Ep 851: From AI Chatbot to Autonomous Coworkers: How Consumer AI Has Changed and What's Next (Start Here Series Vol 10)

Welcome to the Everyday AI Podcast. My name is Jordan Wilson, and for the past 3 and a half years, we've put out more than 800 episodes. Yet one of the most common questions I get, I didn't really have an answer for. Where do I start on the Everyday AI Podcast? And that's why we started the Start Here series. And with fall now back in full swing, the Everyday AI Podcast is going back to school and playing back the entire Start Here series. From front to back. We've hit pause on our normal Monday to Friday programming to run back our most popular series ever for the next 30 days. We made the Start Here series for beginners and AI champions alike. So whether you're just trying to get a grasp on large language models or grappling with the best coding harness for multi-agentic workflows, the Start Here series covers it all. Plain language, no jargon, and easy to follow along each day. So make sure to subscribe to the podcast and check back each day for new insights, day by day. The series is a culmination of spending more than 10,000 hours covering generative AI over the past 3 and a half years. So you don't want to miss a single episode of the Start Here series. Let's get into it.

3 and a half years ago, AI was kind of like a party trick. I mean, if your entire company or department was using AI in late 2022, people probably would have thought that you were crazy, or you could have easily gotten into some hot water. Oh, how the times have changed. I mean, now if your entire department isn't using AI, you're definitely in hot water. That's because there's a good chance that you, your company, or your department might be sinking if you're not. That's because the rise in prominence and capabilities of AI has been astronomical over the past 3 years. And we've gone from AI chatbots that were kind of fun to play around with to straight up autonomous coworkers that are doing the work for enterprises. So that's what we're going to be tackling on today's show, volume 10 of our Start Here series, giving you a quick overview of how AI has gone from a chatbot to an autonomous coworker and how consumer AI has changed and What's next? All right. And if you are brand new here, awesome. Great time to jump in. This is part of our Start Here series. This is our essential podcast series to both learn the AI basics and to double down on your AI knowledge. So like I said, we are on volume 10 of our Start Here series. And if you want to go listen to the entire series in order, I actually just dropped a Spotify playlist so you can do just that.

Super easy.

So make sure you go to starthearseries.com. So that's also going to give you free access to our inner circle community so you can go learn and network with other business leaders like yourselves that are learning AI. Go listen to every single, uh, Start Here series, read about it. It's all in a dedicated space there. And you'll get access to our free Prime Prompt Polish course. And hey, Talking about starting here, right? Because after 700+ episodes, I didn't have an answer for what people said, where should I start when listening to your podcast? I was like, I don't know. Well, listen to the Start Here series in order, but along the way, make sure you go listen to episodes 713 and 712. That is our 2026 AI prediction and roadmap series. All right, last episode in Start Here, we talked about agent risk, security, and AI sprawl in 2026, why AI that acts changes everything. So make Make sure you go listen to that, the last episode of our Start Here series. And now let's jump into today's, From AI Chatbot to Autonomous Coworkers. So on today's show, all right, I'm gonna go over the 5 phases of AI-enabled work and why most teams are still stuck in phase 1 or 2. I'm gonna tell you how and why every major AI company is suddenly racing to put agents on your desktop. Uh, you're gonna learn about the hidden layer that almost no one is talking about that changes how AI works for you or your company entirely. And then we're gonna end with my 5 actionable takeaways for what's next, because yeah, it's been a fast sprint from AI chatbot to autonomous coworker, and we are not done, obviously. So let's first start with a zoomed-out overview of what the heck has happened. So large language models are not new, right? They've been around for a long time, technically well before even the ChatGPT moment of November 2022. But let's start there, shall we? Okay, so that's kind of when the whole, right, generative AI and large language model phase in the corporate world kicked off was in November 2022, right? Many people, including NVIDIA, CEO Jensen Huang called that the line in the sand, right? So November 2022, ChatGPT launches. You know, it's a simple Q&A chatbot in a browser tab. All right, then let's fast forward to where we are today, February 2026. I mean, we have AI agents that can run recurring tasks on your desktop with local file access. Right? So at that point, they can use your browser, they can move files, they can upload, they can download, right? At that point, they are an actual autonomous AI coworker. So