Building the Data Brain for AI Agents: Toby Mather, Co-Founder of Rig, on Data Infrastructure, Enterprise Sales and the Agentic Future

0:00This episode is sponsored by Cpoint, the business account built for startups. If you're a startup founder fed up with finance admin, you need Cpoint. Cpoint is different to neobanks because it connects all your bank accounts and your strike accounts to your cash position, burn and runway instantly. It automates bookkeeping by pulling invoices and receipts from your whole team's inboxes, so you just sync everything to zero and pay outstanding bills with a click. It has a 3.49 % yield on treasury and real human customer support. Find out more at seapoint.co. That's S-E-A-P-O-I-N-T dot co. Use code UNICORN for a free month.

0:35Seapoint Treasury is a money market fund. Rate recorded at 1st of October 2026. Rates are variable and subject to change. Capital at risk. Hello and welcome back to another episode of Riding Unicorns. Today we're delighted to be joined by Toby Mather, co-founder of RIG. Toby's journey offers a unique perspective on navigating enterprise sales, building products for big clients without becoming a service agency, and pioneering the concept of data brains in the AI stack. So Toby, welcome to the show. Maybe you could explain what RIG does.

1:06Toby Mather:RIG is AI native data infrastructure. So we build the building blocks that help companies create a data system that is usable and built for humans and AI agents to use together. So when we started, people knew us really as a data context layer or data brain that they could run on top of their data warehouse, on top of Snowflake or BigQuery or something like that. And it would make sense of all of the complexity and mess of all their data and let them add role-based access controls on top of it. And then their business teams could plug in Clawed or Cursor or whatever tool they wanted and build custom data applications, custom automations, just chat their data, that kind of thing.

1:44Toby Mather:We've since extended to also include managed warehousing and data ingestion because lots of our customers said, oh, I wish I could get X data or Y data into my data warehouse, or I don't even have a data warehouse. we're just YOLOing all of our data through MCPs. So now RIG really is an end-to-end data stack built for kind of AI native teams. And looking at your previous company, Lingumi, which obviously was acquired, what are the sort of key lessons that you learned through that journey that you feel have been useful with your journey at RIG? Yeah, I made the non-obvious pivot from preschool education to B2B data infrastructure.

2:16Toby Mather:I was always very data obsessed and was the one-man data team at Lingumi. And when we were acquired, I then ran the product and the data teams at Novakid Aracquira. And that was a much bigger role. So in terms of learnings, what I brought from the Lingumi nine-year journey into RIG is number one, pick a very large market. And actually, I remember having a meeting with some colleagues in the US sector nine years ago, and then saying, we really believe in big markets, and this isn't a very big market. And it's true that back then in 2015, preschool, pre-AI tech-driven education was very niche. And so I sometimes describe it as a bit like sailing in a canal.

2:52Toby Mather:You can sail in canals, but you might hit the sides and you can't go that fast. You have to be a very good sailor as well to not hit the sides. And RIG is really sailing in a massive ocean. A great customer is a company with lots of complex data, and that describes a lot of companies. And so sailing in a vast ocean is great because you can go very fast, you can go very far, but also if you lose wind in a given area, you'll probably pick up wind somewhere else. So as a sort of foundation for finding product market fit, one of the learnings was move into a very, very large market and build something out of space.

3:22Toby Mather:Number two, the second thing I brought across was really build a product that I use every single day. At Lingumi, I can never be my own customer. I didn't have children and I was a native English speaker. So I just had to sort of imagine a great product. And I think we did a pretty nice job and we got to millions of users and millions in revenue. But I don't think that level of sort of product obsession got to the stage it's at with RIG where we spend eight hours a day, we run RIG on RIG. It is our own data brain. it runs all our automations. And we're a much smaller team as a result of having a rig.

3:51Toby Mather:And so we then push the product ourselves really to the edge of what's possible with it and then extend it. And when most people think of AI, they probably think of an LLM. Most people will have used ChatGPT or Claude, which is essentially a system that predicts the next word. And that's how it works. You guys are in the data space. So maybe you could just explain why LLMs are maybe not the perfect solution for data reasoning and data understanding and why rig is really changing a lot of work that can't be done by llmai we see a lot of teams who are plugging clawed into 15 tools through mcp and it's brilliant in the sense that kind of unlocks access to lots of data straight away and they don't have to work through an analyst or a data engineer to build a pipeline to build a model to build a dashboard for them so i can quickly empathize with why teams do that but it's a bit like instructing a linguist to do a lot of maths the linguist might do some of the maths right they might be doing the maths wrong it's very hard to know if the linguist is doing good maths when a mathematician is readily available should you give it you know all the numbers and that's because when you connect an LLM like Claude to lots of tools that MCP that connector that Claude is using is pulling in data through APIs and in its context window which is sort of a linguistic well actually at the end of the day it is actually maths but different type math.

