#28: A new AI-driven spreadsheet arrives

This is going to be a fun one. We're seeing a new kind of spreadsheet. Who are you?

I'm Anna. I am the founder of Paradigm and I recently studied computer science at Penn and graduated early because I was very frustrated by seeing a lot of my friends that were spending like you know, 80 to 90% of their day in spreadsheets doing tedious and repetitive work. And I wanted to build something to solve that problem.

Very cool. So why do we need another spreadsheet? We got Excel, we got Google spreadsheets. Why do we need another one?

So I think that AI is not only changing the small parts of of how people do things, but actually changing the entire user experience of how they're using certain tools. So Google Sheets and Excel, these are tools that were built pre AI really. They were built for workflows that are just do not need to exist anymore that center around having to do things that LLMs enable you to be able to automate or augment in your workflow. And the way that I see it is it's not just spreadsheets, it's actually every primitive of a product. Take an IDE Cursor is a great example. They built a new IDE from the ground up even when VS code existed. Because the workflow of an AI native product is very fundamentally different than previous existing products. And I think this is going to happen in every single sector. And Paradigm is doing this for or a spreadsheet. Redefining the user experience around an AI first way of working with data.

Give us a little bit of a taste. Maybe we should get a demo because you know, it's hard to understand how a spreadsheet is different in this post AI world than before. But give us a taste of what kinds of things would be much easier with your new kind of AI spreadsheet than with Excel or Google Sheets.

Yeah, happy to hop into a demo. But to quickly explain first, one of the biggest things is data collection and aggregation. So for example, right now data is really dispersed across a lot of different sources and a lot of workflows for gathering like hyper specific data in a research context are centered around, oh, let me go to like multiple distinct different sources and bring this information back often in like a manual way. Like I'm literally going to gather every single piece of information and this is something that with a combination of LLMs and intelligently selected third party data sources you can completely automate and then focus more on. Okay, once I have this data, how do I take action on it?

Very cool. Can we see a demo.

Yep, let's do it. This is Paradigm. As you can see, there's a bunch of different templates because our product can be used for a lot of different use cases. But let's start from just a completely blank sheet. And what we're going to do is here we're going to say we're going to use chat to create the basis of our spreadsheet. Create. Create a schema or a sheet where I have companies and company descriptions and I want to find out more information about the companies and founders. So let's say you have some existing data, right? Paradigm, normal spreadsheet, they have column headers that define what information is. Every single column. Paradigm also has column prompts. For example, some are very simple, like for CEO, founder, name, it's just find the name of the company or CEO or founder and what will happen is we can essentially we can add a row and we can paste in some data that we have. So let's say that this data is a list of YC companies and you already have this information. Company names, company descriptions. What you can do with Paradigm is you can select on the cells that you want to generate and you click enrich. And what you'll see is very quickly the information that you need is populating in. And the interesting thing about this.

That'S cool. All right, so auto, fill in your spreadsheet. I don't have enough data over there. Let's create a couple more columns.

And the interesting thing is we have full data transparencies for every single piece of data we're generating. We're not just generating the data, but we're saying how the agent has gathered this information, like what it's doing. And a lot of the times it's very simple and a lot of the times it's more complicated. Right. If you're getting all of the social media links, it has to check Instagram and Facebook, et cetera. And yeah, there's just full transparency so that you know that the data is as accurate as possible. But let's say I want to do something more specific, right? Let's say I want to do founder technical projects and work history. So I want to. It will generate a prompt for you and then you can also edit it. So go to this founders, LinkedIn and GitHub summarize their past work experience from their LinkedIn in a few sentences and also go on their GitHub and see if they have any interesting technical projects. Explain them and give me the links. So you can do these hyper Specific things that are almost unfeasible to do manually. Right. I'm not going to go step by step, data point by data point and find this complex information and you'll see it generates in extremely quickly. Right. This is a hefty summary that has the links and information that we actually need right away. So I could demo all day. Want me to show one more use case or.

Yes, please. This is amazing. Okay. Does it do all the stuff that Excel does, all the formulas and all the sums and all that kind of thing?

So right now I can't say that it does every single thing that Excel does, but it does the most important things that Excel does. And the interesting thing is eventually it's going to be able to do all of the things that Excel does. But to start, we're really hyper focused on let's do things that Excel couldn't possibly do. So we have this like these behavior patterns of people using our product where they, they can't stop using it because they can't even conceive of how to do this task any other way. Right. So I think starting with these highly differentiated use cases has been very powerful for us. And then from there we have the bulk of what's important. And you can also export data. So you can like get data in, export it to Excel and sheets for now, and then in the future we'll just be able to do everything.

