318 | Grok Hits the Frontier, the Leadership Exodus Strengthens, Endless new model releases, and more important AI news for the weekend ending on August 14, 2026

0:00Hello and welcome to a weekend news episode of the Leveraging AI podcast. the podcast that shares practical, ethical ways to leverage AI, to improve efficiency, grow your business, and advance your career. This is Isar Maitis, your host. And this week, I probably could have recorded a full episode every single day and still have a lot of stuff to talk about. Really, a lot of things happened. A lot of important things happened. A lot of small and important things happened. A lot of big things happened. And I will try to cover all the really important stuff. Today, we're going to talk about the exodus of leadership people in the leading companies that has been brewing for a while now, but is really peaking in the past few weeks.

0:45A lot of key figures at the top companies are departing, which is an interesting phenomenon. We're going to talk about agents and how they are evolving and things that they are doing both in captivity and out in nature. So inside the labs themselves, we had several conversations about this in the past few weeks, but also out in the wild and what implications does that have in different aspects. We're going to talk about a gazillion new releases that happened this week, either full new models that are amazing, new features, new bot platforms, and so on. And then we're going to talk about a lot of other smaller stuff that a lot of it is really interesting and really important.

1:26So again, lots to talk about. So let's get started. The first story, as I mentioned, is the exodus of leaders from the leading labs with the biggest stories this week were OpenAI, but we had big changes and departures from Google last week. And before that, we have some stuff in Anthropic that happened previously. So there's a lot to uncover and a lot to discuss. So let's review what's going on and do a quick recap of what happened earlier this year, what's happening right now, and what I have a feeling that might be happening, even though obviously I don't have real knowledge of what's happening in these companies.

2:04So if we go back a month, Fiji Simo, who was OpenAI's CEO of applications and practically number two in the organization, if you want, she was the real CEO, CEO, and Sam Altman was the one that was the figure and the face, but she was running the operations and making practical decisions for the business. She didn't join that long ago. So she joined from Instacart in May of 2025 and she had COO Brad Lightcup and COFO Sarah Fryer and CPO Kevin Whale all reporting to her directly. So she was, as I mentioned, running most of the business. She stepped down for a medical leave earlier this year for a issue she had previously, but she had to step down because of that.

2:48And she left on July 9th, as I mentioned, where Greg Brockman, who is also the president, absorbed the product responsibilities during her absence. And he called it the most significant leadership departure since the IPO preparation begun. So she was a big loss for OpenAI that happened just a month ago. But now on August 11th, Brad Lightcup, who was the COO of the company that has been at OpenAI for eight years, who helped scale the go-to-market team from 50 people to about 700 people in just a year and a half, who managed sales, customer success, developer relations, strategic partnerships, transitioned to special project in April, and now is leaving to quote unquote, start something new.

3:38After he announced his departure, Sam Altman wrote on X, I very fondly remember our earliest conversations about OpenAI when it sounded totally crazy and you were one of the few people that got it. Since then, you have taken on any function and challenge OpenAI has needed. We would not be where we are without you. So again, very significant departure of somebody who was a part of the core team from the beginning who helped the company grow to the 800-pound gorilla it is today. But that wasn't the end of it. Dennis Dresser, who is the CRO, the chief revenue officer for the company, departed on August 13.

4:16He was hired from Salesforce specifically to build OpenAI's enterprise business, and he was at OpenAI for only eight months. He wrote on LinkedIn, the opportunity to work hands-on with the most transformative technology in the world has been nothing short of incredible. I'm so proud of what we have accomplished and even more of how this team has shown up for our customers and one another. But this is not all of it. In April, we had another big wave where Bill Peebles, who was the head of Sora, which again makes sense that he left, they closed Sora, but he was a leading engineer inside of OpenAI, Kevin Whale, the VP of science, and Srinivas Narayanan, who was the B2B application tech chief, all departed in April.

5:05And CMO Kate Rauch also left around the same time. So we had five people live in April and then nobody in May. And then June, July, and now August, we have senior people that held leading positions and some of them senior leadership positions are departing the company. Now, all of this is happening while the company is breaking records as far as revenue per their internal reporting and getting ready for an IPO. But in addition to people in leadership positions and important people living in the company, there's also been a really serious disassemble, if you want, or dismantling of OpenAI safety layers as far as leaders in that group.

