --> 00:02:05.680
They're all going to be right there in that space, as well as you get access to our free Prime Prompt Polish Chat GPT prompt engineering course.
00:02:05.840 --> 00:02:07.359
So make sure you go check that out.
00:02:07.519 --> 00:02:10.400
And if you're very, very new here, well, this is everyday AI.
00:02:10.479 --> 00:02:13.840
And we've done now almost 700 podcast episodes.
00:02:13.919 --> 00:02:16.240
And one of the most common questions I get is where do I start?
00:02:16.319 --> 00:02:19.360
And that's why we're starting the start here series.
00:02:20.080 --> 00:02:27.039
So uh if you missed our last uh start here um episode, it was episode 693.
00:02:27.120 --> 00:02:30.240
So you can go click back a couple of times on the podcast player there.
00:02:30.400 --> 00:02:32.479
But we talked about AI without the jargon.
00:02:32.560 --> 00:02:38.319
So we kicked this whole thing off by talking about what is generative AI, what are large language models, volume two.
00:02:38.400 --> 00:02:44.000
Uh we dove into just the language and the lingo, because if you don't understand that, you can't compete.
00:02:44.080 --> 00:02:53.120
Uh and now today we're going to be talking about how I think you should be looking at AI as an operating system, not as a tool that you selectively go to use.
00:02:54.319 --> 00:02:56.479
So why?
00:02:56.879 --> 00:03:02.879
Well, I think AI is no longer an app you visit, it's becoming how you work.
00:03:03.039 --> 00:03:16.879
You know, if we think back to the early days of Chat GPT, which I think is most people's introduction to generative AI, but even to last year, I think still most people using AI went and sought it out.
00:03:17.120 --> 00:03:19.919
It's not going to be like that for very much longer.
00:03:20.159 --> 00:03:20.639
Why?
00:03:20.879 --> 00:03:26.800
Because wherever you work, AI has already infiltrated it and you can't really get around it.
00:03:26.879 --> 00:03:31.120
Just like you can't get around doing knowledge work without using the internet.
00:03:31.599 --> 00:03:32.639
Think about it.
00:03:33.120 --> 00:03:38.479
If you're in Microsoft Excel, Microsoft Copilot's there.
00:03:38.719 --> 00:03:43.280
If you're opening up a blank uh Google Doc, Gemini's there.
00:03:43.439 --> 00:03:46.400
If you're in Salesforce, Agent Force is there.
00:03:46.560 --> 00:03:54.319
It's very hard now, regardless of where you work, for there not to be AI already embedded there.
00:03:54.719 --> 00:03:58.000
Yet you still have to kind of go out of your way to use it.
00:03:58.240 --> 00:04:11.280
But we're going to talk about how I think that's not only changing, but how you as business leaders need to shift your mindset and understand that what I think is gonna happen, it's definitely gonna happen.
00:04:11.360 --> 00:04:15.840
Uh, front-end large language models are going to be the future interface of work.
00:04:15.919 --> 00:04:18.560
And yeah, we got we got receipts, y'all.
00:04:18.720 --> 00:04:19.680
Don't worry.
00:04:20.000 --> 00:04:30.480
So, uh, in today's show, we're gonna talk about how the internet is becoming a background service for your AI assistants and probably nothing more, and how that changes knowledge work.
00:04:30.639 --> 00:04:53.519
We're gonna be talking about the protocols and platforms that are making this shift to an AI operating system happen, such as ChatGPT apps, uh, the new Claude Co-work, uh, MCP from uh Anthropic, their model uh contacts protocol, and that how that helps agents kind of not need us for the things that we've maybe been doing a lot over the last year or two.
00:04:53.680 --> 00:04:59.519
And then last, but how business leaders need to change.
00:05:00.079 --> 00:05:01.519
That's the reality.
00:05:01.920 --> 00:05:07.519
A lot of people think of AI as a tech implementation, it's not, it's people management.
00:05:07.680 --> 00:05:13.519
And you need to know the adoption stats, the risks, and how to choose your AI operating system.
00:05:13.600 --> 00:05:15.439
So we're gonna be diving into all that.
00:05:15.519 --> 00:05:17.040
But let's start at the top.
00:05:17.360 --> 00:05:20.160
Why do you need an AI operating system?
00:05:20.319 --> 00:05:22.000
And what the heck is it?
00:05:22.240 --> 00:05:24.399
Um, it's not an official term.
00:05:24.560 --> 00:05:31.120
I don't know if I made it up, but I've been calling, I've been talking about AI operating systems now for more than two years.
00:05:31.279 --> 00:05:39.040
Uh, but the reality is right now, there's almost 2 billion people that use AI almost weekly.
00:05:39.120 --> 00:05:46.879
And 90% of knowledge workers who use AI report higher productivity than those who don't, obviously, right?
00:05:47.199 --> 00:05:54.560
Um, and not just how we use AI, but AI's capabilities have obviously greatly changed.
00:05:54.720 --> 00:06:00.959
I think, especially um in the year 2025, more so than the two years prior.
00:06:01.360 --> 00:06:08.160
And the main reason for that is large language models mainly did there's two big advancements.
00:06:08.399 --> 00:06:11.839
Number one, bringing in your company's dynamic data.
00:06:11.920 --> 00:06:13.279
And that's big, right?
00:06:13.360 --> 00:06:17.759
That's what cuts down on hallucinations, but that's also what increases collaboration.
00:06:17.920 --> 00:06:20.079
And that is the second one, right?
00:06:20.160 --> 00:06:26.480
Uh large language models and front-end AI chatbots are no longer, oh, I'm just chatting with a helpful assistant.
00:06:26.639 --> 00:06:32.079
No, now your business's data is there and you can collaborate with team members, right?
00:06:32.319 --> 00:06:39.360
Work in the same chat, like you're sharing a uh Google Doc and you're all collaborating there live in in real time.
00:06:39.600 --> 00:06:46.399
And that's because right now integrated AI systems are a baseline requirement, right?