And can you import from Excel and Sheets?

You can, yes.

Okay, cool.

You can import, export data, all of that stuff.

And can you hook up to any other data sources? Can I import like a PDF?

Yeah.

So can I do something weird?

I can't import a PDF right now, but that's coming very soon. And one other interesting example that I want to show is let's say you don't have some data to import or export, you don't have starting data and you want to use paradigm to generate in the starting data. You can do that. So let's say I want a list of founders. Find founders of startups, startups in the Bay area with over 10mil revenue. This will populate with a data set that matches your criteria. So what this is going to do is it's going to give me a starting data set that's pretty comprehensive and this works with essentially any data set and then you can go and build on top of it and do the specific things that you want to do.

Now how did it do that? Did it go out to Crunchbase or something like that to find out who has big companies and who's in San Jose, California or whatever.

So essentially what happens is it uses integrations that are built into our products. So like we're integrated with things like Apollo and Crunchbase and we have our own data scrapers as well that can go out and get information from the Internet. And what we do is our product is able to choose the right tool for every task so that it will use the data source with the highest possible data accuracy for your task. So if you're asking for fundraising information, it's going to pull that from something like Crunchbase. So that's the most, most verified possible.

Wow. I want to bring my ex listen to this and start going to town. Yeah, I might have to manually import. That's an interesting question. Can I bring in a list of media people in here and start going in town like this or do I have to manually interim for now?

No, no, you can import, export all that stuff. You can even send outreach emails through the platform. Like it's very holistic. We're trying to create this like end to end workspace where you can do a lot of things that really help accelerate your work in an AI first way. Yeah.

Can I create charts like Excel does? You highlight a bunch of cells.

Right now you cannot, you can't create charts. Okay. So we're more focused.

That's one reason to go back to Excel.

Yeah, exactly, exactly.

You can see that's going to come though. You're going to hook up some data flux or something like that, also creating some crazy ass charts.

The interesting thing is I think our chat is right now it can do a whole bunch of things but eventually we are going to try to have it be able to do everything so you won't have to mess with these super tedious formulas and you'll literally just be able to say what you want and get a drafted output. Just like you would in like cursor for coding, for example.

And how much does this cost or is it free or what's the business model?

So we have a couple different plans right now we're focusing on mostly letting on like larger enterprises that are paying upwards of $500 a month. But we also have a wait list where we're just giving access to individuals that we think have interesting use cases for free. So we'll be opening up that more and more after the launch.

Very cool. Wow. What do you find the most interesting uses? Are you give me a couple. Are there others that people have found that? Because you can talk to your LLM this way? Because I assume there's an LLM running underneath here.

Yes, definitely. There's a bunch of super interesting ones. I think one that, that I really. It's super powerful is you can basically we have this feature called macros, but it's different than like Excel macros in the sense that you're able to combine cells, columns and, and spreadsheets as context within a natural language prompt. So what this means is you could say, hey, I want to rate this candidate on a scale of 1 to 10 compared to my other database of accepted candidates. Or hey, I want to see if anyone on my team has connections with this list of people based on this database of my team's contacts. And that allows you to do things that are really powerful and also not really feasible on other platforms.

Wow, that's really cool. Congrats again. This is, I think, I think you're crazy. They're trying to take on the Microsoft Excel and Google spreadsheets. It's really a new world though, isn't it? With this, with AI? How does it work with other AI tools? Do you have a suite of tools that you use to do this kind of things other than your own tool? And are you thinking of ways to build like an Office suite, a post AI office suite kind of thing? Yeah, I think I know there's people who do writing tools, there's company, there's lots. I have lists on X of all the tools and there's a couple hundred writing tools.

Definitely. I think right now our product already has so many use cases with just the this spreadsheet as a primitive that we're focusing in on it for a bit. However, we definitely will get more broad as far as use cases quickly. I think a lot of that will come from just understanding like what is these most desired use cases of people using the product and what's the natural extensions like? For example, one thing that we're thinking a lot about is how do we make this more of an end to end process in the sense of can we hook into data sets that are constantly changing over time and continuously doing actions on those data sets? Things like this are very interesting to us.

Can I hook it up to corporate data sets? If I work at a bank, there's a data lake. I don't know if it's on Snowflake or whatever, but is there a way to hook up to that data set and then start using your tool to pull data out of that?

Essentially, yes. So this is some of the stuff that, that we're working with a lot with potential like enterprise Partners.

