Agentic AI for Patents: The Future of Patent Drafting, Prosecution & IP Strategy

Hello and welcome to the IP Author podcast. I am your host, Saskia van Reinerveld. And I'm Samir Rayani, the CEO and founder of IP Author. And we're talking all things AI and IP.

So, Samir, tell me what has been going on in the world of AI and what has been happening that we can have a little chat about? Well, it's been all things Greek this week. interesting so i don't know uh why it's all greek do you have an idea like why it might be all greek oh i wonder i mean well i think maybe whenever we're all feeling a little bit insecure about ourselves and uh and civilization we like to go back to the roots of of things and the greeks are quite quite far back so maybe maybe that's what we're doing i think you're onto something because Okay, so the big one announcement was Anthropic, right?

The company that makes Cloud, the chatbot. They came out with this announcement that they have this new model. It's called Mythos, Mythos Preview. And this model is so good at coding.

It's so good that it's finding vulnerabilities in all sorts of old software that has been around forever and ever. It's so good at finding these vulnerabilities that the world is at great risk. Is it sort of like a Trojan horse? You know, maybe this is the Greek connection.

They're trying to take down the competitors, sneaking in behind the defenses. What do you think? I don't know, but they went to an extent and they created this new project Glasswing, right? Like a consortium of companies like Microsoft and Google and NVIDIA.

Most of them have some economic interest in Anthropic, by the way. So, I mean, you know, I am not so sure how real this is, but that's the big announcement that these people are trying to protect the world from this really intelligent new model. Protect the world. So they're sort of coming forth as the Avengers defending from this one.

And why? Why do you think that that's so important? They're saying that they really have... So, okay, they came out with this tool called Cloud Code, which we can talk about later.

But Cloud Code is this tool that is doing really, really well amongst the coding types, right? Like everybody who writes code is into Cloud Code. They like it. It's a real advancement in the world of coding.

and in trying to improve cloud code, they train the model to do better at coding. And one of the coding tests that you put these coding models to is to say, oh, are you able to secure an existing piece of code or are you able to write software that is very, very secure that will not be cracked or that will not be broken into? And apparently they're saying that it's gotten so good at finding these security issues that now the world is at risk. That as soon as this functionality comes out, we're in trouble.

But isn't this riding off the back of the news of some code that was leaked either two weeks ago? The irony. Oh, it didn't hit me until now. But you are right.

These are the same... I can't swear on a podcast. But these are the same people who basically are saying that their super intelligent mythos is so good at finding and preventing sort of vulnerabilities that they will save the world. But that same tool basically released all of that Lord Code code into the wild just a week ago.

Yes. and then wasn't there a kid who then went and rewrote it and tell us the story because you were telling me that was last week though that was like a lifetime ago but it was a fun story so apparently this cloud code this super intelligent tool somehow ended up and super secure also apparently but it ended up inadvertently releasing all of its code into the wild some people noticed obviously and they let the word out on the internet immediately a whole bunch of people downloaded the code including this University of British Columbia student who was running a hackathon for people in Korea.

He's in the Bay Area. He was running the hackathon in Korea. So he was up all night. So he saw this plot code sitting there and he just decided to give it to another coding agent called Codex from OpenAI.

And it said, can you please translate it from TypeScript, which is a language Cloud Code was written on, into Python so that I won't be infringing on Cloud Code's copyright. And so it was rewritten in Python and then it was rewritten in Rust and then it was remixed 20 times over. And now everybody knows how Cloud Code works and all the clever decisions they made. So that was last week.

And now this week, they're saying that their platform is this super intelligent being. I think you've found a problem. Yeah, yeah. The ventures of AI and the AI industry.

But now, okay, talking about Claude Code, you've been getting quite into Claude Code and then also using OpenClaw and creating all of these different kinds of agents. and now kind of taking AI to the next level. Can you tell me a little bit about that and how you're using it? Absolutely.

So, I mean, let's leave out all the hype. Let's leave out all the shenanigans of these companies. Cloud code is genuinely good. It's a genuine improvement on whatever we were doing on the coding front up until recently, right?

So in the past, we were writing code, like we would sit down and write the code and create lots of different individual programs that will all work together, et cetera. Then came large language models that will help us write individual pieces of code better. So if you take one piece of code and give it to the large language model and say, can you fix it? It'll fix it or it'll improve it.