00:06:46.800 --> 00:06:55.199
Um, I like to think most companies at some point in the uh in the 90s made a decision, right?
00:06:55.279 --> 00:06:57.279
And early 2000s as well.
00:06:57.519 --> 00:07:02.879
They said, we're either a Windows organization, we're a Mac organization, we're a Linux organization.
00:07:03.120 --> 00:07:16.720
You chose your operating system, and then you kind of your front-end processes that happened before the computer, during the computer, and obviously what happened after, were largely based around what operating system that you chose.
00:07:17.040 --> 00:07:23.519
And I think the same can be true for the choices that we have now ahead as everything becomes AI native.
00:07:23.600 --> 00:07:32.399
Uh, that's because a single person with good AI skills, using the right agents, the right co-pilot, can accomplish the work of a larger team.
00:07:32.639 --> 00:07:40.879
So naturally, we have an extremely strong orchestration layer that's already in place.
00:07:42.079 --> 00:07:44.240
And it's just begging, right?
00:07:44.399 --> 00:07:52.240
The smartest organizations I talk to, I've, you know, lucky enough to get to talk to hundreds of very smart people in AI every single year.
00:07:52.480 --> 00:07:55.839
The smartest organizations have already done this.
00:07:56.079 --> 00:08:09.839
They've already brought their entire team, their entire processes into, you know, ChatGPT uh enterprise or you know, Google Gemini Enterprise, uh, Claude Enterprise, obviously Microsoft 365 Copilot.
00:08:10.079 --> 00:08:12.160
And they've changed their processes.
00:08:12.319 --> 00:08:20.399
Not just, oh, we're gonna use uh, you know, we have a 10-step process, we're gonna use AI at step two, four, and seven.
00:08:20.800 --> 00:08:27.040
No, they said that 10-step process process is now a three-step process, right?
00:08:27.279 --> 00:08:33.200
Uh human orchestrates something, agents go and do something, human checks it on the back end.
00:08:33.360 --> 00:08:39.360
And yeah, we went from 10 steps to three steps, and from 10 hours to 30 minutes.
00:08:39.679 --> 00:08:44.240
Those are the moves that the business leaders right now are taking.
00:08:44.559 --> 00:08:47.120
And that's why you need an AI operating system.
00:08:47.360 --> 00:08:49.919
All right, and we're not going to dive into this too much, right?
00:08:50.080 --> 00:08:50.960
Oh, what do you choose?
00:08:51.039 --> 00:08:51.600
What do you choose?
00:08:51.679 --> 00:08:51.919
Right.
00:08:52.000 --> 00:08:58.639
We'll maybe do another uh show later in our start here series on this, but I think you essentially have four choices.
00:08:58.879 --> 00:09:03.759
And well, if you're a Microsoft co-pilot organization, you might be able to choose two.
00:09:04.080 --> 00:09:05.200
So let's start there.
00:09:05.360 --> 00:09:08.799
Microsoft 365 Copilot, hats off to them.
00:09:09.120 --> 00:09:14.799
They started technically the enterprise AI craze, right?
00:09:14.960 --> 00:09:19.519
This is back when large language models essentially you had front-end chatbots, and that was it.
00:09:19.679 --> 00:09:28.159
Microsoft was the organization that brought enterprise AI into the mainstream by launching Microsoft Copilot.
00:09:28.559 --> 00:09:30.720
It is extremely robust.
00:09:31.200 --> 00:09:33.519
It is wildly flexible.
00:09:33.840 --> 00:09:38.320
Yet the learning curve is bonkers, right?
00:09:38.480 --> 00:09:45.600
That's one of the reasons why I think Microsoft Copilot hasn't just run away with it and they haven't been running away with it for years.
00:09:46.080 --> 00:09:51.120
Yes, there's a lot of um technical intricacies that I don't even understand.
00:09:51.200 --> 00:09:53.919
And the majority of people, unless you work in IT, don't, right?
00:09:54.000 --> 00:09:57.759
But there's mainly, I think, permissions, right?
00:09:57.919 --> 00:10:06.799
Anyone I talk to that's a Microsoft 365 copilot organization, they see all these great uh you know features that Copilot has.
00:10:07.039 --> 00:10:11.600
And the overwhelming majority of enterprises don't know how to get it set up, right?
00:10:11.759 --> 00:10:14.639
It's essentially you have gatekeepers and bottlenecks, right?
00:10:14.799 --> 00:10:18.240
Usually, you know, IT is looking for approval.
00:10:18.320 --> 00:10:19.759
People are like, why don't I have this?
00:10:19.919 --> 00:10:24.879
Someone else has access uh to this, you know, this in you know, copilot studio.
00:10:25.039 --> 00:10:25.519
I want that.
00:10:25.679 --> 00:10:26.639
No, you can't have that.
00:10:26.720 --> 00:10:27.600
How do we get that right?
00:10:27.759 --> 00:10:29.120
That's what a lot of it is.
00:10:29.360 --> 00:10:38.159
For the organizations that have robust uh, you know, Microsoft training already in place are the ones that have obviously done really well here.
00:10:38.320 --> 00:10:43.039
But uh Microsoft 365 Copilot obviously works at the OS level, right?
00:10:43.120 --> 00:10:47.279
Uh, it's the only uh AI that's baked into an operating system.
00:10:47.440 --> 00:10:51.600
Obviously, Apple is not there and they're very far from ever being there anytime soon.
00:10:52.000 --> 00:10:53.919
Next, Google Gemini.
00:10:54.320 --> 00:10:59.039
They won 2025 and they're innovating faster than anyone, uh, right.
00:10:59.440 --> 00:11:06.159
And similarly, like Microsoft 365 Copilot, you can get Google Gemini in a lot of different ways, right?
00:11:06.240 --> 00:11:07.120
People don't even know this.
00:11:07.200 --> 00:11:14.080
As an example, probably the Gemini that you use and have heard of is not even the best Gemini for Teams, right?
00:11:14.240 --> 00:11:23.840
Uh, about four months ago, Google actually came out with a business version of Gemini called Gemini Business and Gemini Enterprise.
