Presentation by Ian Schick, Co-founder and CEO of Paximal

Ian Schick, CEO of Paximal, presented Idea Clerk – an AI tool for generating patent applications. He noted that innovation is accelerating exponentially while the patent bar is shrinking, with fewer engineers entering patent law. Traditional patent applications take 20-30 hours and cost thousands. Paximal uses an “alignment” process instead of prompts, where users answer questions and review AI-generated content at key steps. The system produces complete patent applications in 2-3 hours, including figures and specifications. Idea Clerk targets startups, allowing founders to file patents same-day for $120 instead of using expensive law firms.

Full Transcript of Presentation

Roland Vogl:

Welcome, everyone, to our Codex group meeting. It is Thursday, October 23rd. We’re excited to have Ian Schick with us. H’s the co-founder and CEO of Paximal, of course, has been a friend and affiliate of Codex for many years, doing interesting work on legal document generation in the IP space in various contexts.

He is here today to talk about Idea Clerk. Really great to have you here. We’re excited to learn more. Welcome everyone. If you have questions, as always, please put them in the chat or unmute yourself. Ian, we have a total time of half an hour until 2:00. If we could sort of split that between presentation and Q&A, that would be great. With that, I’ll just turn it over to you.

Ian Schick:

All right. Great. Thanks so much. And thanks, everybody, for tuning in today. Let me get my screen shared. So hopefully everybody can see my screen. I’m going to start with just a quick slide overview of Paximal and Idea Clerk. Then I’d like to give a demo, and then we can open it up for some Q&A if we’ve got time left.

Patents at the speed of innovation. If you think about innovation that we’re experiencing today, it’s like we’ve entered a new paradigm where AI-assisted innovation is just putting things into overdrive. We’ve been experiencing exponential growth in technical innovation ever since we’ve been measuring it. That’s kind of the classic Moore’s Law. But really, anything you can measure, innovation shows some type of exponential growth.

It turns out patent filings have also shown this exponential growth ever since the first Patent Act of 1790. We’ve seen displays of exponential growth in patent filings. Just a few examples. In the first decade of the patent system in the US, we issued 229 patents. Now we issue that many every couple hours.

In 1837, the electric motor was patented. The 46 years before that, 10,000 patents were granted. In the 46 years after that, almost 300,000 patents were granted. Then the time between the one millionth and 2 millionth patent applications. So the one millionth application took 120 years. The 2 millionth application or patent only took 24 years.

Then finally, with semiconductors, through 171 years, we issued 3.3 million patents. After that date, it only took 38 years to double that number of patent filings. Now we are kind of past the elbow. We’re past the curve, and innovation is just accelerating faster and faster and faster every day.

The patent system really can’t keep up with it. There’s many things going on that are kind of affecting how we protect inventions. But traditionally, this is a slow, manual, and expensive process.

I was a patent attorney for several years, and so I know this intimately. But traditionally, preparing a patent application requires a trained professional, 20 to 30 hours of time, literally sitting at their desk and typing up this document. So they’re generally 25 to 45 pages long, and they explain how to make and use the invention in kind of a legal way. In a way that would enable one of ordinary skill to make and use the invention, but in kind of a legal parlance.

I’ve got this graph to the right here. This is interesting. It shows the number of currently active practitioners binned in a histogram based on their years of experience. You can see that the patent bar is really collapsing over the past decade or so. There’s just fewer and fewer people joining the patent bar. I think you have to be an engineer to join the patent bar. Engineers these days can make tons of money in industry. So going to law school and working in a law firm, it’s just less appealing. But what it means is we’ve got this sort of glut of highly experienced professionals in the field that are getting close to retirement, their kind of end-of-career folks, and there’s nobody to fill the shoes.

We’re on the precipice of this real drastic, even more drastic collapse in the patent bar and the ability of humans to be available to write these patent applications. But then, of course, at the same time, innovation is kind of going off the charts right now. The result is that innovation is moving much faster than ownership.