that's where we started, right? This is like one of those social media posts, like how it started, right? An AI chatbot that was a party trick that didn't really know anything to, okay, this thing runs just like a human would. It can work on your desktop. It can use your files. It can move your files, upload, download, use your browser, right? Access all your data. Oh, how the times have changed in even just the last few days and weeks, right? Again, you may be listening to this in late 2026 or early 2027. I don't know. But if you're listening to it in February, right today, February 27th, I mean, the last 48 hours alone, we've seen a ton of movement that I think signal where this is headed, right? We got Perplexity Computer, uh, Copilot scheduled tasks from Microsoft. Uh, we got remote control and scheduled tasks from Anthropic. So, you know, 3 of the big 5 players there, uh, in Perplexity, Microsoft, and Anthropic just released pretty big updates that signal toward this more scheduled autonomous desktop computer worker. All right. So here's the 5 phases. All right. So phase 1, online chatbots, right? They answer your questions in a browser tab. Phase 2, business chat. This is where, you know, answers started to come with your company's data and using apps. Phase 3, we have the agentic era, right? So this is both agentic models and multi-step execution toward an outcome, right? So more outcome-based with multiple steps. Phase 4, this is your AI coworkers, right? Delegate a goal and get finished work. And then phase 5, desktop AI agents that work with your local files, right? And these are kind of in order, but as you'll see as we talk about each phase a little bit more, technically there's some overlap, right? We're still obviously getting updates to phase 2 and phase 5 at the time. Right? It doesn't mean phase 2 is over just because phase 5 has started, right? That's not where it is. It's, it's more of kind of the official eras of when things started. And I think, you know, for the most part, I'd say phase 1 is probably phased out. Although unfortunately, I still think people use ChatGPT just as an online chatbot to get answers without, you know, using any of the agentic capabilities, without using any of the business capabilities, without using any of the more coworker-type capabilities. So I think for the most part, people have kind of, uh, phased out of that era, but not completely. And then there's the feature or the layer on top of these 5 phases that I think might be even more important in really changing how works get— how work gets done. And that's scheduling, right? So I just mentioned that in the last 48 hours we've seen literally 3 big updates just In this space. But this does live across all phases, so it's not its own phase necessarily, but right, uh, and Google and OpenAI have had this for a pretty long time, uh, right? About 6 to 12 months, a little bit longer, but it's rarely talked about until recently, right? If you go back to my, uh, original ChatGPT Tasks show, uh, and it's a huge bummer that OpenAI changed how tasks work in, uh, In ChatGPT. It's actually weird. If you're on the Pro plan, you don't even have scheduled tasks anymore, which is one of the things I use most right now. It's just Pulse. And scheduled tasks you can only really use in agent mode anyways. It's, it's rarely been talked about until recently. That's because of all the changes in the last 48 hours with Perplexity, Anthropic, and Microsoft coming in with some new offerings there. Alright. Well, now you have the answers. But if you wanna dig in a little bit deeper, now's where we're gonna do it. So let's talk a little bit about phase 1. So that started with ChatGPT launching in November of 2022, and it was obviously mega viral, right? I remember at the time, right? So my company, we had been using the GPT technology since it came out in 2020. So a lot of people overlooked that before there was ChatGPT. OpenAI put their product out there. GPT-3 in other pieces of software. So when ChatGPT came out, at least for me, and I used it in November, I'm like, this stinks, right? It's not— wasn't— it was terrible. It was worse than using the other platforms that had been out for 2 years, obviously, because they had the time to improve their platforms. But regardless, this is what put today's AI and large language models on the map. And it did technically, I think, change the course. And this might sound corny, but it's true. I think it's going to change the course and has changed the course of business history. All right, then you had a little bit later, right, about 3, 4 months later, you had Google launching Bard in March, and then also Anthropic launching Claude in March, although Claude was not generally available outside of kind of an enterprise pilot, I believe, until that summer. But at the time, this is just online chatbot era, right? Ask a question. So a couple huge reasons why in this first phase, this wasn't suited for the business world, right? Uh, one of the reasons large language models at the time did not have access to the internet, okay? And that made them extremely dangerous. And I think, depending on how you look at it, fortunately or unfortunately, it