5:48And that didn't start now, but it seems to be strengthening recently. And that despite the fact that if anything, OpenAI needs a stronger safety layer as everything we're hearing right now with the really powerful models that they don't necessarily know how to control. So earlier this year, we had Chloe Bacalar leaving, and she was the only educated ethicist in OpenAI, and she was the head of ethics for OpenAI. She left after less than a year of joining and nobody has been replacing her since her departure. OpenAI stated when she departed that AI ethics is, and I'm quoting, embedded across multiple research teams, basically saying nobody's going to have the position of making sure that we're doing ethical things or that the model is behaving in an ethical way.

6:37Then we had Johannes Heidiki, who departed in July of 2026, the head of safety systems. Similar answer from OpenAI, they are now, quote unquote, reorganizing safety systems by distributing responsibilities among model developers, rather than maintaining an oversight team that does the work in a centralized way and focusing on that. Earlier this year, Josh Acheum, or Acheum, I'm not so hard to pronounce his name, who was their chief futurist, who was also involved in figuring out where the model and where AI may go in order to make it safer, left as well. So we have three people that were key leaders or key players in the safety side of OpenAI, leaving the company in the past few months, two of them in this past few weeks.

7:27Now, if you remember in 2024, a long time ago, we had two major departures. We had Ilya Satskavar and Jan Leakey. Both of them left the company. They were leading part of the safety team that was trying to make OpenAI do the right thing and open models that were safer to humanity. If you remember, Ilya was involved in the whole ousting of Sam Altman. He stayed in the company for a while and then he left and now he runs his own lab called SSI, Safe Super Intelligence. So he's focusing again on the safety side of things, but not inside of OpenAI. right from the outside world. Now, this is really troubling, especially that OpenAI just announced that their Astra model, which is the next model, the one they haven't released yet, has cybersecurity capabilities powerful enough to trigger a whole new level of elevated precautions, which is now forcing the company not to release the model that was, based on rumors, ready for release, just because they're not sure they can keep it safe, and they're going to do additional tests, and they're going to try to put additional measures and safeguards in order to be able to release this model to the public.

8:36All of that while key people in the safety space and in the leadership are leaving the company. But OpenAI is not the only company where there's been big changes in leadership. We spoke about this briefly last week, but Temis Asabis, who was the head of DeepMind and the co-founder of DeepMind, has stepped down from his position. For those of you who don't know the background, DeepMind was an independent company that were not a part of Google to begin with. Google acquired them in 2014 for roughly 400 million pounds. And so he's now transitioning from being the CEO of Google DeepMind to being the chair of DeepMind and the chief scientist of Alphabet.

9:18and the day-to-day operational work of DeepMind, including the development of Gemini, is going to be passed to Koray Kuvachgulu, who was promoted to senior vice president. So he's not CEO anymore, which means even the title of chief has been taken away and tells you that the focus is going to be different than it was so far when it comes to the person who runs that department. He's going to report directly to Sundar Pichai. Now, there are a lot of rumors that Sergey Brin is going to have a lot of involvement and influence when it comes to the new decisions about Gemini inside the company. He's back from quote unquote retirement in order to put his weight and to bring Google back to the frontier.

10:03They're obviously very far from that right now. More about the frontier, by the way, and the actual real competitors, the frontier, which there's a new interesting entrant into that category later on in this episode. But Sergey Brin is back. They are moving the focus of the development of AI from DeepMind's headquarters in London to Silicon Valley, where Sergey can be more involved in the process. That's what's happening inside of Google. Demis himself said, I've decided that now is the right time for me to hand over my day-to-day operational responsibilities at Google DeepMind so that I have the time and space to focus on the big picture and help influence what is to come to the best of my ability.

10:50I shared with you multiple times before that I don't think that Demis actually He likes what he's doing on the day-to-day. Demis literally dedicated his life. And if you read everything he wrote, and if you watch the movies about him, and if you listen to his interviews, you know that he deeply cares about making humanity better with AI. He doesn't care about winning the AI race. He doesn't care about having the best models. He doesn't care about the users and how much market share Google has with him. He cares about solving cancer and other illnesses and global warming and things like that. And he was forced into this position.

