The Future of Browser Automation and AI Agents (Nathan Ricks)

Full transcript

Welcome back to the podcast guys. Today we are joined by Nathan Ricks, a bisops lead at Annen. Nathan, welcome to the podcast. Thanks for having me. To start, could you share a quick version of your background and what pulled you into an? >> Yeah, so my background has been primarily in finance. I studied finance at Brigham Y Young University. I worked in investment banking with Goldman Sachs uh for a period of time as an analyst. I then worked in venture capital. But for a long time, I've known that I would be led into the world of entrepreneurship and in startups and especially the world of tech. That's the where I spent most of my time when I was at Goldman and working in venture capital as well. So for the last few years, I've been working at a couple different startups. I also had my own company that I ran for three years that I recently sold and now I am leading business operations here at Anonymity Labs, also called Anon. And if you had 60 seconds to explain what Annen does, what would you say? >> Yeah, we are currently in a pivot.

Historically, our company has been based around a GETIC technologies, especially as it's applied to browser automation. So being able to take action on the internet with using browser automation to be able to do highly repeatable workflows, especially in systems that require authentication. So anything that requires a human to go in and put in a username and password, then wanting to go and take automated actions within that system, whether that's reading data, writing data, our agents are able to go and do that. So basically a system that doesn't have any API, we can be the unofficial API for that system. So that's the base of our technology and that's what we've been doing for the last couple of years. But right now we are looking for new areas where we want to apply that technology and we've explored a bunch of different areas from healthcare revenue cycle management to also now focusing on financial advisors and automating their back office.

So we're we're still kind of figuring out our direction currently. >> Where do you think is the most potential with this technology? >> I mean really any industry that is highly driven by business process outsourcing. If you know of the BO market, it's a multi-billion dollar industry across the world. And it's basically these kind of back office tasks that require today a bunch of humans to do these highly repeatable, highly predictable workflows, usually within applications or on the internet. Those are the type of workflows that are highly poised to be automated by artificial intelligence. And really any industry that is backend or back office heavy, there is a big opportunity to use browser automation like we've developed to automate that type of workflow. As these AI agents are actually getting more capable, the internet still runs on login and sessions and permissions. Why do you think authentication and authorization is becoming the bottleneck for agentic workflows?

Well, this is a very good question because right now there are a lot of people using tools such as OpenClaw. We use it openclaw internally and these agents want to be able to access various software systems and applications but they don't have a natural way to be able to sign up for these services because all those services were designed human first. They weren't designed to be agent first. And so there really is a big question around how do you build the authentication that is designed specifically for agents. In the past, we've had to work around it. We had to do authentication that was designed for humans. But we had to build our agents entirely for a world that was designed for humans and for human eyeballs that were looking on a screen. But I think the way the world is moving and the internet is moving is one where it'll be an agent first internet. one that is designed with agents at the top of their mind.

Everything from authentication to data retrieval to reading and writing from systems. I think the world is moving in that direction. >> When it comes to user consent, when software take actions on someone's behalf, trust becomes the product. How do you think about the user consent and the credential safety? What exactly do you mean by user credentials and user consent? >> Like how do you manage that the access stays secure and sane? >> It's still something that I think we're we're still trying to work out, right? I mean in the past we've used basically a one password vault and the one password has API integration so that it makes it easy to store credentials keep them safe and only give agents access to those credentials when you need them to. when we use our own like open claw agents here internally, we kind of all have a one password account that we use for these agents and we make sure that it's only when you're explicitly trying to use a password do you give them that ability.

But the other thing is this really like trying to build in more logging and observability so that you're clear as to what actions an agent is taking on your behalf uh so that you can go back and audit that information later. We're actually working to build out that infrastructure around open claw today so that you can track those sort of things. >> And uh in production when you hit two-phase authentication or capture, what has been the hardest reliability problem to solve to make these integrations dependable? Yeah, a two-factor authentication has been relatively easy if the authentication method works well for the agent. Like authentication methods that are email-based are relatively easy. Ones that require a OTP password, those ones that are generated within like one password, for example, you can store those OTP codes within one password and generate them automatically and put them into the system. ones that require hard tokens or require an authenticator app on the phone. Those ones are more difficult and we kind of need to develop workarounds for those in those instances. But I think in the past it was to prevent bots from entering in a system.

