Agentic AI in the Boardroom

Welcome to Applied AI Australia. I'm Ramon Rodriguez. This show cuts through the noise and delivers only on what matters, growth, margins, and time. Each week, I translate AI into practical outcomes leaders can actually use. Now, OpenAI just released their browser last week. That morning, I woke up to multiple messages from Australian executives asking me the same question. Ramon, should we be using this browser? Should we deploy it? And whilst that may seem like the right decision, especially for someone who is pro AI as me, and my response to them was, you're asking me the wrong question. What you need to be asking yourself is, who owns the decision your browser makes when it's authenticated in your profile and making decisions on your behalf? Now, none of them had the answer, and I don't think many people do as of now. Let's talk about Agentic AI browsers. Okay, so what are these agentic browsers that we're all speaking about? Comet browser came out, which was Perplexity's product and when I saw that, it just blew my socks off. And I've been experimenting with it and using it for the best part of 6 months. I was one of the first people in Australia to get access to it. So pretty much daily in my workflows. But the key thing that I will reiterate is I've not used it in a workplace setting, in an enterprise setting with controlled data due to some of the reasons that we will cover today. So the key differences with those browsers that you've been using is you control everything, the clicks, the initiations, the web pages, what you do. You've got your hands on the steering wheel 24/7. But with a Gentic browsers, you give it a goal and it figures out how best to get there. We'll go through it in this video shortly. When you see this thing take control of your browser, it's crazy. It's cool. It's fascinating and it's extremely powerful. But the risks are real. And the reason I'm making this video is because I've been using this for the best part of 6 months. And now that it's been rolled out at scale to enterprises, I need to make sure everyone understands what this is for users, but importantly for boards, CFOs, and CEOs. If you don't know what this is, there is a high chance you could be having a breach or a cyber security incident come up. Because the hardest thing about this is even if you know it, it's extremely hard to even know which employees are using it. Everything that you are reading, everything that you are consuming, it has access to. Okay? So, whatever you trust your employees with or whatever you trust yourselves with, you're essentially saying, "I trust this browser with." So, it holds context. It is going to start to understand you, which is amazing. It's great. But it's equally going to start to understand your work, your information, and probably some confidential content and context that you probably don't want getting out. One thing I will say now using it after 6 months, at the start when I got it, I asked it to do some work and go for a coffee and uh come back and my work was done or my tasks were done. And for some things, it actually did that. Social media, it did got really good at that. Research, it got really good at that. things that are super simple that I can't get wrong, it was amazing at, but multistep, it's not there. And even when it did get there, it took it six or seven times longer than I would have done it at my own time. But the thing that we all need to consider and remember with that comment is we're in the infancy. It's just come out. I know there's a lot bigger and better things going on in the world of AI. I think the whole browser thing is actually quite stupid now 6 months in but it is blowing up and it presents opportunities but it presents risks. Now if we talk about who's using it going back to my point within one week Atlas came out 28% of enterprises noted that their organization had downloaded 28%. I'm not saying consumers and the general public because that's going to be significantly higher. But I'm talking today to CEOs, executives, board, CFOs. Now, the key thing is there was no formal roll out, no IT approval, even people that are heavily into this stuff like me. No one knew OpenAI was going to drop this. No one. Okay? But we all know about the usage and the database Open AI has. All of you, if you go into chat GPT in the top lefthand corner, you're going to be getting a prompt to use Atlas and it's going to be asking you as well to make it your default browser and you'll actually get some credits and premiums if you do that. So, it's really encouraging users to do that. So, OpenAI is obviously getting into this browser war. So, you bet your bottom dollar within your organization right now, your cyber security and data privacy laws are getting completely breached. Not because they're trying to do anything malicious, but obviously they don't know. This technology is so new, coming up with guard rails to actually monitor it is extremely hot and as of now there's no real single tool or way to see if an employee is actually using it. So it's absolute governance nightmare. And the thing about this is I'll call it the accessibility paradox to build agents. Agentic AI. As much as we make it sound simple, and I'm probably guilty of that, I make it sound simpler than it is to get anything going. It takes a while building an agent and building one that's actually good and worth your time. It is right now it is still pretty complicated to execute. But the thing about these agentic browsers are if your employee can use Google, they can use aic AI, they can use these browsers. One of the things I did recently is I need to book a trip up to Sydney. So I asked it to find me cheaper flights, which it did. It saved me $340. It fully cancelled my trip to Sydney and it rebooked it at a better price, better rate, better seats. You bet. It did a great job. Now the thing that it didn't take into consideration is my calendar. So whilst I got cheaper flights and a better flight up to Sydney, I was 4 hours late to a meeting or I'm going to be 4 hours late to a meeting cuz it didn't take that full context into consideration. So the technology works brilliantly. There's no doubt about that. But if it still can't get my travel right and just looking at my calendar, there's no way in hell I'm trusting with my financial data system, CRM or procurement. Now, if we go into the risks in detail, so the first thing I want to talk about is credentials. Whatever you trust your staff with, trust yourself with, where it be your banking data, your financial modeling, your spreadsheets, your CRM, your human resource management system, if you are able to log into it with a password, chatbt, perplexity, comment, whatever it is, can have access to it. Whatever you are signed into in a browser like you would be in Google Chrome, wherever you access your documents, when you access it through one of these new browsers, it has access to everything you have access to. The first thing you're going to notice is I'm now in my chat PI account. I'm going to authorize my Gmail account so we can access my confidential information and do some pretty amazing things with it. Okay, but I don't want to do that. So, I'm not going to hit approve and go through the authorization. Instead, I'm now going to go into Gmail, and I want to use my Gmail account like I would use any other browser, but see how comma browser can help me move faster and surface some strong