00:11:23.919 --> 00:11:24.720
It is separate, right?
00:11:24.799 --> 00:11:35.200
So I have a paid uh Gemini account for my personal Gmail, I have a paid Gemini account um for my work email, and then I have a paid uh Gemini business account.
00:11:35.279 --> 00:11:37.120
All different work a little differently.
00:11:37.200 --> 00:11:40.480
Um, and that is also one of the pros and cons of Google Gemini as well.
00:11:40.559 --> 00:11:43.279
They all the versions work a little differently as well.
00:11:43.440 --> 00:11:50.240
There's pros and cons to them, and that's not even the tip of the iceberg of where you can use Google Gemini, right?
00:11:50.320 --> 00:11:54.799
Uh in AI Studio, uh, they have so many different, you know, vibe coding platforms.
00:11:54.960 --> 00:11:58.480
Uh, but it's uh Google Gemini is great, it's all over the place.
00:11:59.679 --> 00:12:02.399
Then you have Anthropic Claude.
00:12:02.559 --> 00:12:10.480
Uh so I will say this I am much more bullish on uh Claude than I was probably two months ago.
00:12:10.639 --> 00:12:17.600
And a lot of that has to do with, well, mainly two big uh pieces that I've seen from Anthropic.
00:12:17.919 --> 00:12:25.279
One is they were the first to really popularize file creation, which might sound crazy, right?
00:12:25.360 --> 00:12:29.519
Like, why haven't large language models just been doing that uh at the forefront?
00:12:29.759 --> 00:12:33.759
Um, well, Anthropic popularized it, right?
00:12:33.840 --> 00:12:44.799
They were the first one to get um Excel uh sheets right, right, which is kind of crazy, even before, in some instances, Microsoft 365 Copilot.
00:12:44.879 --> 00:12:46.320
The same thing with PowerPoint.
00:12:46.720 --> 00:12:48.720
Uh Claude really pushed the boundary there.
00:12:48.799 --> 00:12:56.559
And then they also uh, you know, just came out with their uh Claude co-work, a great um, a great tool, FYI.
00:12:56.639 --> 00:12:58.320
We're gonna have a show on that tomorrow.
00:12:58.480 --> 00:13:05.759
But I think that in 2024, in the early part of 2025, they were overly focused on coding and software engineering.
00:13:05.840 --> 00:13:13.759
Uh, and I do think that they've made a pivot over the last quarter or so uh to really go after the non-technical business leader.
00:13:13.919 --> 00:13:17.759
So uh they also have a great enterprise team product as well.
00:13:17.919 --> 00:13:20.480
And then last but not least, ChatGPT, right?
00:13:20.559 --> 00:13:23.039
In terms of users, no one's coming close.
00:13:23.200 --> 00:13:25.440
In terms of retention, no one's coming close.
00:13:25.600 --> 00:13:27.360
Stickiness, no one's coming close.
00:13:27.519 --> 00:13:30.960
Um, but it seems like they're really focused on consumers.
00:13:31.120 --> 00:13:34.080
Yes, they have a great uh Chat GPT business tier.
00:13:34.159 --> 00:13:42.000
They have a great, you know, Chat GPT enterprise tier where, you know, reportedly, I think 92% of Fortune 500 companies uh use one of their products.
00:13:42.320 --> 00:13:49.279
Uh, but I do think uh that right now, OpenAI is maybe a little more focused on the consumer side.
00:13:49.360 --> 00:13:52.559
Uh, you know, recently announced, you know, ads, right?
00:13:52.720 --> 00:13:55.039
A lot of things, ChatGPT health.
00:13:55.120 --> 00:13:56.720
Uh, but maybe it makes sense.
00:13:56.799 --> 00:14:05.840
Maybe that's their strategy to bring teams into the fold is to, you know, become so sticky in their personal life that it's hard for people to use anything else.
00:14:06.000 --> 00:14:17.600
So when you talk about bringing your team, and like I said, all four of those obviously have great options for, you know, individuals, small teams, large teams, uh, enterprise organizations, right?
00:14:17.759 --> 00:14:19.679
You have to choose which one is right for you.
00:14:19.759 --> 00:14:24.320
We're not gonna go too far into that, but in my opinion, those are your choices, right?
00:14:24.399 --> 00:14:25.759
Especially if you're in the US.
00:14:25.840 --> 00:14:28.960
Uh, yeah, you can look at open source solutions, right?
00:14:29.200 --> 00:14:30.879
Uh, I'm not gonna get into that.
00:14:31.039 --> 00:14:39.039
I think you have to choose, just like you know, you choose uh Windows, you choose Mac, you choose Linux.
00:14:41.519 --> 00:14:44.720
And this is actually a very telling sign.
00:14:44.799 --> 00:15:01.919
Uh, so we covered this a couple of months ago when this was first reported, but there was a report that Microsoft's CEO Sadia Nadella uh essentially warned his employees that Microsoft and Windows, their office suite, could become obsolete.
00:15:02.320 --> 00:15:07.440
Yeah, he said that they had to reboot and focus on AI or the company.
00:15:08.000 --> 00:15:10.399
The company could become obsolete.
00:15:12.559 --> 00:15:15.759
Now, if that sounds kind of like a wild statement, well, it kind of is.
00:15:15.840 --> 00:15:27.440
I mean, Microsoft has, you know, for the past, you know, three to four years traditionally been uh top one, two, or three company in the world when it comes to market cap.
00:15:28.159 --> 00:15:39.840
Yet, Nadella warned that without a fundamental overhaul and a focus on AI, that Microsoft's most successful businesses could become irrelevant as AI agents replace traditional software.
00:15:40.080 --> 00:15:40.320
Right?
00:15:40.480 --> 00:15:42.720
Yeah, saying no one's gonna use Office anymore.
00:15:42.799 --> 00:15:45.600
No one's gonna use, you know, all of these different tools.
00:15:45.759 --> 00:15:51.840
And that's why Microsoft reportedly is transitioning from a software factory to an intelligence engine, right?