was really this phase 1 that I think is still kind of setting the tone for how a lot of people think about AI today, right? That's because hallucinations were rampant in phase 1 because number 1, uh, well, 3 reasons, I think. Number 1, you couldn't upload data. Number 2, it couldn't access the internet and models couldn't reason. So unless you were really good at prompt engineering, there's a good chance that if you were trying to use a large language model in 2022 or early 2023 for any business context, it was probably pretty bad. All right. And that's why in 2022 and 2023, RAG was all the rage, right? Retrieval augmented generation, because if you wanted trustworthy outputs from AI, AI, you had to build your own RAG pipeline because the outputs were absolutely terrible, just riddled with hallucinations because, well, the models were not capable. It was at the time, I think, more than a party trick. So no file access, no integrations, no real-world actions, right? And the models were pulling on very old data, right? At the time, those first models that came out, right now, the people training the models are legit superstars, right? Some of your heads of research are making tens of millions or hundreds of millions of dollars, right? At the time, I don't think that was the case, right? Obviously, they're very smart researchers and seasoned researchers, but just the emphasis that has been placed on getting good, clean, reliable training data and just the work that can be done at inference has completely changed. All right, so that was phase 1. Phase 2, I'd say probably started in like 2023, and it Kind of, I won't say again, it didn't end in 2025, but I'll give this phase or this era 2023 to 2025. So this is team AI, this is business chat, this is when it was born, right? So team AI that's connected to your company's data. So first, I think technically first was Microsoft Copilot in late 2023. And then also you had ChatGPT Enterprise in August 2023. And then you had Gemini for Workspace in February '24, and also Anthropic for Teams and Enterprises came out in 2024 as well. But it wasn't just going team first, right? That was a big first step. So along with that, companies found out, right, or big AI labs, they're like, okay, this is great to get, you know, 100 million consumers to check out your AI chatbot. But if we wanna make this a tool that companies are gonna buy, Right? Because at the time, I think it was Microsoft that was cleaning house with enterprise AI, right? Because when they came out in 2023, there weren't a lot of other options if you wanted to get 10,000 people to use a product, right? You weren't going to use a consumer chatbot that didn't have data. So this is really the maturation phase. And I think things like ChatGPT connectors, Claude connectors, Google Apps, right? All these things in 2025 really, I think, helps solidify this space and also cut down on hallucinations. And this is where AI could finally access your email, docs, and drives all in a dynamic way, but it really couldn't act on it, right? At least in the earlier parts of 2025. That's kind of in the next phase. But this is the beginning of, you know, AI going from fun party trick to, okay, this could work, right? Kind of the experimentation phase of 2023 and 2024 to, I think by 2025, the business world knew, I would say, I was hoping or assuming it would be in 2024, but I don't think it was maybe until early 2025, maybe late 2024 that everyone realized, oh yeah, the future of work is 100% AI and there's no way around it. I think you still had fence sitters in 2024 at the enterprise level, which is absolutely nutty to me because now I think those fence sitters are paying for it. But phase 2 really pushed it. Phase 3, I'll say, started in 2024, kind of through 2025, although it's still technically going on, right? And it's ever improving. But this is the agentic area. This is where you have AI that plans and executes, right? I think technically you could say 2023 because Microsoft had Copilot Studio at the very end of 2023, but I'll say 2024 because I think Nothing was really, you know, adapted at large until 2024. But this is when you had things like OpenAI launched Operator in January 2025, right? That was a big moment, although Operator was fairly bad and slow. But it signaled a big shift toward AI that could think and do human work, right? The big thing was being able to access a browser, which is what Operator could do. You know, it could, it was an agent that could navigate the web autonomously. Then I think you had this right thereafter, shortly thereafter, you had this push for agentic browsers, right? So different than an agent, it is a browser powered by a reasoning model. So, you know, an agentic browser, one of its main features is to agentically browse the web. So I think some tools like ChatGPT Atlas, Perplexity Comet. Google announced their agent mode, Project Mariner. They had their Gemini in Chrome. That was probably more late 2025. But the key shift here, I think, was reasoning models, right? They powered all of this. You don't have worthwhile AI agents without reasoning models. All right. Models that can think, plan, use logic like humans can.