We need patents that move at the speed of innovation. We need those documents to have the same quality, the same compliance with all the requirements that exist today. We just need that orders of magnitude faster and more accessible. That’s the mission behind Paximal.

Paximal is a fully agentic solution. We are prompt-free, and instead of a prompt experience, we go through an alignment, an alignment sequence with the user. Paximal is designed for turning users. The platform is built for enterprise deployment. We produce complete documents. Again, this is not a copilot. It’s not a kind of chat experience. We’re actually creating the entire documents beginning to end. We’re also generating the figures. And I’ll show you that in a few minutes in the demo. We can customize our output so it can be customized to a law firm’s preferences or portfolio-specific requirements, and that kind of customization is possible throughout the document. So the layout of the document, what it contains, the structure of the document, even the way it looks and reads.

I mentioned that we go through an alignment step that effectively is a human in the loop. So this is an AI-driven process, but we insert the human. In this case, the patent attorney, at critical parts of the process, to make sure that we’re generating a document that’s going to be in line with what the project requires a little on our traction so far. We have been in development for a few years now in sort of a semi-stealth mode. But went to market at the beginning of this year, and since then have signed up 70+ organizations. That’s majority law firms but also many corporations.

Our drafts, in terms of project time, are up to 90%+ faster than the traditional approach. So to put that in the numbers, a 20 to 30 hour project traditionally, we can get that done in 2 to 3 hours of attorney time. Our quality is on par with leading law firms. We bake in the best practices that are accepted across the industry, and our best documents are very high quality. Our worst documents are also very high quality. So we manage with guardrails to keep a very consistent, high quality across our documents. We do this in a seamless workflow. It does not require retraining. A patent attorney doesn’t have to become expert in prompt engineering. It really kind of simulates the process that attorneys are already used to in their practice.

Excited to announce that we are extending access to early-stage startups. About a month ago, we launched a service called Idea Clerk. Idea Clerk is the same core tech as Paximal, but it’s reimagined for startups. This has already taken off in the startup scene. We’ve got over 100 startups that have signed up now in the first month. We did events at Tech Week in San Francisco, in LA. We’re here today presenting and will be at the Legal Tech Fund Summit in a few weeks as part of their startup showcase. So what we’ve been hearing back, founders love the simplicity. They love the investor-grade output that we’re providing. But most of all, they love the ability to get on file in the same day for a fraction of the cost that they would otherwise have to spend at a law firm.

If you’re familiar with Clerky for entity formation, a kind of self-serve service for startups, Idea Clerk is kind of like a Clerky for patents. It’s as if you’ve got a patent attorney in your browser.

I am going to switch over to a demo and take you through the process. Go to IdeaClerk.com, will get signed in, and it’s just as easy as clicking New Invention.

The process starts off with a few questions and the type of question that an attorney would need to know if they were working on your project. We start with the type of technology. So for today’s demo, I’m going to be using a kind of a fake invention that I made up. It is augmented reality co-counsel for patent interviews.

This is a document that I assembled that describes the invention. I just really quickly will point out our blog on Idea Clerk includes tons of articles. They’re short, quick, easy to read, but kind of answer all of the typical questions that early-stage startups tend to have, including how do I document my invention.

This is an article that I go through the process of documenting your invention in a way that would be useful for about an attorney, in a way that would be useful for Idea Clerk. Even include some sample prompts. So I use this process to generate this document that kind of lays out all of the technical aspects of the invention, including a couple of figures.

This figure is, so it’s kind of the environment in which the invention is practiced. So we’ve got an inventor talking to a patent attorney. The patent attorney is wearing AR goggles here. Then the second figure is what the attorney would be seeing through those AR goggles. So there’s the inventor, of course. Then some AR fields that show information about the invention. The idea here is that while the attorney is talking to the inventor, the conversation is being forwarded, transcribed, and the relevant information is being presented to the attorney so they can better understand the invention and ask better questions of the inventor. So with that, I will go back to the demo.