00:15:52.080 --> 00:16:09.279
Sadia Nadella, one of, I think, the smartest tech people probably to ever live, is saying that, hey, even we as the largest company in the world, that we power the enterprise when it comes to an operating system, we're not gonna make it.
00:16:09.600 --> 00:16:13.120
Because the business world is transitioning away from that.
00:16:13.279 --> 00:16:15.519
And it's all about your intelligence engine.
00:16:15.600 --> 00:16:20.639
It is about how you put your knowledge to work using AI, right?
00:16:20.799 --> 00:16:30.720
And he said that Windows uh becoming an AI first operating system, uh where the keyboard and the task bar become secondary.
00:16:30.960 --> 00:16:36.879
So his goal was for every Office application to function as a development environment for AI agents, right?
00:16:37.919 --> 00:16:38.799
Think about that.
00:16:40.080 --> 00:16:52.480
That's right, like billions of people use Microsoft in his goal that all that's going to be is a base for agents to work.
00:16:53.600 --> 00:17:02.480
That's the shift that I think a lot of people, uh, right, you should probably be listening and paying attention to what he's saying, uh, because that's the truth.
00:17:04.400 --> 00:17:09.920
And it's also because, like I said, large language models are much more than that, right?
00:17:10.319 --> 00:17:12.960
They are now agentic.
00:17:13.839 --> 00:17:15.279
They're now agents, right?
00:17:15.359 --> 00:17:16.000
That's the thing.
00:17:16.160 --> 00:17:19.119
You go into Chat GPT, people don't know this.
00:17:19.200 --> 00:17:28.400
You can schedule an agent to go, you can give it access to you know your uh connected and synced data.
00:17:28.559 --> 00:17:36.400
Uh, you can log into certain websites and you can say, hey, every day at you know, 7:30, go do these five tasks for me, right?
00:17:37.119 --> 00:17:47.680
People are still looking at uh large language models as you know, fun little toys to help you write better emails, and they're not that.
00:17:47.839 --> 00:17:56.000
Uh, the orchestration, the tools, uh, you know, with the browsers, code executors, app connectors, uh, bringing in your company's dynamic data.
00:17:56.079 --> 00:17:57.279
Uh it's it's huge.
00:17:57.440 --> 00:18:07.759
And now as memory and context improves, so done so does an agent's ability to orchestrate, to plan and execute taps, right?
00:18:07.839 --> 00:18:16.799
As the memory, the context window, uh the personalization and the ability to properly use data uh with the harnessing of these models.
00:18:17.039 --> 00:18:19.039
It's it's it's nuts, right?
00:18:19.279 --> 00:18:25.359
And and this is the cognitive layer that Saudi and Adela, the intelligence layer that he's talking about.
00:18:25.599 --> 00:18:32.160
As the models get exponentially smarter and they get more and more access to our data and they get better and better at using tools.
00:18:32.400 --> 00:18:33.440
Hello, people.
00:18:34.000 --> 00:18:35.680
This is what humans do, right?
00:18:35.759 --> 00:18:39.519
Uh, I've talked one or two times uh through this start here series.
00:18:39.599 --> 00:18:46.559
Um, and I'm gonna talk about it probably a little bit tomorrow when we talk about Claude Cowork, very much uh related uh to this show.
00:18:47.039 --> 00:18:58.559
I think a lot of the work that humans in companies that are ahead in AI is acting as duct tape, right?
00:18:58.880 --> 00:19:02.240
We're duct taping and connecting different agentic flows.
00:19:02.319 --> 00:19:04.319
That's a lot of what I do all day, right?
00:19:04.720 --> 00:19:09.839
I have a bunch of agents going to do things, I orchestrate them, you know, I taste make.
00:19:09.920 --> 00:19:11.440
I'm like, good job, agent one.
00:19:11.519 --> 00:19:14.240
Uh hey, agent two, that's not what we wanted, right?
00:19:14.400 --> 00:19:15.519
I'm duct taping.
00:19:16.000 --> 00:19:19.680
I mean, I think even though that's weird, I think that's where we're headed.
00:19:20.000 --> 00:19:28.640
Because large language models are not chatbots, they're literally that human intelligence layer that Sonia Nadella was talking about.
00:19:30.720 --> 00:19:33.279
So let's get into the web, right?
00:19:33.519 --> 00:19:36.960
And how we can't work without the web.
00:19:38.240 --> 00:19:43.440
Right now, if you're listening to this, if you're watching this, it's because of the web, right?
00:19:43.519 --> 00:19:44.240
It's because of the internet.
00:19:44.720 --> 00:19:45.680
Everything we do is the internet.
00:19:45.839 --> 00:19:51.519
If you're a knowledge worker, it is because of the power of the internet.
00:19:52.319 --> 00:19:52.640
Okay?
00:19:53.119 --> 00:20:01.839
I like to tell people you need to replace the word internet with AI, because the two are gonna become one.
00:20:02.319 --> 00:20:02.559
Right?
00:20:02.640 --> 00:20:06.240
You think of anything, think of a Google search, right?
00:20:06.400 --> 00:20:09.759
Um, eventually it's gonna be just an AI mode search, right?
00:20:09.920 --> 00:20:14.240
Um they've already said Google has said that that is going to be the end goal.
00:20:14.319 --> 00:20:16.240
Everything's going to go to AI mode.
00:20:16.720 --> 00:20:19.839
So are you gonna browse the web in the future?
00:20:20.000 --> 00:20:20.799
I'd say probably not.
00:20:20.960 --> 00:20:23.759
Browsing the web is an absolutely terrible experience.
00:20:24.400 --> 00:20:25.119
The same thing.
00:20:25.279 --> 00:20:30.960
I think for the most part, the web will be made for agents, right?
00:20:31.200 --> 00:20:39.279
And in the same way that we have these protocols that allow agents to talk to each other, they're gonna be doing the web browsing as well.
00:20:39.759 --> 00:20:40.319
Right.
00:20:40.480 --> 00:20:42.319
Um, I mean, look where we are now.