5:12Toby Mather:It's a linguistic sort of process of trying to fudge all that data together and make sense of it, rather than taking all that data and putting it into tables against which you perform maths using a coding language called SQL, which has been around since 1970, where the answer is a deterministic calculation against data in tables. And so when you're working with data, you want to use as little non-deterministic methodology as possible. And the use of AI is just designed to get the data into a shape where you can perform a mathematical calculation on it. So that's how we think about AI. AI is used to get the data into one place, to make sense for the data in that place, so it's explainable to the AI.

5:49Toby Mather:So the AI at the end of that process can write a really good calculation in SQL across that data. and that's if you are building a startup today or in a scale up thinking about the data stack the thing you should be doing is using good old-fashioned tables a data warehouse getting everything into those tables and then harnessing ai in a certain way whether it's with rig or another tool to perform accurate calculations across those tables it's fascinating because i suppose the i mean there's just an explosion of unstructured data as well which is kind of as a result of LLMs that companies like RIG are now having to deal with providing some structure around.

6:27I'm really wondering whether you're having to like re-architect your platform regularly, because you're running your own business off RIG. But you know, we're speaking to companies who are in the sort of agentic space, and they're finding best practice changes every few months. And so are you guys having to change your product super regularly? Or is this far enough down the stack that actually you can configure it in such a way that it's pretty future-proof today.

6:52Toby Mather:Yeah, I think there's this great besaucism around Amazon of like build for the thing that doesn't change. I think the thing that hasn't changed in 55 years is that tables are a very good way of storing and calculating against data. And often I meet teams who say, oh, but we have our SOP documents. Surely we need the agent to have those in document format. And the answer is if data has properties, like who created it? When did they create it? Has it been updated? When was it updated? Who is it updated by? Who's allowed to access it? Is there a summary that you want to write of it? We sort of took a bet at the beginning of Rake based on all our experiences.

7:24Toby Mather:And my co-founder had spent some time at Palantir doing the sort of foundry and ontology deployments that use a similar structure. Our bet was putting data in tables, relational tables, where you have ideas that join them together. Started in 1970. It was a big step up from what happened in the 1960s. And we've been doing it exactly the same way ever since. And we don't think that's going to change just because of AI. I think AI is going to get amazing at getting it out of the tables though. Yeah, we're hearing a lot about the sort of context layer and the reasoning layer. Like are people building that on top of you guys or is that the space you are in and you're competing with them?

7:57Toby Mather:Yeah, RIG started as a context layer and now we do a bit more than that. But at the heart of the platform is this context layer, which keeps looking at your data and making sense of it and updating itself as the data changes. And that's why we call it sort of a brain for your data. I know it looks good, but it's actually the wrong field. and we sort of take away all that unstructured skill information and we put it into this really structured format so the whole organization can use it and you can decide who's allowed to use the data or you can keep it up to date and so on. And what's the most common output for clients?

8:25Toby Mather:Somewhat, not depressingly, but there's this writer on data called Ben Stancil who founded a business called Mode Analytics. He's a very funny writer on Substack and he has this blog post from a few years ago called Every Data Platform Becomes a BI Platform. And the answer is a lot of people just build dashboards. They were doing it with clawed artifacts. Now they sort of build it on rig because you can vibe code something in clawed and host it on rig. And there's a bunch of reasons you want to do that around security and governance. So a lot of people just build dashboards. And we see a lot of go-to-market teams will build sort of custom go-to-market command centers where rig makes sense of all their CRM data for deal velocity and time and stage.

8:58Toby Mather:But you can also pull in product analytics and customer support tickets and kind of put that into this really beautiful system. I think that's great. but I think the kind of cutting edge of what people are doing with the rig are more fully agentic systems where it pulls in data from all their different tools but then it can push out to another tool. For example, we have teams who have replaced Gainsight with rig. So they take data from usage analytics, from customer calls, from customer emails, from the contract with the customer and any mentions of competitors on calls over time, renewal dates and so on.