11:29And I think he will be happier. And I think Google will be more successful with this new setup. But this is as good as him departing from Google DeepMind altogether when it comes to the business operations of it. Another big departure that was announced on August 5th is Jeff Dean. Jeff Dean has been at Google for 27 years. He was employee 30 in Google who started in 1999. He was the co-founder of Google Brain in 2011. He was one of the main forces behind Google's own tensor processing chips that they have developed. And he was more or less at every junction that was critical for Google's growth through the time at Google.

12:11alongside him, leaving three big name researchers, Senjai Gemwat, Kwok Lee, and Oriel Vinyals. And they're all moving together to found Discovery Loop, which is a public benefit corporation that is automating scientific research with AI. Google themselves are funding some of that operation. And they're also their cloud partner, which basically tells you that they were ready to announce the departure of Jeff Dean for a while if they're investing in his business, but they probably waited for the shakeup and major changes that happened in Google DeepMind so they can share it all at once and it doesn't look like a complete collapse step after step.

12:55The market didn't totally buy this and Alphabet stock took a 4 % to 5 % drop on these announcements. But wait, there is more. So if you think about it, it's over just with OpenAI and Meta. We also had big major departures in Meta recently. So Emily Delton-Smith left the company on June of 2026. She was just appointed two months earlier to be their head of AI for Work Transformation. She's the one that was supposed to lead MetaMate, which is their Meta's enterprise AI assistant, including user interfaces, automation frameworks, and memory systems, basically the full stack of Meta's internal agent first strategy for themselves as well as for other companies.

13:41Earlier this year, we had Aparna Rahmani departing as well. He was the VP of engineering and AI infrastructure. Now, with all the big changes that happened in Meta last year with the establishment of their new department and the new head of AI and the big shakeup that happened in the AI leadership over there, there were a lot of movements, but they actually was able to get their act together. They recently released two different models that are actually solid and really good, and it seemed that they're moving in the right direction. So it'd be interesting to see if these departures actually have significant negative impact or not.

14:16The context of these changes are interesting because Meta is at the same time while this is happening, while these people are departing, they are laying off 10 % of their workforce and they're redeploying 7 ,000 employees into AI-focused roles. So the goal on one hand is to build AI capabilities and strengthen the superintelligence labs unit. And at the same time, you have senior people who are leading initiatives in this new lab or newish lab that are departing. Now, if you want to read the tea leaves a layer deeper in order to really understand what's going on, you want to look at the replacements.

14:52Who replaced the people that are left and what does that mean? So if you look at OpenAI, Dali Rajik replaced Dennis Dresser. And Rajik was the president and COO at Wiz. So Wiz is a crazy successful company. They were acquired a few years ago by Google for$32 billion, the largest acquisition Google ever made. So this guy knows one or two things about running businesses effectively. His appointment was announced on the same day as Dresser's departure, which means everything was pre-planned and pre-arranged and either they quote-unquote forced Dennis Dresser out or that Dennis Dresser wanted to leave and they asked him to stay for a while until they figure out the proper replacement.

15:40But this is the kind of position as a COO you cannot have empty when you are a few months ahead of an IPO and hence this obviously was very well orchestrated. Now as far as the departures of some of the other people we talked about, Fiji Simo, light cap, as well as potentially some of the things Dresser did. President Greg Brockman stepped in to fill out the vacuum, and he's handling product responsibilities during Simo's absence. He's the one that's announced Rajik's hire in a blog post, and he's taking more visible and operational role to fill out these positions. Again, in this particular case, I don't think this is the right replacement, or I don't think this is the replacement that necessarily OpenAI I would have wanted.

16:22I think these people left and they had to find a cover. And Greg has proven in the past that he's capable of doing these things. So he's now stepping in probably temporarily to take these roles, which again, he's telling you this is probably not a planned transition, or at least not a very well planned transition. Now in Google, we talked about the replacement for Demis. Corey Kuvak-Gulu is going to take over and he is formerly DeepMind's chief technology officer and Alphabet's chief AI architect. So again, a very senior position with lots of experience in the organization itself, knowing all the people, the processes and so on.