I’m going to focus on the software aspects of this system. Can do mechanical inventions. So that’s like physical objects, software inventions, and then a blend of the two, electromechanical. So like a device that’s operated by software. See that a lot in the medical device field. The entity type. We are a company, the Paximal Inc.

The invention name? This is AR co-counsel for inventor interviews. This is just kind of a short shorthand name. It won’t be the name of the patent, but just a name for the user’s reference. Okay, so let’s go ahead and get this created.

It asks for a list of inventors. The threshold question. Here’s some language from the patent office. You know how to determine who is an inventor. It all comes down to conception. Who thought of the idea? If the CEO of a little company comes up with an idea, shares that idea with engineers that build that technology, the engineers would not be considered inventors because they didn’t conceive of it. They just reduced it to practice. So I’ll put in my own name here. I’ll put in my co-founder’s name.

Click to the next window.

Documentation. Here’s where we’re asking for documentation on the invention. It could be things like system diagrams, flowcharts, schematics, wireframes, really any documentation that you have that describes the invention. This doesn’t need to be something that includes the secret sauce. It doesn’t need to be an exact recipe. We don’t need code-level detail here, but a high-level description of the invention. I will go ahead and drag that over the document that we just looked, get it uploaded, and click next.

This field is for additional information about the invention. Coming soon. Kind of the idea here is that sometimes it’s useful to hear the inventor just describe it in their own words. Kind of go through the main topics, in the topics that a patent attorney would like to hit on during an interview with an inventor. So what’s the environment in which the invention is used? The problem it solves, prior attempts to solve that problem, if any, and more details on the invention itself. I’m going to go ahead and skip this step for the demo.

We are ready to generate. This is kind of a multi-phase process. We’ll start out by generating a formal invention disclosure document. This is a document that kind of helps suss out the critical aspects that our system will need to generate the application. It’s also the type of document that would be useful to conveying the information in kind of a bite-sized way to either your colleagues or investors, or for documenting trade secrets if you decide not to file the application. Next, we’ll go through the application generation itself and a final review of the documents. So let’s get started.

I mentioned that we don’t use prompts. Instead, we go through an alignment process. So you’ll see a field pop up in a few minutes where we start that alignment. In the first alignment step for the formal invention disclosure, we’ll be looking at the problem being solved. So we want to make sure that our system is on the same page with the user as to specifically what is being accomplished with this invention and why does it matter?

Okay, here we go. So this is our system’s stab at the technical problem being solved. This is a free text field. So any changes that might need to be made, refinements, can be made by the user here. If you’d like to see kind of a rearticulated version of our system’s understanding, you can click regenerate. But typically this is right on or close to right on, but it’s an opportunity for alignment with the user. I will continue and accept this and go to the next alignment phase, which will be kind of a high-level description of the invention. So it’ll be 3 or 4 sentences that describe what the invention is.

Here  the invention provides a system for delivering augmented reality context overlays to a patent attorney during an interview. Blah blah blah, blah, blah. This kind of hits on all the main points. Again, we ask the user to read this carefully, make any refinements that need to be made. Again, rearticulation can be, again through the regenerate button. We’ll go ahead and accept this and go to the third and final alignment step for the formal invention disclosure. This will be the technical details of the invention. So it’ll be, let’s say, 4 to 12 paragraphs that go through each of the kind of main components of the invention and describe it in a nice, clean, pretty way.

All of these details are details that were pulled from the invention disclosure document. You know, this is not an inventing system. We’re not trying to, you know, extrapolate on what was provided to the system. It really kind of keeps it within the bounds of what was provided, what was described by the user initially. So again, this can be refined if there’s any details that need to be added or nuances that don’t come through. Clearly, the user can modify them here and click accept.