9:29Toby Mather:And then it pulls it together into emails to send three months before renewal, two months before renewal, one month before renewal, or to really make sure they're booking a workshop and a check-in call and so on. That's sort of perfectly aligned to the context and usage of that customer. The other way that we try to spearhead rig usage internally and the way we use it is it runs our CRM for us. So after every sales call, it processes the call following a sales framework that will run in a model in rig. And then it suggests what stage to move the deal to, what context to attach the deal in our CRM and then sends us a Slack message in a channel saying, hey, here's my pros changed.

10:02Toby Mather:Do you agree? and we can reply to it or it can just click approve and then it updates in the CRM. So sort of human in the loop agentic control of tools that are grounded in lots of data contexts would be the bleeding edge of what we see today. I'm fascinated by the timing of this change from human in the loop to fully autonomous because we're going through this ourselves episode one at the moment and we're using a lot more agents ourselves as individuals. And currently we're all kind of in the human in the loop phase, but I'm just starting for certain workflows and processes to phase that out and to go fully autonomous.

10:35But I'd be curious to hear what you're seeing. I mean, both internal use cases, like at some point, I'm sure the proposed changes are accurate enough to basically say, okay, just make these without our input anymore. But like what you're seeing from customers, how important it is for them to have human in the loop at the moment.

10:52Toby Mather:I think we sort of missed a step. Like we're all sprinting before we've learned to walk properly, but we can't do that. We can't go from 25 different systems with 25 different sets of data to agents. It just doesn't work. The agent is too much space and too much data and too many systems to reason over. I would say that the quality of an autonomous agent is defined by the clarity and context and consistency of the data it's working against. So if you give an agent a single table of data, you say this is your entire domain, be autonomous, it will do a brilliant job with that small constraint and well-explained data.

11:23Toby Mather:The reality of most businesses is you have a very messy data environment. Your data warehouse is maybe 50 % complete and 50 % model. So let's say your 25 % of your data is well represented. And your agents will reason against that, assuming that's the entire universe of data when actually 75 % of it is in raw or it's out in other tools. So I'm not seeing in the startup environment truly autonomous agents doing jobs, except in a couple of spaces like encoding, where I think obviously we're all kind of using tool or cursor or whatever it might be. And actually in like recruiting, I would shout out Jack and Jill.

11:55Toby Mather:I think their agents are like genuinely agents. It's really, really nice experience to use on. For everybody else, the walking stage is sort out the data infrastructure beneath your agents. Then you can go through human in the loop quite quickly to basically make sure it's doing what you expect against the tables you expect. And then lastly, just let it run itself. Yeah. And you are still quite a young company, but you've landed really big clients very early on. How did you do that? How did you build the trust to get access to this level of data? And what advice would you give to other founders who are starting out, who want to land enterprise clients, but don't want to become an agency to those clients, want to build trust with them?

12:40Tell us a little bit about your sales processes and how you kind of got such amazing traction so quickly.

12:46Toby Mather:I know absolutely nothing about B2B sales. and so when I started working on this I knew I wanted to do something at B2B and I wanted to be my first dog fooder I'd use my own product a lot and I love data and data engineering data architecture has some strong views on it so I turned to this guy called Rob Snyder many of your listeners may know Rob Snyder but he has a blog on Substack called the physics of startups great podcast as well his methodology because I couldn't understand medic or bant having not been in the sales world was if sales feels like hard work something's not working and he has this thing called the pull framework.

13:19Toby Mather:And to just explain it briefly and borrow from Rob, it's P-U-L-L. P is what's a project on your to-do list. And it's always a person, not a company. It's the executive that you're talking to. Is it unavoidable? What is the list of options you've considered to achieve that project? Because there'll always be things they've considered or they are considering. And then what are the limitations, the last L limitations of those? And if someone has a project on the to-do list that's unavoidable, they have a list of options and that list has limitations. If all four of those are true, you have pull and someone will probably buy.

13:48Toby Mather:Even if your website doesn't exist or your product's a bit janky or you've got lots of bugs and so on. And with each of the customers we've taken on, they have had a project. So if we look at Geve and Suri, they had lots of data in lots of different places. They're 3PL providers, they're warehouses, they're Shopify, they're Boots, Target. And it's a really fast growing business with lots of sales outlets and data systems. It was unavoidable because they're scaling so fast. They need a way of bringing all that data together and making sense of it. They had some options, including traditional data consultancies and a couple of other platforms that were sort of pre-AI generations of what we do.