Give it back to you. You can use it, right? Cloud code is more orchestrated. You tell it like, hey, I want to build an application to track all my customers or something like that who come to my shop or who come to my website.

It'll go and write a front end. It'll have an agent writing the back end. It'll have an agent creating the database. Then all of those agents will make mistakes.

They'll all talk to each other. They'll fix the mistakes. And then they'll give you something that's working. And it's really, really exciting.

It's absolutely fantastic to watch when it works. so that's what we've been sort of working with and do you think that you have to be somebody who understands coding to be able to take advantage of this so there is the open claw which is the other tool which is not as coding focused where this guy peter steinberger has made this tool that will actually has people and people have written all sorts of agents that will go and you know purchase tickets for you etc So you don't need to know coding to use OpenClaw.

You don't even need to know coding to use Clawed code. But you need to be able to use terminal GUIs. They look ugly, right? So, you know, it's like your old DOS terminal, MS-DOS terminal.

What is a terminal? I can't picture this at all. So, yeah. So it's like if you see some movie, right?

Like, and in the movie, they will have like this screen with a lot of text just going up, up, up, right? Like, you know, somebody's typing furiously and the text goes up, up, up. That's how we used to talk to computers before. That's how coders talk to the computers.

They mostly communicate in text. And large language models are good with text. And therefore, they found out that a terminal GUI user interface is the best way or the fastest way to bring something cool to the market. That's ironic being that we've become so advanced.

Let's go back to how it was done right at the beginning and make you feel like you're back to the old coding days, but actually in a completely different level. I have a question for you. Okay. So we were talking about the whole hype engine, right?

And it reminds me so much of the world that you come from, like the world of movies and theater and all. Every time a new model comes out, it's about the same level. The naming convention is very similar to Marvel movies now, if you can tell. Then the hype engine is on overdrive the same way.

Probably their budgets are very similar of course you know the hit rate is about the same as you know Hollywood right Like it not Pixar hit rate It more like you know the hit rate of Marvel or? Well, Marvel's hit rate is, I think, better than that of like, let's say, OpenAI or something like that, right? So how do you deal with this? Like, you know, It's interesting because, you know, I think there's also there has been a bit of a plateau that has been reached in the entertainment industry in terms of storytelling.

And, you know, you see how there's a lot of rinse and repeat happening, particularly where I think Hollywood has got the idea of adapting books all the time. And so there's less original series or original things that are being created. But it's always kind of looking at, OK, well, what's been successful and let's just recreate that in different formats. and it begs the question of, you know, can we really tell original stories or make original things or have we oversaturated ourselves with really good work and we just have to keep reinventing that work and finding new ways of presenting it?

You know, of course, you have something like my background coming from Shakespeare. Shakespeare has lasted years and years and years, hundreds of years and we just tell it again and again because the story is so good. Now, I suppose this is where I could throw that back at you in terms of your evaluation of these models. Do we think that it's kind of like iPhones?

You know, you kind of get to a model where it's really good. It does everything that it's, you know, it's been asked to do, that it needs to do. It can't really get any better except for a slightly cooler design or maybe a slightly more efficient camera. Do you think that that's where the large language models have reached.

You know, they kind of, they've got to their Shakespeare point where they've told their story, their good story has been told, and now it's just about how many different iterations you can get of that story. What do you think? Really, I mean, really good question in terms of where this is going. I personally believe that in spite of all of those claims, I think the large language model technology itself has reached its zenith for this iteration.

I think the incremental improvements that we are getting are going to be substantial. They're going to be very beneficial. There's a lot of what they call capability overhang. Man, this hype machine creates such good, like so many good terms, right?

Like capability overhang, what they mean. What does that mean? Like, you know, they say that it has capabilities that we haven't exploited fully yet, right? Like things that we may have some, there may be some advantages like cloud code, right?

They used exactly the same model as they had before, but they wrote these agents very cleverly. They used memory and all that so cleverly that they were able to eke out some really substantial improvements in terms of, you know, to make a developer's life better. Okay. But you think that it's kind of, it's reached that moment and now it's more about any kind of development is going to be a sideways or a sort of lateral development instead of a peak or reaching.