And we will get to final form here. This is just a few questions on public disclosure. This has no effect on our system, but these are the types of questions that you want to keep track of as a company. And then a patent attorney would be interested in knowing. Given that patents are a date-driven business, click submit.

In a moment here, you will see the invention disclosure document presented. It shows up as a Word document that you can download. But it is a nice, clean, as I mentioned, bite-sized description of the invention. So it’ll be 3 to 4 pages that concisely describe what you have invented. This document, plus the document that you uploaded, will be used to generate the patent application. I will go ahead and jump into that.

On the invention disclosure, we’ll also be going through a couple of alignment steps. The first is the target for protection. The target for protection is two sentences describing what specifically we would like to protect with this patent application. So a little bit more targeted than the prior alignment steps.

Here we go. So the target for protection being nice, clean, short, setting terms again, that can be refined or regenerated as needed. Assuming that this looks good, we will click accept and continue. The next alignment step will be the figures that were provided by the user. So we will suck out these figures from the disclosure document. Present them to the user so the user can either select, deselect, or reorder those figures. We will be generating several additional figures, figures that are commonly seen in this case, software patent applications. But at this alignment point, we are only looking at the figures that were pulled from the documents uploaded by the user.

I’ll go ahead and select both of these, and we could reorder them if we wanted, but the original order was just fine. So I will click and continue. The final bit of alignment is on each of the figures individually, so there will be two steps per figure. So in our case, we’ve got two figures, so there would be two steps each. The first step for each figure will be identifying the components in the figure that should be described in the patent application. So if the figure has labels, those labels will be read off. If, like in our case, there are no labels, the system will interpret the drawing in the context of the invention and attempt to identify the components to describe in the application.

It does this in kind of an over-inclusive way. So oftentimes you’ll want to turn that down a little bit. In this case, let’s just say it’s an inventor, a patent attorney, and payer. And so we’ll get rid of the rest of this stuff. Click accept and continue. The next alignment step will be the entire description of that invention as it will appear in the patent application. So it will be several paragraphs, paragraphs that describe each component individually. There will be examples of that component, alternatives to that component. And then there will be a few paragraphs on how those components work together when practicing the invention. So this is the same type of format and layout that is considered a best practice in the patent world.

You can see that here. We’ve got several paragraphs describing each component individually and then the collective description. We will go through this exact same process for the next figure, but we’re running a little short on time, so I’ve got one already backed up that I ran earlier.

But let’s take a look at that. So once that final alignment stage is complete, the documents will be generated. So this is the document package. This delivered editable figures, the formal invention disclosure that was provided, the documents used as input, and then the figures in the specification, which would be the documents that are actually being filed.

So look quickly, the figures. So this is our figure one. This is a system diagram that you would expect to see in a software patent application. However, it’s custom for each application is generated. You’ve got your inventor-provided figures. And then a number of other figures. So some diagram, block and component diagram. So this is typically how software is described in a patent application, sort of as functional buckets where these components are the functional buckets that are described in the patent application. A device diagram for hardware support. And then a couple of flow diagrams. This is exactly what you would expect to see in a quality application written by a big law firm.

The specification. So the text component of the application, in this case, 44 pages. All of the sections that need to be in here are in here. They’re all written according to best practices. And all the way through, figure captions, detailed description, detailed descriptions of each of the figures, and all the way through to the end, where we’ve got claims that are provided. And, yeah, let’s see, claims that are provided somewhere and we’re a little bit running, coming up on time, but, okay, sure, sure.

So anyways, that is the result of the process. So the idea here is that inventors, startup founders, can generate their own applications, file themselves through the patent office website for $120, and be patent pending in the same day. So with that, happy to answer any questions.

Q&A Session:

Roland Vogl:

Right. Well, that’s great. Ian, thank you so much. I think since the AI revolution started, like 3 or 4 years ago, a lot of people have been saying, oh, you can, you know, soon we’ll be able to draft patents with AI and ChatGPT will be able to do it, but it has, you know, because ChatGPT is like kind of random, it has all kinds of issues.  I shouldn’t be surprised that you seem to be the first person to build a real such an elegant system around it. And it’s really, really impressive. So congratulations.