00:20:42.400 --> 00:20:47.200
Look at some of the I think the best uh advancements of 2025 were agenc browsers.
00:20:47.279 --> 00:20:51.440
I think uh if you had to give categories a grade, right?
00:20:51.599 --> 00:20:54.720
I think niche agents got an A.
00:20:55.039 --> 00:20:58.400
I think general agents that we thought were gonna take over got like a C.
00:20:58.880 --> 00:21:01.839
And I think that agenc browsers got an A as well.
00:21:01.920 --> 00:21:02.400
Uh, right.
00:21:02.480 --> 00:21:09.279
I think agentic browsers performed much better or at higher uh aptitude than agents did.
00:21:09.519 --> 00:21:18.880
So working in uh autonomous capabilities into your browser and then pairing it with memory, which I think OpenAI's Atlas does really, really well, right?
00:21:19.200 --> 00:21:28.720
That gives your browser the ability to number one just perform tasks for you, uh, right, but to bring your context and your history over as well.
00:21:29.119 --> 00:21:29.440
Right.
00:21:29.599 --> 00:21:33.519
Think about, I don't know, an example, filling forms.
00:21:34.319 --> 00:21:36.079
Why should we do that anymore?
00:21:36.400 --> 00:21:37.519
It's stupid.
00:21:38.160 --> 00:21:38.720
Right?
00:21:38.960 --> 00:21:48.480
If if if an agentic browser has all of that in its memory, has all that information and can navigate these, we should never be filling forms, right?
00:21:48.559 --> 00:21:53.440
That's why I think so much of what's happening on the web is just going to become for agents.
00:21:53.599 --> 00:21:59.599
And I don't think that we're gonna be browsing the web much longer, which does change the future of work because.
00:22:00.640 --> 00:22:01.599
That's what we do.
00:22:01.759 --> 00:22:02.960
That's what we do as humans.
00:22:03.119 --> 00:22:04.880
We go do work on the web.
00:22:06.319 --> 00:22:08.079
Asians are better.
00:22:08.240 --> 00:22:11.599
If you know what you're doing, if you're using the right model, right?
00:22:11.920 --> 00:22:16.960
The the the frontier state of the art, they're smarter than us all, right?
00:22:17.039 --> 00:22:27.200
You look at offline IQ tests, they're nearly at genius level, but smarter than probably 99.99% of anyone listening, a thousand times smarter than me.
00:22:27.680 --> 00:22:34.559
Um so what used to take many searches, clicks, and tabs is now just delivered as one concise answer.
00:22:37.519 --> 00:22:46.240
So let's get back a little bit to how using AI as an operating system can solve that.
00:22:46.400 --> 00:22:51.119
Because one of the things that I think we waste so much time for is context switching, right?
00:22:51.599 --> 00:22:54.400
Every extra login, every copy and paste.
00:22:54.559 --> 00:23:01.119
Oh my gosh, this formatting's wrong from Copilot, uh, Chat GPT, I, you know, I need this in Markdown, it's not working, right?
00:23:01.359 --> 00:23:08.160
We spend so much time reformatting, copying and pasting, you know, switching uh, you know, tabs.
00:23:08.240 --> 00:23:12.160
Uh you know, analysts call this the app hop tax, right?
00:23:12.640 --> 00:23:18.079
Going from all these different apps, all these different softwares, you lose so much time, right?
00:23:18.400 --> 00:23:29.519
And to be honest, aside from summarizing information and synthesizing things and personalizing information, that's what knowledge workers do, right?
00:23:30.000 --> 00:23:37.839
We use our brain to go and summarize, synthesize, and rewrite information.
00:23:38.079 --> 00:23:40.400
And then we context switch, right?
00:23:40.559 --> 00:23:42.640
Okay, I'm gonna read this industry report, okay?
00:23:42.960 --> 00:23:46.880
I'm gonna personalize that context, take it by to my team, okay.
00:23:47.039 --> 00:23:50.720
Uh, whatever the team says, then I'm gonna go put it in a blog post, okay?
00:23:50.960 --> 00:23:52.640
Then I'm gonna go put it on the web, okay.
00:23:52.720 --> 00:23:56.319
Our customers are gonna read it, I'm gonna sign up calls right with the customers, okay, right?
00:23:56.480 --> 00:24:01.839
All you're doing is you're just slightly switching that context over and over and over and over again.
00:24:03.920 --> 00:24:05.359
It's gotta be agentic, right?
00:24:05.440 --> 00:24:15.599
Right now, that's friction because agents can handle most of that right now, right?
00:24:16.000 --> 00:24:26.880
Um in a in an agent native flow or an LLM native flow, an agent can execute everything inside of a single thread.
00:24:28.559 --> 00:24:28.880
Right.
00:24:28.960 --> 00:24:34.559
Think of those multi-step tasks that you do every single day, right?
00:24:34.640 --> 00:24:38.240
I'll even give you a very quick personal example of how it's changed for me.
00:24:38.480 --> 00:24:38.880
Right?
00:24:39.119 --> 00:24:43.759
When I first started this show, there wasn't a lot of AI out there.
00:24:44.079 --> 00:24:44.400
Right?
00:24:44.559 --> 00:24:50.160
There wasn't, you know, there was no Google didn't have AI, Anthropic didn't exist, there was no Metallama, right?
00:24:50.240 --> 00:24:52.559
There was Chat GPT, it was a nice little chat bot.
00:24:52.640 --> 00:24:57.440
So, how this used to work, I'd have spent a lot more time reading and synthesizing information.
00:24:57.599 --> 00:25:02.960
The quality of my shows wasn't able to be as good because it took way too much time.
00:25:03.359 --> 00:25:04.160
Um, right.
00:25:04.319 --> 00:25:14.799
I wasn't able to go as deeply into topics, you know, it was a lot more interviews, a lot more topical things without being able to dive deep and I think have a much higher quality.
00:25:15.119 --> 00:25:27.119
Now I have multiple agents going out, looking at my context every single day, looking at my podcast stats every single day, looking at what people are clicking on in the newsletter every single day.