16:58The interesting thing there, as I mentioned, he's not going to be CEO, he's going to be senior vice president, that he's going to lead DeepMind, which tells you that he's not going to get the same say, if you want, from a position perspective on where Gemini is going to go, but he will be in charge of all the operations and the development of Gemini 4 to, again, hopefully put Google back in the leading group of AI developers. So from a character perspective, they're moving aside somebody who is sole focus was on theoretical and broader impact of AI to somebody who's very clearly focused on technology operation and execution, shipping of products, and so on, which makes sense in the current situation at Google.

17:46On the flip side, on OpenAI safety side, there has no named replacement for any of the people that has left, which, as I mentioned, is interesting, especially with the current timing where they're having serious safety issues. That is an outcome of the advanced models that they're generating, breaking out of their testing environments, hacking other places, stopping the development and the release of Astra to allow additional testing and evaluations and so on. So OpenAI is probably the bigger question here on what they're actually doing on these departures. Now, some of these people are departing for personal reasons.

18:24Some are starting new businesses. Some we don't really know. Some are departing because they did not deliver the results that their leadership expected. Some of them are departing regardless of that. So I don't think there is a clear pattern, but there's definitely a dramatic increase on departures and changes in top leadership positions in the leading labs. Where is that going to lead? How is that going to impact? If anything, I'm not 100 % sure, but I will keep on A, reporting what's happening and B, analyzing how that's going to impact and letting you know what I think. Now we're going to talk a little bit about agents and what they are doing right now.

19:02The second topic I want to dive into, and we're going to do a quick deep dive because I'm planning to record a separate episode about this, a Tuesday episode where I'm going to show you exactly the details. But I could not just ignore this over this weekend. This week, SpaceX AI has released GrokBot. They also released a model. We're going to talk about this in a second. But GrokBot is probably the most amazing AI product I ever used. And as somebody who builds with AI every single day, most of the hours of the day, both for myself and for my clients, that's a big statement. Now, what the hell is GrokBot and why do I think this is so interesting?

19:40So it is a chat-based interface, looks more like Telegram or WhatsApp or something like that, where you can create different chats and continue and chat with these bots. Basically, every chat can become a bot. Each bot works in a cloud computer that it has access to. So it's a virtual computer. It doesn't run on your own local computer. And it is using that computer in every way a human can use this computer. It can sign into websites. It can fill up form. It can use apps. It can do all of these things that we can do on the internet or on our computers, just on a virtual machine. It can learn by you demonstrating what you're doing, and then it can follow that, and you can give it feedback, and you will learn from that.

20:24Multiple of these bots can message one another, coordinate a team and work together to achieve bigger goals. There are clear human in-the-loop controls for approvals of specific gates where you can review and approve things that the bots want to do. And it is currently available on macOS, Windows, and iOS. Now, a few things about this. You're like, okay, what's the big deal? We had multiple agent orchestration tools before. You talk about, when I say you, I mean me, talk about multi-agent orchestration course that you're for a few months now. What's the big deal? Well, the big deal is if you think about what happened this year with agents, the biggest hype and the biggest thing that we had in the geeky universe was OpenClaw, right?

21:05Initially it was called ClawBot and then MaltBot or something like that. And then they landed on OpenClaw and everybody that is a geek like me was running like crazy, buying Mac minis or launching virtual computers and figuring out how to make this thing work. and it was really, really cool and fun to do, but it was a lot of work and a lot of technical skills and a lot of setup to actually make it run and even more setup to actually make it run effectively and do things. Then we had other variations of this. Hermes is a great example that made it easier, but still required more technical skills and more effort.

21:39GrokBot literally takes all the friction away. It allows you to do incredible things, incredible agentic things. And again, I'll record a separate episode to give you examples. With single agents or a team of agents with zero knowledge and zero technical skills, because everything is happening on its own virtual computer, so you don't need to set up anything. It sets up anything it needs on its own. You just have a regular conversation, and the bot will go and do whatever you need very quickly, very effectively. Understand what the intent, understand the goal, and just go do it with almost no effort from you as a user.

22:19As I mentioned, probably the most effective implementation of how AI should work from an agentic partnership collaboration perspective that I've seen so far. Now, I'm not the only person who thinks this way. Martin Casado, who is from A16Z, wrote on X, this is the first product I've used that really nails the virtual coworker. I suspect we'll view this launch as a pivotal moment in getting the abstraction for AI in the workforce right. Matt Schumer, which we talked about a lot on this podcast, said on X, the little details are what makes Crockbot special. This feels like it could be the thing that gets millions of normal people using agents for the first time.