All right.

And that's really what I think separates, you know, at least in my opinion, it was the agentic era that I think is ultimately going to be the most impactful. It wasn't ChatGPT, right? Because if we never kind of had models that could reason, or which that would mean we probably would never have AI agents, right? We would have more, human duct tape than agentic output. So it wouldn't have been worth it for any businesses to truly invest in AI over the long term if we never entered into the agentic era or reasoning models, right? So some of the models, maybe you've heard of these, they're a little older by now, but, you know, 01 Preview really kicked it all off in September of 2024. Then you had Gemini 2.0 Flash in December '24, and then Claude, Saunet 3.7, I think, was their first reasoning or hybrid model from Anthropic in February 2025. So a pretty big like 8-month period there where we kind of left the quote unquote old versions, you know, the non-thinking transformer models, right? Even though they're still transformer models, they just think and reason, right? But I like, I really say that's like the old school AI versus the new school AI because I think what agentic models can do and their capabilities, the scaffolding, the harnessing that continues to be improved, right? You can make the argument today, and I've talked with very smart people about this, right? Like the head of Microsoft Research that's been working in agents for 20 years, the head of agents at Cloudflare, right? I've had so many conversations with extremely smart people in the space that have eventually essentially agreed that, yeah, if you're using You know, uh, GPT-5.2 Pro and you have all your business data connected to it, that's an agent, right? Especially when you can schedule it and it can act autonomously. It's like, yeah, that's an agent, right? Or if you're using, you know, Gemini 3.1 Pro and scheduling things and it has access to your data, that's an agent, right? So, uh, but it really started with the reasoning models. Uh, then we have phase 4. This is AI coworkers. So where you delegate a goal and it just finishes work on its own, right? And yes, I know and understand that some of these phases start to blend together. I get it, right? But in my opinion, the big difference between the agentic era, I would say that's more model and browser-based, right? And that's kind of the foundation or the stepping stones for today's AI coworkers or agent coworkers. And this is more or less general AI agents that have a virtual computer and can access your data. So, right, you can say that's, you know, ChatGPT agent mode, you know, Google Gemini agent mode, obviously. But I'd say things like Manus, right? Huge successful launch for Manus, you know, recently acquired by Meta. Same thing with GemSpark. Right? GenSpark, another kind of general purpose agent that can browse the web, right? It can access all your data, but for the most part, you know, scheduling, a lot of these AI coworkers are more browser-based, right? But you can give access to your data and then they can go use the web. For the most part, they can, you can log even though it can be risky, right? But, you know, they have a virtual computer, a sandbox, a terminal, all these things. You know, they can run code, you know, a lot of them can run different models, uh, you know, use, uh, subagents, right? So, uh, this is kind of the, uh, the version of a virtual AI teammate that works more in the cloud. And then we also— the newest entry here, uh, you know, Perplexity Computer, uh, that was just launched, uh, that uses 19, uh, AI models. Yeah, let me know if we should do a show on Perplexity Computer. I've been thinking about it. All right. And then that leads us to phase 5. And this is kind of where we're at now. Although yes, we're still in phase 4 with AI coworkers. That's not going anywhere. We're still in the agentic area, uh, era. That's not going anywhere. But I think where we are today, right? The era that is maybe most recently started, that is desktop AI agents, right? So this is a little different and a step, uh, both in a Um, technically a more opportunistic direction, but also in a much more dangerous direction as well, right? Um, it is probably in most cases, and I think most experts would agree, somewhat safer, uh, to use a cloud, uh, based agent that maybe can't access your local files, right? Even though they can oftentimes, uh, if you grant them access, they can