But what's even better than rather than getting to a point where you have to put in two factor authentication is to avoid being detected as a bot in the first place. And there's a lot of different techniques that you can use to be able to avoid that. But I think this is just a temporary solution really. I really think the world is moving in a way that is going to be agent first. >> What do you think is the biggest misconception that companies have around AI agents? >> I think a lot of people who are not familiar with the technology tend to think it's much more unpredictable than it really is. I think a lot of people are fearful that agents will go and run and do something completely wrong or do something rather egregious without you knowing about it or realizing it. And even though there are of course horror stories of like for example an agent like leaking certain data for the most part agents I feel are a lot more predictable than humans. So if you're thinking about whether you wanted to hire a human to do this task or an agent to do do this task, I think overall agents are are more reliable and less likely to cause mistakes than humans.

>> You mentioned risk. What do you think is the biggest risk that these agentic workflows have? >> I mean it depends on what systems you are dealing with. Uh you know it depends on what data you're dealing with, how sensitive that information is, what it's connected to. I mean, obviously when you're dealing with like personal health information and other things that are covered by HIPPA within the United States, obviously that data is pretty sensitive and you want to make sure that it's handled properly. I would say there's very little risk that it ends up getting leaked in an inappropriate way. But you could certainly have the case where the agent thought it did the right thing and put the information in the right place or in the right way when it actually didn't and it put it in the wrong place, in the wrong way. And it's more of an inconvenience rather than it being like a total data hazard or data breach hazard, but it's still one that needs to get better. The observability and making sure that you have the ability to know whether the agent completed the task successfully the way that you intended it is still a big question.

>> And Nate, what are you guys focused on proving this year or what's the goal for this year? There's a lot of small things that we're working on right now like different hypotheses that we have that we're wanting to prove out that we are still unsure what the real answer is like whether for example financial advisors whether the open claw technology that we've been using internally would be able to really help accelerate their workflows and help these financial adviserss automate their back offices in a way that is both helpful and compliant. That's still kind of a big open question that we're trying to figure out. But I guess at a broader level, we're always just ready ready and waiting to see where the industry of agent technology is moving and what the future of the internet looks like. I think we have like a pretty big hypothesis that the future of the internet is going to be agent first and human second.

And I think this year is going to be a big proving year for that hypothesis. >> And how are you guys preparing for it? >> We're preparing for it by uh adjusting our business model to match that potential new reality. I mean there's other places where we're exploring in terms of where we want to develop our product and other products that we want to develop. We're starting to kind of move in that direction so that we can be ready to catch the ball when it lands. And in terms of your marketing strategy, are there any effective ways that you use to get new clients? >> Right now, a lot of it has been LinkedIn based uh sending out connection requests on LinkedIn and starting conversations with people on LinkedIn, which has actually been pretty effective uh in the industries that we're targeting. We have also done some cold email campaigns, but cold email has not been a very productive place for us thus far. A lot of it at this stage of development for us has been through warm introductions and first-deree connections.

And I think that's the kind of the future of go to market in a lot of ways is that it's going to be through warm connections first because there is just so much noise that cold outbound really isn't going to be getting through. >> For people listening who want to learn more about Anon or follow your work, where should they go? >> Yeah, go ahead and follow me on LinkedIn. I post quite a bit about what we're building here at Anon as well as my work with OpenClaw and other agentic frameworks. Uh you can also go to anon.com, see what we're building, see our research and our papers and the different tools that we're putting out there. I think those are the best places to kind of keep on track of what we're up to. >> Great. I will leave a link so people can check you out and thank you guys for watching. Thank you.