information for my board meeting and acquisition that I'm currently working on. Okay, so what I've written there is surface the context to my unread emails. Find me growth and risk opportunities that I can use for my board meeting. I need some really powerful insights and learnings. So, as you can see, it's going through all my emails. It's analyzing my unread information, seeing some pretty interesting stuff. It's seeing a deferred capex, valuation, market metrics, equity, DSCR dips. It's looking at a scenario analysis. Okay. So, it's looked at the deal metrics. So 1.175 billion acquisition 11.9xida forecast 245 million IBIDA margin 40% equity returns 16.8% 28% okay it's found some risks subtle exposures there's energy pricing volatility but if you see the other side of the coin it's also pretty risky so we've seen the power of using it in your Gmail with multiple emails so what I'm going to ask it to do is go through my board paper give me growth and risk opportunities give me three points to raise an investment call and impress me when I come back I need this done hit enter and as you can see it's gone through my papers Now fast forwarded a little bit. It's taken under two minutes. Okay. So it's noticed margin and revenue expansion, monetization opportunities, yet risk across customer concentration and contract clauses. The last thing I need to do is I need to go through the financials in this Google doc. We've got the six standard tabs that most of you are probably familiar with. We've got a summary. We've got forecast. We've got some evaluation information. We've got some scenario planning KPIs and some risks and assumptions. This is going to take it's not a 5 or 10 minute job. It could take me a few hours to go through this in a level of detail that I'm going to be happy with. So, what I want comments browser to do is to help me prepare and surface some risks and opportunities. I'm also asking it to step in as my personal CFO and analyze risks and opportunities and a clear framework I can take to my investors this afternoon. So, I'm going to pop in this prompt and ask it to take over. As you can see, it's thinking. When it goes blue, that's when it is taking over the screen and I step back. Now, this is going to take 5 to 10 minutes, I would presume. So, I'm going to put it on fast forward and let's see what it comes back with. Okay, so it's going through the pie chart, going through financial information, key assumptions, revenue forecast, power cost inflation. Wow. Ebida, net income, free cash flow, 10ear performance metrics. It's now worked through four tabs. It's going through the KPIs, interest cost ratio, the assumptions and risks. Now, this is probably taken about 10 minutes cuz we've got it on fast forward. And it is nearly done. It's got analysis and it's nearly ready to present some clear talking points and framework for my afternoon meeting. Now if you think about all those tasks and activities that you've got that are really time consuming if you think about how you can leverage comet browser as your agent in your daily workflows the opportunity that presents so there you go on the right hand side it's pulled out all the information it's given me a framework okay and there you go now the last thing I wanted to do is we've gone through Gmail we've gone through board papers we've gone through detail financial M&A what I want to see is what is the total picture how do I connect the dots between the different threads the different points of information and evidence and give me a concise picture so it's given me first highle deal structure risks the high leverage the deferred vendor note and the closing condition sensitivities bottom line talking points bank gone through all of that now this whole exercise has taken maybe 25 minutes. This would be half a morning's job at least. To put it in perspective, to get access to this information, previously you would have to go into settings, go into configurations, maybe go through some authentication, definitely some secondary approvals needed. So, whatever you see, it sees. Whatever you read, it reads. So it creates a whole accountability problem that is pretty hard to reverse from when it goes in and does something silly like it did with my flights. Now that was at a minor personal scale. But when it goes out and does something really bad like it breaches a law or sends something out of confidential information or there is a leak, who's accountable? Is it AI company? Good luck with that. the agent, the CFO who implemented it and authorized it. Is it open AI? Who is it? Because if you think about this from an Australian governance perspective, most Australian firms, 42% I believe it is, don't have AI governance procedures in place. And even if you did, there's not really much governance really available for a browser like Comet or uh Atlas. So there is no real AI playbook or guidelines as of such that is going to be able to control or factor for agentic AI because the reality here is so much of stuff that we're governing is yesterday's information and what's coming out tomorrow we don't always know. So it's very hard to get a lid on this thing. The other thing is prompt injections. So what it is is you're on the internet, you're surfing, you're looking at the news, digging into some competitive research, looking at baking, whatever it is. When you are on a normal browser, if behind, you know, the website that you're looking at had malicious code saying, steal this person's email address and try and get their banking data. It's going to be pretty hard to get. You might have to download something. There's going to be a few steps you have to take. all the things that cyber security says 101 not to do. But with an agentic browser like Comet, Atlas, because it is prompted to take action, when it reads the code, it will take it on face value. So, it's very easy for a hacker or a cyber security villain to have some code behind the back end of the website you're looking at unintentionally where it may say, "Give me that person's email address and any other confidential data with it." And it's that easy to steal your data when you use an Aentic app browser through a prompt injection. Now, when any of these browsers have your information, the other thing you need to think about, even if there's no prompt injections, even if everything's working aokay, largely a lot of these servers are located overseas. So, the question is, are you okay with your Australian confidential data going to serve in another country? Does it impact the Australian Privacy Act? because if you're moving data from Australia to the US without explicit consent, there could be a few problems that come up along the way. Now, I want to give you some guardrails and rules to ensure the browser that your employees are likely using or even your using don't go off track and there's only one. Right now, as of what's the date? It is the 1st of November. I would not be using the Gentic browser in an enterprise setting at all unless it's sandboxed in a particular environment, but that gets pretty complicated quickly because there is a lack of clear guard rails when and how to use these tools effectively and policy and organization vulnerabilities aren't worth it for the potential upside. Now remember, you can't control the speed of change, but you can control how quick you adapt. Thank you. My name's Ramon Rodriguez. This is Applied AI. Thanks for listening and I'll catch you soon.