access your Gmail and You know, your, your OneDrive, your SharePoint, right? Your Notion, right? You can add, you know, all these different connectors and apps to the kind of virtual AI coworkers. But in phase 5, this is agents that are running on your actual computer. So they can do everything that an AI coworker can do, right? A virtual AI agent. But they can control your actual computer. All right, so I think technically you could say, oh, this is more 2026 starting, but technically I think Claude Code kicked this off in February 2025, but it really got popular, I would say, in the 4th quarter of 2025, right? So Claude Code was essentially a desktop coding program that really what it turned into. I think some of the popular use cases that really exploded this category was actually people using Quad Code, a terminal tool for developers. Well, non-developers started to use it for, well, non-technical work. And I think Anthropic realized that early on and capitalized on it, and that led to the launch of Cowork. Claude Cowork in January 2026, which I absolutely love. I'm a big Claude Cowork fan. I use Claude Code as well on the desktop version, not in the terminal, not really a terminal guy myself, right? But this is now you have in the desktop version of Claude Code and Claude Cowork. This is a desktop AI agent that can control your computer and work for hours, right? I'm never one that's intentionally trying to push desktop AI agents to go longer and longer, but I got Codex to work for 10 hours once, right? I've gotten Claude, usually is a little faster, sometimes not as thorough. I personally prefer Codex. I'm gonna talk about that here in a minute. But these are now desktop AI agents that have access to literally everything, your computer, your files, your notes, your browser, everything, right? So you can't talk about Claude Code and Claude Cowork without talking about OpenAI's Codex. All right. And I am crazy bullish on Codex. I freaking love it. If you've been listening since Codex was released, I probably have it running 8 to 10 hours a day. All right. I'm, I'm really getting my money's worth on the, on the pro plan there. And I know they've had like double usage and I think that might be going away sometime soon, or maybe it went away sometime soon. That's why I'm hitting my limits, right? So OpenAI launched Claude, their Codex desktop app, in February, and this is more of a multi-agent command center. And the great thing here, they've had kind of skills. So Anthropic popularized skills kind of protocol. So they've had skills support, automation support, so you can schedule things, right? So If you wanted to, right, you could schedule Codex to, you know, go on a certain website, grab some information, you know, create a Word document on your computer, you know, organize your computer every day, right? So any task that you could do in your local terminal, on your local computer with your files and folders and your browser, Codex can do and it can schedule it, right? So I think this is, and also, Important to note, Claude Cowork just literally hours ago added scheduled task support as well. So I think, you know, that's something that was kind of like low-key. I don't think people are using— I don't think for the most part Codex people are using for non-technical work, and that's the majority of what I'm using it for. So that space, the desktop AI agent space that can control your browser, access all your local files, Upload, download. I mean, it's huge, right? And you can't talk about this space, the desktop AI agent, without talking about OpenCLAW, right? So OpenCLAW technically can run in a virtual environment. So you could say it's a phase 4, an AI coworker, but many people are buying OpenCLAW their own or its own computer, right? And that's all that happens. This is OpenCLAW's computer, right? And this is where, you know, it has its own phone number and its own Uh, you, you know, email, right? But that's where we're at now. The, right? This is the journey from an AI chatbot that hallucinated and was just kind of a fun party trick and no businesses would touch. Now, literally, you have companies buying dedicated computers for multiple. I've seen stories of this, right? People buying multiple dedicated computers for every single employee so they can have You know, desktop AI agents running more and more. So the hidden kind of phase that impacts it all, right? So there's our 5 phases, but the hidden one is