There’s a couple of people who have a couple questions in the chat, and I’ll get to that, but right away. But I think, you know, one of the questions, obviously, Paximal’s business is to work with patent attorneys, but then you’re rolling out a product that allows people to go directly to the patent office and circumvent the patent attorneys. There’s a little bit of a tension there. And it’s kind of I guess threatening to the patent bar. How do you deal with that? Ts the idea then that people, that this is then just sort of like a provisional patent, the patent pending type of thing? Or is this like a full-on patent?

Then the other question I had is around evaluation. Right. You showed like it puts it all together nicely. Right. But what kind of testing did you do and how did you compare it to like, you know, best-in-class human patent attorneys who would go through the same process and say like, oh yeah, this is like, here the machine has whatever, you know, like in litigation, it’s like, you know, the machines hallucinate cases. And I don’t know what the equivalent would be in the patent context. And how do you deal with that? You mentioned alignment steps? I think maybe that’s the way for you to kind of make sure the AI is staying on task and is not producing outputs. Anyway, so I keep rambling. There’s so many questions there.

Ian Schick:

So in terms of Idea Clerk kind of inching into law firms’ business, you know, we’re really targeting early-stage startups. So these are companies that typically are not funded or have, you know, angel funding, $100,000 or something. It’s a segment that has been historically underserved. They’re not kind of to the point where they can afford a big law firm. Patent applications are still $15,000 if you go through a law firm. And this gives that kind of early, early innovator a way to get their priority dates as early as possible, get something on file that they can be confident with, do it themselves.

But the idea is that, you know, they will use these to protect their inventions and ultimately need an attorney at some point down the line. So, for example when these are prosecuted, they’ll need an attorney. It’s really to get the patent pending early.

The documents can be filed as provisional patent applications. That typically makes sense for startups because it’s a way to get on file quickly but delay the costs of prosecution. But they meet all of the requirements of a nonprovisional application. So they are fully enabled and kind of include everything that you would need to file as a nonprovisional.

The documents are benchmarked against the best of class patents. And so if you look at our output, it looks a lot like Qualcomm patent applications, looks a lot like Salesforce applications, kind of the leading portfolios that are out there.

And in terms of dealing with hallucinations, you know, this is not a creative writing exercise. We’re not, and it’s also not being used as an oracle, as a source of truth. I think those are areas where you get a lot of hallucinations. This is more akin to a translation system. So, you know, classically translators go from English to French sentences, for example. It’s one piece of information transformed into that same information in a different form. So in this case, we’ve got the invention disclosure document that we’re translating into a patent application. So we kind of have our universe of information established and don’t need to have the system get creative. 

Roland Vogl:

So translation into patent legalese, so to say.

Ian Schick:

Yep. Exactly.

Roland Vogl:

So okay, so Benjamin is asking how the system delineates between obvious and non-obvious claims in the patent. Would an  AI agent produce, some solution to a problem. If the idea that you had is generated by the language model, but not explicitly prompted to use a specific solution.

Ian Schick:

Yeah. So it’s not an inventing system. Like I mentioned earlier, we are kind of taking what’s presented to us by the inventor and translating that into a patent document. In terms of obviousness or the other bars of patentability. The patent office, you know, ultimately that’s up to the patent examiners. We don’t do a prior art search. That’s not a part of our process. For Idea Clerk, you know, we rely on the inventor’s expertise in the field to, you know, know that this is new and unique and something no one else is doing. For Paximal, there may be prior art searches performed by the attorney. But again, it’s a guided process. It’s different than what you saw here. But it’s performed in a way where the attorney has the opportunity to kind of make sure that what is claimed is, you know, there’s enough daylight between that and the known prior art.