But then what technology is going to be the next exponential rising kind of technology? No, but I mean, let's take a step back. You gave the iPhone example, right? It took a while for the iPhone to plateau, but once it plateaued, right, only then, and once it reached a particular point, only then did we get some of the new age applications like Uber and Instacart and DoorDash and all of those came only once the iPhone reached a certain level of maturity.

Okay. So the cloud code is kind of the first of the Ubers if you will, built on top of the large language model technology. The fact that the large language model isn't improving at the exponential rate doesn't mean that people have creatively utilized them at all to their fullest extent. There's just a lot of different ways in which we are going to...

Like, for example, I was showing, we put all of our IP author tools and made them available to this agentic architecture with cloud code. And people are able to do amazing things. Like, you know, you can remix all of our search and claim charting capabilities and say, oh, do some white space analysis for me and tell me where I should invent new patents. And then when you give the ideas, you can say, okay go and draft some patent applications and give me some claims and so on And all of the IP author tools that we have currently are being orchestrated by Cloud Code And it works beautifully.

Okay. So these sorts of, so what's happening is we've been given this technology and it's reached a certain point. And now it's for people like IP author to come and take advantage of it and see how it can be incorporated and see how it can be used as a sort of springboard to do all sorts of other things. So the large language models kind of being the glorified search engine chatbots that they've been, but actually it's more about other companies then leveraging it within their own kind of technology to create something new and then to kind of create that exponential development.

Is that? Yeah. The AI ecosystem, right? Like the mobile ecosystem, you know, today consists of all sorts of things, right?

Like people like TikTok. Can you imagine TikTok without the iPhone? No, right? It wouldn't have existed.

But we don't have the TikTok of the AI world yet. It's coming. It is coming and it's going to be even more addictive. But do you think, so kind of going back to the beginning of our conversation, do you think that there will be a feeling within the AI companies to kind of protect and to stop other people taking advantage of their technology and using it to build something else?

Or do you think that actually the community is very open-minded and thrilled for people to build and take advantage of it? What do you think is the attitude? Is it the Trojan horse? Or are they excited and are they developing it?

Okay. Two stories from today, like from the news today, that are very funny, right? So the American technology platforms, the American large language model platforms, basically Google, Anthropic, and OpenAI, they got together and they went to the federal government. I don't know who they went to, but they are announcing that they're going to try and prevent the Chinese companies who supposedly train their models using these US models, right?

So they're like, hey, you're just distilling our models and stealing our models. Okay, great. So that's one story. The second story is that this new, again, Greek-named Muse Spark model from Facebook that was announced this morning, that was built on top of, or that used Alibaba's QN model as its primary sort of building.

How interesting. Okay. So, I think the world, everything is going in circles, right? Everybody is in this tight loop.

Nobody has a monopoly on superintelligence or intelligence or anything like that. So I don't know what to make of it. It's almost the Wild West of AI technologies at the moment. Okay.

Well, to close us off, why don't you tell us what's the most exciting thing that you've got an AI agent doing for you at the moment? Yeah, I have an AI agent right now working on improving our office action response. So one of the things these AI agents can do is they can work for longer and longer time periods now. So you can set them up, you know, before you go to bed and they will keep churning until you wake up in the morning and they'll tell you they did something.

Was that something useful? Well, you know, about three times out of 10, they do something useful. So last night I had them working on trying to improve our office action response, right? Like, you know, today, IP authors' office action response is already very good.

Like, you know, it would suggest some strategies. Most of the strategies are very sensible. Most of the customers seem to like them and they pick one of the existing strategies we recommend. But we want to get better.

So we have given these agents all of the data that we have, the prosecution data, the patent data, and all of the insights that we have developed over time in terms of what our subject matter experts have given. And he said, go figure out how to make the office action response better. And that's what the agent has been working on over the last several hours. Another agent is helping me with my travel bookings.

Fantastic. Fantastic. Well, next time, maybe you can help me to set up my own agents and I can have my little army of assistants working whilst I sleep and for me to wake up and see wonderful, great things done. Three times out of ten.

Three times out of ten. Indeed, indeed. Okay. Thank you very much.

Thank you.