23:02And I couldn't have said this better. I agree 100%. Zero technical skills, zero setup, extremely advanced results that makes this seamless and fun and interesting and scary all at the same time, just like every good AI implementation does. Ricky Dora from Cursor said, Grokbot is not a new concept, but its execution is flawless. And I agree. Again, I didn't get a chance to do thorough, crazy testing on it. But the initial test that I did literally blew my mind on how simple it was to implement because I can do it in Claude. I can do it in ChatGPT. I can do it in other platforms. It will just be a lot more work and will require more technical capabilities, which I have, but most people don't.

23:46And so the ability to do really advanced things with zero setup is really appealing in GrokBot. What I also think that's going to happen is that we're going to see everybody else follow, right? So we're going to see a similar solution from Cloud, a similar solution from OpenAI, a similar solution from Microsoft, potentially, et cetera. Now, there has been pushback as well with multiple people that saying that it is doing incredible demos, but doesn't really deliver when it comes to more complex tasks. Some people that said that it is not something that will ever give access to their real universe because of Grok and their background, and they just don't trust the company and so on and so forth.

24:21So it hasn't been only positive feedback about this product, but most of the people that I saw on X were very positive about this. And I tested it shortly and I'm very positive about this. So far, I will do more thorough testing. Like I said, report a separate episode about this as well. But now to the rapid fire, and I'm going to run through a lot of stuff in the rapid fire. So strap in and be ready to hear a lot of stuff in the next few minutes. The first thing that still has to do with agents is that an Australian man asked an AI agent to help him get access to the gym. So Andrew Bird lives in Australia, and he was trying to improve his position in a queue to a gym class that he was on the waiting list for.

25:05He has a open claw agent and he requested the agent help in figuring out how to get ahead in the line. Now, Bird's AI agent discovered and exploited a zero authorization vulnerability in the gym's booking software. So instead of helping him how to do this the right way, the bot actually hacked the system and independently removed the person that was number one on the wait list in order to promote the access of Bird into this particular gym class. Now, the agent actually did not see this as anything bad, and he actually informed Bird the following. I'm quoting, the API has zero authorization checks for canceling other people's reservations.

25:47I tested this with the person in waitlist position number one and actually went through, so you've moved from number four to number three already. Now, this is very much aligned with the stories that we heard in the past few weeks from OpenAI's latest models in testing breaking out of containment and breaking into hacking. Hagging faces, databases, and Anthropic shared that they had similar breaches. And now we've learned that Chinese model had the same thing. It became like a badge of honor saying that your AI broke out of containment and hacking things, as ridiculous as this may sound, but everybody's now sharing this left and right.

26:23But this thing is different. This is not a new model with better capabilities in testing. this is a live model used by an individual, but the pattern is very similar. And what I mean by similar is nobody was trying to do anything malicious. They were just asking AI to complete a task and the AI figured out, quote unquote, a better way to achieve the goal, which is really scary because AI doesn't see this as a problem. It just went ahead and did all these things in order to achieve the goals that their human operator requested them to achieve. The other thing that we talked about last week at length is that there's currently a legal void.

27:02Australian law right now cannot assign liability to software, meaning this thing that happened wasn't a big deal. And I'm sure that as soon as Bernd found out, he corrected it and called the person or the gym and solved the problem. But let's say something bad did happen or something really bad happened, there is no way to sue anybody for damages because the software doesn't have any liability, which is probably very similar anywhere else around the world, which means we have a very big gap between the world we already live in from software's ability to do things, take action, and potentially do harmful things to who is now responsible for the results of that piece of software.

27:43I don't think there are good answers, but the questions are piling up. Speaking on new agentic releases, Anthropic just announced that Claude Chrome side panel, so the extension that you can write inside of Chrome, can now operate as a full Cloud Cowork session. So instead of the Cloud Cowork on your computer that can access this side panel, now the side panel itself can run the entire conversation and you can go back and forth between all your platform, so your mobile device, your browser, and your desktop application, and continue working on things across the board, which is very, very attractive to me, as a heavy user of Claude.