Roland Vogl:

Got it. Okay. Lucianne is asking about will one of the features be to run a search for competitive ideas? I suppose that would be, you know, sort of a prior art search.

Ian Schick:

Yeah. Yeah, that’s, that’s, you know, something that comes up from time to time, you know. Like, we’re not trying to ideate we’re not trying to kind of help people be creative. There are many tools out there for that. We’re really concerned with the legal part of it.

Roland Vogl:

And how do you protect privacy and confidentiality of the idea if you’re using ChatGPT or Claude?

Ian Schick:

Idea Clerk uses the same system, same backend system as Paximal. We’ve got a managed relationship with Microsoft. So we’ve got our own models running on Microsoft servers that do not store anything. We do not train off of any information that we receive. The information we receive is not accessible to any other customer or any other kind of any other party. But we definitely have that as locked down as possible in a way that makes the largest law firms in the world comfortable. 

Roland Vogl:

Adam is asking about, how do you make money on this? Is it on a very reduced basis or on a subscription basis? 

Ian Schick:

So Paximal is an enterprise platform. So we have seat licenses or enterprise licenses available for Paximal. Idea Clerk is subscription-based. Based on the subscription level, you have a certain number of invention disclosures and patent applications that you can create periodically.

Roland Vogl:

Okay. Doss is asking for the URL and I think, yeah, someone else asks for your contact info. Maybe you could put that in the chat. That’d be great. The final application could be construed as a legally competent filing. So in a sense, you know, I mean, you sort of like said, sort of legal advice. And could somebody challenge this being sort of as malpractice or like UPL?

Ian Schick:

Sure. So we’re not actually performing the filings. So with Paximal, a patent attorney ultimately signs the document and files. It kind of makes it their own. With Idea Clerk, it is set up for pro se filing. So the inventor themselves files the application. And so in neither case are we acting as counsel.

Roland Vogl:

Got it. Okay. Well, Nancy is offering some advice on TOS use terms. I think we covered most of the questions, but yeah, I think we covered a little bit over time already. But yeah, I really appreciate you coming in and launching this here. It’s like, I mean, clearly, I feel like I’ve sort of seen the future. I mean, I think one of the questions I have is, you know, let’s see, what people are gonna use this. The startups are going to use this. It’s going to increase the amount of patent filings. It’s going to overwhelm the patent office even more. I think there’ll be an opportunity to sell similar software to the patent office so that then eventually there’s got to be an AI deciding and it’s going to be like sort of high-frequency trading or something on the stock markets, the market of, you know idea trading and you know, protecting. And so, I don’t know, what’s your vision on that?

Ian Schick:

Yeah. Well, the IP5, so the top five patent offices in the world, which makes up about 80% of filings worldwide, they’ve been working on this, anticipating this for many years. Really, since we first launched. US specifically to the first kind of company in this space. But they are putting an AI across the examination process from prior art search, categorization, classification, and prior art analysis. I think they’re getting ready for it.  think that they’ll be able to meet the demand.

Roland Vogl:

Cool. All right, Bruce, final question goes to you.

Bruce Cahan:

Thanks. And this is awesome. Like, like Roland says, how are you able to kind of give a scorecard for patent filings that ultimately get approved that started with Paximal? So there’s a kind of a sense if the community is working well or poorly. 

Ian Schick:

Once we generate the documents, we don’t have visibility on them anymore. O they’re filed, they’re kept private for 18 months. So there’s a little bit of lag on the feedback that we can get on performance. But we do have about a half a dozen or so applications that have been shared with us that have been allowed by the patent office. So we are kind of working on some communications on those statistics. So be on the lookout for that. Probably through my LinkedIn newsletter.

Roland Vogl

Beautiful. Thank you. Yeah. Awesome. So round of applause to you, Ian. Thank you so much for sharing, you know, here with our Codex group, and, yeah, this was really cool. Thank you all for joining us.