Okay. And yeah, what else do I need to know? Because I'm trying to think back to the days when I actually did spreadsheets a lot. I did a lot of pivot tables, a lot of charts. Right. And then just a lot of manipulation of, of rows and columns and trying to make them do what you want them to do. Add them up.

Yeah. I think another interesting thing is the concept of evaluating the data quality. We've been thinking about this a lot and I think across every data provider and even across the Internet. And this is a huge problem with existing LLMs. It's how do I know if I can trust this data? This has been something that we've put an immense amount of focus on and how we're tackling it is in a couple ways. The first you already saw it is just this full transparency. Right. So instead of LLMs, which are more like a black box right now, let's just show everything. Let's not even just show the sources, let's show the agent's thought process in making these conclusions and coming to these answers. Which is important when you're having the agent do things that require more multi step reasoning. Like maybe I want to write an entire investment memo in a cell which people are doing on the product, which is very interesting.

That's wild. Look at this company and tell me how good this company is. Compare its key to E ratio versus its sales versus its future predictions and stuff like that and give me a rating of the company.

Yeah, exactly. So there's, there's ensuring the data quality and then there's also our focus on multi step reasoning. And where paradigm becomes really powerful is when you're doing tasks that require this multi step reasoning because you have this really great integration of the highest quality data. And also agents that are able to do complex sort of manipulations and reasoning on data, which is very interesting.

Wow. What's the reaction of the customers ban? How does this change their lives?

It's been really good. I think the best thing is beta.

Testers or whatever you call early, early testers.

Yeah. I think the most exciting thing for us to hear is just I have no idea how I would have done this any other way or this would have taken me like hours. And starting to see people like becoming dependent on this new cycle of working that I feel like in our minds we're such a small and talent dense team and we've been moving so quickly. So it's like this thing was just a concept in our head just merely a few months ago and we were able to make it possible and see that the impact is actually giving people their time back to focus on more meaningful work, which is pretty cool.

If you have a data set with a lot of sales data, like sales from every state, let's say I work at Walmart and I have sales from every store, can I start asking it for patterns like why is this sales down in this one store? Maybe it can go out to the Internet and go, oh, there was a hurricane last week or it was really hot and everybody stayed home. Can it start figuring out that kind of data to make your sales data more useful or more accurate?

It can definitely do things like that. And we're seeing people add entire columns with these sort of like enrichment data points that are complicated and require a lot of different contexts to figure out and get. Which is pretty interesting.

Okay. And so yeah, that's crazy. So reasons I would still go back to Excel is if I needed to do a really weird ass set of formulas because a lot of accountants, man, they have like multiple tabs and multiple formulas going, grabbing numbers from this set and bringing it over to this one to start making their, their profit and loss statements or whatever.

You know, definitely I would say that if, if your workflow is primarily financial in the sense of like calculations, you likely would want to use something like Excel. But if it's financial in the sense of doing research on some due diligence or some other set of information that you need to gather, then you would use Paradigm.

This is really cool. You went through Y Combinator with this company.

Yes, we did.

What was that like? Because. And how did that improve your product? Tell us a little bit about going.

Through there because it was awesome. I think at Penn I had, I really just had this core insight and this like obsession with building that like brought me into startups. And the core insight was I had been bootstrapping a couple different businesses pretty successfully throughout college. And what I did was I just manually created different automations that made a lot of like my work flows more efficient and more manageable to do while I was in school. And then I started doing some work in the agent space, like particularly working alongside some of the like brightest, like PhD students at Stanford. And I really started to understand that agents had this ability to automate the most tedious and the most repetitive of work. And not just that, to do it in a way that was better than humans could themselves. This is a big focus with gathering the data. Like it can really always be more accurate than a human because we can just cross reference more sites, et cetera. And I just had this core insight and I applied to Y Combinator with the concept of let me build an AI intern to automate your most tedious and repetitive tasks. And this was a very loose vision. And I spent a lot of time throughout Y Combinator just really making it more concrete and going through a lot of iterations with, with the group partners and with early customers to shape it into what it is today. So I think YC was incredibly helpful in just giving the mindset of you need to test and iterate and get this out in the world extremely quickly in order to build something that is right. And now I feel like we've ended up on something that is right because of that process.

Very cool. Where do I learn more about you and get on the wait list?

The best place to find out about me and Paradigm is Twitter. My Twitter is Anna Rmonaco and I guess other ones would just be like LinkedIn or our company's website. We have a wait list, so that's paradigmai.com join so anyway, email, wait list, Twitter, any of those methods are great.

Very cool. Thank you so much.

Thank you.