00:25:27.279 --> 00:25:32.480
And then based on all of that, right, based on my context, my point of view, right?
00:25:32.640 --> 00:25:42.160
What stats, you know, what podcasts are performing well, what people are clicking on, they go in real time, base all that, do reports autonomously.
00:25:42.240 --> 00:25:52.079
I have an agent that looks at multiple agents' reports and they get back to me and they say, Hey, based on what you cover, based on what your audience cares about, here's what you should be covering.
00:25:52.160 --> 00:25:53.599
And I look through that, right?
00:25:53.759 --> 00:25:59.440
As the orchestrator on the front end, the taste makes tastemaker on the back end, and I say, okay, great.
00:25:59.680 --> 00:26:00.000
Right?
00:26:00.079 --> 00:26:01.519
I still take the time to read it.
00:26:01.680 --> 00:26:08.960
I always, you know, go through and read, read the chain of thought, uh, you know, the observability and traceability 101.
00:26:09.599 --> 00:26:16.559
But it's no longer a step that myself and my team have to go do over and over and over.
00:26:17.920 --> 00:26:21.200
And that's the promise of an AI operating system.
00:26:24.000 --> 00:26:28.160
And it's a lot of the recent technology that's actually going to make that work, right?
00:26:28.240 --> 00:26:33.200
And I think one of the things you have to call out by name is Anthropic's model context protocol.
00:26:33.359 --> 00:26:36.559
And actually, a pretty uh important recent update to that.
00:26:36.640 --> 00:26:55.279
So in December, Anthropic actually donated the MCP, made it open source to the Agentic AI Foundation under the Linux Foundation, with also uh the that same foundation, the Agentic AI Foundation, also supported by OpenAI, Google, Microsoft, and AWS.
00:26:55.440 --> 00:27:02.720
So, right now, MCP, the easiest way to think of it, it's like uh, you know, people call it the USB C of AI, right?
00:27:02.799 --> 00:27:07.039
It's just everything plugs into it and you know, it just converts and it's good, right?
00:27:07.119 --> 00:27:14.799
It's it's it's like a US, you know, one of those dongle converters where you can, you know, put anything on one end, put it anything on the other end, and it just works and it's great, right?
00:27:14.960 --> 00:27:19.920
This is kind of like the same thing with uh, you know, how websites have APIs and can talk to each other.
00:27:20.079 --> 00:27:24.160
Websites built on all kinds of different frameworks, it doesn't matter, right?
00:27:24.240 --> 00:27:30.319
If if one thing's on, you know, React, one thing's on PHP, one thing's, I don't know, JavaScript, HTML, CSS, right?
00:27:30.559 --> 00:27:35.920
All those websites can talk to each other, even though they're built in different languages because they have this API.
00:27:36.079 --> 00:27:42.799
It's a common way to transmit languages, and that's what you have now with the model context protocol.
00:27:42.880 --> 00:27:49.200
It allows AI agents, AI models, AI tools to all talk to each other, right?
00:27:50.160 --> 00:27:52.160
Part of it's great, right?
00:27:52.240 --> 00:27:53.519
That's amazing, right?
00:27:53.680 --> 00:28:05.519
And it's also great that OpenAI, Google, and Microsoft, uh Anthropic's technically three biggest competitors, have all adopted it and have put it uh you know, put it to use in their products for their users.
00:28:06.319 --> 00:28:17.359
But at the same time, you also have to think wait, doesn't that just in theory take away a lot of the work of what a human would be doing?
00:28:17.680 --> 00:28:19.039
Absolutely, it does.
00:28:19.200 --> 00:28:21.680
I don't think that's a bad thing, right?
00:28:22.000 --> 00:28:24.720
But that's a lot of whether you know it or not, right?
00:28:24.880 --> 00:28:40.240
If you took time to really go deep in model context protocol and look what's capable as an example, you know, in uh Claude Co-work, what's exam uh like what's available with uh you know OpenAI's uh like agent SDK, right?
00:28:40.480 --> 00:28:50.079
And being able to use in in co-pilot studio model context protocol, you probably wouldn't know it, but probably just about any different website, any different app that you use.
00:28:50.160 --> 00:28:56.160
And normally right now, maybe a human is interacting with that data via AI and they're going back, right?
00:28:56.400 --> 00:29:05.680
Human one is is working with you know, uh tool one with AI, human two working with tool two with AI, human three working with tool three with AI.
00:29:06.400 --> 00:29:11.200
You don't need that because now you don't have to have this vertical integration, right?
00:29:11.359 --> 00:29:16.000
Humans working vertically, going just up and down with a single tool, right?
00:29:16.240 --> 00:29:25.119
Now those AI tools and softwares can all uh work horizontally across the plane as well because of the model context protocol.
00:29:25.200 --> 00:29:26.400
They can talk to each other.
00:29:26.559 --> 00:29:35.519
You don't need a human, you know, multiple humans uh to use different AI processes to talk to these 20 different uh you know pieces of enterprise software that you use.
00:29:35.680 --> 00:29:39.279
You can have an agent that changed them all together using the model context protocol.
00:29:39.440 --> 00:29:43.440
That's where you know expert-driven loops uh you know come into play.
00:29:43.839 --> 00:29:49.519
Uh and right now, some estimates suggest that 90% of organizations are already using MCP servers.
00:29:49.759 --> 00:29:50.160
Yeah.
00:29:50.640 --> 00:30:00.319
You're probably, especially if you're uh a claw user, you're probably using NP like M uh sorry, MCP model context protocol servers without even knowing it, uh right.
00:30:00.559 --> 00:30:04.319
Especially if you're using their desktop uh application, which is really good.
00:30:04.400 --> 00:30:07.680
That means you're using some of their built-in MCP servers.
00:30:10.000 --> 00:30:19.759
So you want more proof that AI operating system is the future of work, that you need to be bringing in all your processes, all right.
00:30:19.920 --> 00:30:23.519
I'm actually gonna uh start with Microsoft Copilot Studio.