scheduling. Alright? And we've seen, like I've already said, you know, Claude Cowork just added this. Microsoft Copilot just added this. Codex has always had this since it was announced. But I mean, OpenAI released scheduling via task more than a year ago. Although, like I said, unfortunately, it's really just in ancient mode now. Google has it, which I don't know why more people aren't using scheduled actions in Google Gemini. It's amazing. Perplexity and Grok rolled out tasks about 7, 8 months ago. OpenAI Pulse, which I'm not a fan of. If they made Pulse better, I think it would be great, right? But this is kind of the hidden layer that infiltrates now You know, phases 2 through 5, right? The ability now to schedule, right? Whether it's scheduling a desktop agent, which is crazy, right? That's why I leave my computer on now all the time because, you know, all of a sudden, oh, it's 2 AM, you know, Codex is gonna go do a 3-hour task for me, right? So this is, if you aren't paying attention to scheduling, whether it's in the business chat context, whether it's the remote Uh, you know, AI virtual coworker, or whether it's Phase 5, you have to pay attention to it. All right. Now let's just go to, let's wrap it up here. Uh, I told you I'm gonna give you what's next, and here's kind of my 5 facts and strategies for what's next. All right. Um, number 1, agents are delivering real artifacts, decks, docs, spreadsheets. Everything better than humans. I think this is probably what num— one of the most overlooked facts and aspects of large language models. And this is models by default. People don't understand that you can literally, if you know what you're doing, if you give, you know, Claude, or I'd say right now probably Anthropic and OpenAI are the leaders in this, at least on the, you know, business chatbot space. You can literally go do your research, contextualize and personalize through your business context, and create spreadsheets and docs, decks, et cetera. All right. Number 2, desktop AI creates a security and governance challenge that most IT teams aren't remotely ready for. So be ready for agent drift and agent crash. That is gonna be a huge trend of 2026 and 2027. All right. The next piece of advice here, tomorrow's AI winners are gonna care less About what actual AI systems they're using, and they're gonna care more about and spending more time rebuilding how knowledge work works, right? I think, you know, oh, today's best model, right? It becomes a commodity, right? They're all getting similar or the same features. Yes, I still think there's always gonna be winners and ones that are slightly better. But if your team is spending more time on deciding, oh, are we gonna use Gemini or OpenAI? Oh, are we going to use, you know, Claude or Google? Right? At that point, if you're spending more time doing that than rebuilding how knowledge work works, you're behind. All right. Number 4, the AI assistant era is already over, right? The AI worker era is here, both virtually and on the desktop. And most companies haven't noticed because I think it's been swift. Right? A lot of these other rollouts, right? Even if you look at the, uh, you know, the business chat context, you could say that took 18 months, right? The, uh, AI worker and going from AI assistant to AI worker has taken 18 days. It is fast. It is here. Most companies haven't noticed. If you're listening to me, you need to pay attention. All right? And then last but not least, scheduled or proactive AI is the sleeper feature that changes the entire relationship between humans and AI, right? And it's still one of those things that's flying under the radar for now. So if you, your company, uh, your department wants to take advantage, that's where you should be spending your time on right now. All right. I hope this one was helpful. A quick journey and recap of how we got from AI chatbots to now we have autonomous AI coworkers. So if this was helpful, please go to startheerseries.com. That's gonna give you free access to our inner circle community, and it's gonna put you right into our Start Here series channel where you can listen to now all 10 volumes of our Start Here series. So thank you for tuning in. I hope to see you back tomorrow and every day for more Everyday AI. Thanks, y'all.

And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going. For a little more AI magic, visit youreverydayai.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers, and we'll see you next time!