00:30:23.599 --> 00:30:27.279
Uh, they just launched their computer use in public public preview.
00:30:27.519 --> 00:30:32.240
What that means, they want everyone to be using this, right?
00:30:32.319 --> 00:30:34.160
It gives them more data, right?
00:30:34.240 --> 00:30:36.799
Uh, but they have a great computer use agent.
00:30:36.960 --> 00:30:44.079
So all the things that us humans do, we've been training models for the last probably two years on a lot of this data.
00:30:44.640 --> 00:30:47.920
And they're gonna get even better at using computers, which is wait, what?
00:30:48.000 --> 00:30:49.680
That's what we do, right?
00:30:49.920 --> 00:30:57.599
The biggest companies in the world are creating uh specific models to go and do the work that we do.
00:30:58.000 --> 00:31:11.599
Navigate between, you know, applications on the desktop, files uh on your local machine, uh, you know, terminal on your local machine, uh, AI models, the web, right?
00:31:12.640 --> 00:31:13.279
That's there.
00:31:13.440 --> 00:31:18.640
That's what Cloud Cowork uh does that was just released uh last week, right?
00:31:19.039 --> 00:31:36.160
It's a non-technical version of Cloud Code that allows you to navigate and orchestrate files on your local uh machine, a virtual machine, virtual terminal, um as well as big websites, right?
00:31:36.319 --> 00:31:37.839
It brings that all together.
00:31:38.480 --> 00:31:46.319
But I will say, I think the biggest nod to the AI operating system is what ChatGPT is doing.
00:31:47.279 --> 00:31:58.400
I don't think a lot of people are talking about it because they're comparing it to the iOS App Store, but their ChatGPT apps are all the proof you need that the internet is going to die.
00:31:58.640 --> 00:32:05.519
And the line between AI and the internet is going to blend, right?
00:32:06.079 --> 00:32:09.839
So what tools do you use on an ongoing basis?
00:32:09.920 --> 00:32:11.440
I'm gonna I'm gonna name a couple here.
00:32:11.680 --> 00:32:12.160
Ready?
00:32:14.079 --> 00:32:27.759
Photoshop, Canva, Agent Force, Asana, Box, Clay, Dropbox, GitHub, Google Drive, Gmail, HubSpot, Monday.com, Lovable, Ramp, Pipe Drive, Replit, Notion, SharePoint, Teams, Slack, Stripe, Zoom.
00:32:28.079 --> 00:32:33.759
Chances are those are pretty much a lot of people's tech stacks, right?
00:32:33.839 --> 00:32:38.240
I probably covered at least half of your tech stack, maybe two-thirds, maybe that's all the tools you use.
00:32:38.319 --> 00:32:39.680
And that's not everything.
00:32:40.079 --> 00:32:40.960
What are those?
00:32:41.119 --> 00:32:45.599
Those are apps that are available today in ChatGPT.
00:32:46.400 --> 00:32:47.920
And let me tell you why.
00:32:48.160 --> 00:33:02.400
This is a huge nod and a big reason why you need to get your team working in an AI operating system and bringing all of your day-to-day processes into, as an example, something like ChatGPT business or ChatGPT Enterprise.
00:33:02.640 --> 00:33:05.039
Those apps, okay.
00:33:05.519 --> 00:33:15.759
Yes, you can technically use ChatGPT to um interact with a graphical interface, but it also brings in all of your data, right?
00:33:15.920 --> 00:33:31.599
So as an example, uh, all of your data if you use Dropbox or Box or Google Drive, uh right, all of your company's CRM data, if you're using HubSpot or PipeDrive for your CRM or Monday.com, there's also click up on there.
00:33:31.839 --> 00:33:38.000
All of your, you know, your Zoom meetings, your Slack messages, your Teams messages.
00:33:39.599 --> 00:33:50.319
All of that is now can be remembered and can be shared in the same context window by multiple people of your organization, by an agent.
00:33:52.000 --> 00:33:53.519
You get where we're going here?
00:33:53.759 --> 00:33:57.680
You don't have to go into those 10 different tools anymore, right?
00:33:58.160 --> 00:34:01.200
And context switch or 10 different people.
00:34:01.359 --> 00:34:06.960
You know, there you have all these broken lines of communication, whether it's yourself, oh my gosh, I forgot what I did yesterday.
00:34:07.039 --> 00:34:11.599
Yeah, I was using an AI tool inside, you know, Dropbox or inside Google Drive.
00:34:11.760 --> 00:34:13.840
And well, I forgot now, right?
00:34:14.159 --> 00:34:17.920
It's because when you're going on all of these separate runs, right?
00:34:18.079 --> 00:34:24.400
When you're using, you know, 10 different tabs, five different apps, you're losing all of the context.
00:34:24.800 --> 00:34:29.039
When you're bringing it inside of ChatGPT or something like that, it has it all.
00:34:29.519 --> 00:34:30.159
Right.
00:34:30.480 --> 00:34:41.360
And the the ability to triangulate all of the different uh opportunities within there, that is the real promise of an AI operating system.
00:34:41.440 --> 00:34:46.800
It it sees the things that you see and it can compound your impact.
00:34:47.199 --> 00:34:55.039
But the biggest thing is it's going to see what you can't see because it has access to all of that data in real time, dynamic data.
00:34:55.119 --> 00:34:58.880
And an agent can just go be working for you around the clock.
00:35:00.239 --> 00:35:08.960
And even if there's a certain piece of software and you're like, yeah, our team, you know, half of our team, this is where we work, right?
00:35:09.119 --> 00:35:10.719
There's never gonna be AI in this.
00:35:10.800 --> 00:35:11.679
Yeah, there is, right?
00:35:11.760 --> 00:35:18.800
A Gartner study said that 40% of enterprise apps are gonna have agents by this year.
00:35:19.039 --> 00:35:23.519
So essentially, all enterprise software is gonna be agent-led, anyways, right?
00:35:23.760 --> 00:35:34.000
So don't just think, oh, well, you know, right now there's not an AI present, you know, our you know, piece of software that we use, you know, you know, whatever it is, you know, no, it's not gonna be touched.
00:35:34.079 --> 00:35:35.039
Yes, it is.
00:35:35.360 --> 00:35:37.519
It's everything's going agency.
00:35:37.679 --> 00:35:47.679
Not only that, but with the MCP protocol and some of these other protocols, at that point, that really takes away the traditional human work, right?
00:35:47.840 --> 00:35:57.039
And that turns us more into orchestrator, uh, orchestrators, tastemakers, and experts who should be driving these loops.
00:35:58.480 --> 00:36:11.440
So choosing your AI platform is now probably more strategic than choosing your actual operating system or saying we're a Microsoft organization or we're a Google organization.
00:36:11.840 --> 00:36:19.039
It is huge, but you have to make sure that you understand the data side as well, the governance, right?
00:36:19.199 --> 00:36:22.320
You have to make sure you understand what controls you have.
00:36:22.400 --> 00:36:41.360
Uh, you know, human approval for sensitive actions, uh, audit logs, how you can uh tackle prompt injections, um, you know, making sure that your data is integrated in the right way and still making sure that whatever systems you're using, that you still adhere to the uh the permissions that you've set in those tools.
00:36:41.519 --> 00:36:43.920
And also portability, right?
00:36:44.079 --> 00:36:46.639
Are you gonna rely on just MCP?
00:36:46.719 --> 00:36:48.960
Are you gonna build modularly, right?
00:36:49.039 --> 00:37:00.480
Because if all of your workflows depend on one single AI and you haven't been testing them and scoping them in any others, if that model has a pretty big change, right?
00:37:00.639 --> 00:37:07.599
As an example, going from a uh GPT-4.0 to a GPT-5, and maybe things don't work like they used to, right?
00:37:07.760 --> 00:37:12.800
Especially when we talk about uh agentic models that are doing a lot of the work by themselves under the hood.
00:37:12.960 --> 00:37:15.119
What if their tooling really changes?
00:37:16.079 --> 00:37:19.039
Then switching becomes nearly impossible.
00:37:19.360 --> 00:37:26.719
So not only do you have to be smart in how you set it up, but you also should always be running backup and building modularly.
00:37:26.800 --> 00:37:37.760
And not saying you need to be working in multiple AI operating systems because I don't think that's the best way, but you should have at least a small set of your team always looking at what's our backup plan, right?
00:37:37.920 --> 00:37:41.199
Because people are always like, okay, well, with all this extra time, what should humans be doing?
00:37:41.280 --> 00:37:46.239
Well, hopefully focusing on higher value work, more creative work, more strategic work.
00:37:46.400 --> 00:37:50.639
But you need people like me on your team that are keeping up with everything every single day.
00:37:50.800 --> 00:37:54.639
You need people who are creating backups and redundancies, right?
00:37:54.719 --> 00:37:56.159
And looking at the redundancies.
00:37:56.320 --> 00:37:58.000
That's what you should be doing.
00:37:58.320 --> 00:38:02.880
Because, like I said, human work is changing.
00:38:03.519 --> 00:38:11.519
So, to quickly recap, the line between large language models and the internet is going out the window.
00:38:11.679 --> 00:38:22.880
So, in the same way that all of our work now happens on the internet, and we don't necessarily make that choice to use the internet, it's just, well, when you turn your computer on, there's your outlook.
00:38:23.039 --> 00:38:24.239
It's on the internet, right?
00:38:24.400 --> 00:38:28.079
Your teams, it's like everything is powered by the internet.
00:38:28.320 --> 00:38:31.039
And that's where we're headed with AI.
00:38:31.280 --> 00:38:37.519
We're no longer, I'll say if you know, listen to this in January 2027, and I think you'll agree.
00:38:38.000 --> 00:38:43.599
We're no longer going to be making an active choice to go use AI like we have for the past two or three years.
00:38:43.760 --> 00:38:56.559
It is going to be there, it is going to be in our face, and more and more smart teams are going to be pushing all of their day-to-day operations into front-end large language models, the team versions, or what I like to call AI operating systems.
00:38:56.719 --> 00:38:59.360
So you need to be thinking about it now.
00:39:00.400 --> 00:39:02.800
What is your plan, right?
00:39:02.960 --> 00:39:06.159
When your competitors beat you to this, right?
00:39:06.320 --> 00:39:12.000
And they can run leaner, they can run faster, and they start gobbling up your slice of the pie.
00:39:12.159 --> 00:39:14.320
You have to start thinking about it now.
00:39:14.880 --> 00:39:25.679
The technology is there, the capabilities are there, the results and the outputs are there, but you just have to meet them there.
00:39:27.199 --> 00:39:29.519
What's keeping you from doing that?
00:39:29.679 --> 00:39:44.159
We've talked about in the first two episodes of the Start Here series, but it is making sure that your team seems your team stays up to date, educated, and you have to be able to carefully sprint as fast as you can.
00:39:44.320 --> 00:39:44.559
All right.
00:39:44.639 --> 00:39:48.320
I hope this show was helpful talking about AI as an operating system.
00:39:48.480 --> 00:40:06.480
If it was, make sure you go to thestartherseries.com and that will give you free access to our inner circle community where you can go uh catch up with all the different Start Here series, all the different uh episodes in that volume, as well as get free access to our prime prompt polish course.
00:40:06.559 --> 00:40:13.519
And you know, you'll get to network with hundreds of other uh just trailblazers in the AI space.
00:40:13.599 --> 00:40:26.559
I mean, we have leaders from Fortune 500 companies, small business, you know, CEOs, uh, you know, marketers, advertisers, anything you can think of, a bunch of great people who are pushing uh just the pace of AI in the inner circle community.
00:40:26.719 --> 00:40:27.840
So thank you for tuning in.
00:40:28.000 --> 00:40:31.119
Hope to see you back tomorrow and every day for more everyday AI.
00:40:31.199 --> 00:40:32.079
Thanks, y'all.