14:19Toby Mather:And those had quite severe limitations, like very expensive to work with consultancies, it's a point-in-time solution, or working with platforms that weren't AI native from the get-go. So it was a really easy conversation. And then we've worked really hard to deliver a great architecture and framework for them. And it's been going really well. So that's how I sell is I don't expect to have to push to close. if someone doesn't have all of P-U-L-L I just say well that's great it's good to chat it and if you come up against this later let's talk again but if somebody where all four of those are true it's not even a sale it's just a conversation okay would you like me to solve that problem for you I have some software but it's not what you're worried about you're worried about your project that's how I do it so my question is how did you even get to that point where you knew that this was a problem for them how did you outreach to them without being salesy sort of say look I want to understand whether this is a problem for you How did you even get the ball rolling?

15:10Because once you're in the conversation, yes, I can see how you go through those steps. But how did you get the conversation in the first place?

15:17Toby Mather:I'd say the last step for Rig on this journey to product market fit is going to be that repeatability of the sales motion and the marketing motion. I don't think we're there yet. So a lot of it has been using my network. Customer referrals has been a big driver for us, which is great. My own sub stack, my own LinkedIn presence and so on have driven most of our business so far. My co-founder has brought in deals as well. I then have a conversation with someone and just say, well, this and this quarter, what's the top of your to-do list? And at the moment, that being a horizontal platform, we work with businesses of all sorts of different shapes and sizes, but they tend to have this type of problem, which is, hey, I've got this data.

15:52Toby Mather:Some of it's in the warehouse, some of it's not, or I don't have a warehouse. I want to use AI in the business to replace Power BI or to replace Looker or to help run agents' automations. I need to figure out somewhere bringing it together and making sense of it. And that then turns into a sales conversation. And I want to talk about beating the competition because you've already mentioned some of the incumbents, you know, they're not AI native, but they're all sort of trying to become AI native. And it is a crowded space. And there are a lot of names and the buyers are receiving a lot of emails saying, try this new tool.

16:19What's your playbook for beating competition in a space like this?

16:23Toby Mather:We're definitely competing with some big boys because we're competing with Snowflake, competing with Databricks. And that's very exciting to us. There's two things. One, there's just the philosophical bet that we're taking, which is by being AI native, we are not selling to a data team in general. We're selling to a business team because we believe that business teams are their own data owners of the future because AI now removes, except in very large enterprises for good reasons, removes a lot of the steps of data engineering and modeling and dashboard preparation that data engineers analysts previously did.

16:51Toby Mather:And those data teams are now becoming more like the context owners and context engineers for the organization. So rather than, you know, Snowflake who sponsored data conferences, we're going on to GTM conferences or RevOps conferences and things like that, direct consumer conferences and meeting business leaders who are stuck with their data. So that's sort of that bet. Then number two is, I think we really believe there's this company called DBT in the data space, and you can use DBT open source or the hosted version. They've now joined forces with Fivetran, who we're somewhat competitive with because we do a much cheaper data ingestion and we don't charge corrobes.

17:23Toby Mather:More shout out for RIG for anyone fed up with a Fivetran bill. We think about DBT sort of playing nice in the industry. So you can run RIG on top of Snowflake. You can also use rig instead of Snowflake and we run a managed warehouse. If you don't have to think about data ingestion and pipelines and warehouse management and instance sizes and compute costs and things like this, we just take care of all that and bring the cost down. So number one is have a different philosophy on the market. And then number two is sort of play nice. So you can plug into people who are already embedded with those existing vendors.

17:49But landing an email in someone's inbox or LinkedIn and getting them to understand what you can actually do for them and whether it's real is quite another question.

17:58Toby Mather:Yeah, and I'm a terrible cold email. I don't know if I'm doing something wrong, or it's just that cold email is like this at the moment, but certainly we've not cracked cold email, especially I think when you're a horizontal platform, you're saying, listen, do you have really messy data? You're struggling with your data systems. They probably are to some degree, but it's very hard to pin the project that's unavoidable where they've looked at list options and your email ads are just the right time. So we have had a couple of deals closed from cold, but if you look at the percentages, I mean, it's ridiculous, like sub-1%.

18:25Toby Mather:I don't mind admitting that. So no, definitely not cracked that, still needs to test that messaging. But what we have noticed is there are certain pockets of demands that are consistent. So for example, direct consumer businesses like Siri or retail businesses, we didn't expect this, but they have lots of data in lots of places. They don't have full-time data teams. They can't afford to run that sort of infrastructure in-house. So we're finding it much easier to have those conversations. And similarly, we've worked now with several companies where two companies emerged in a private equity deal, and they used to do systems integration, data warehouse integration, CRM integration.

18:53Toby Mather:and because rig maps out messy data where it lies and plans ways to get it to join together it turns out that's a quite a repeatable motion for us is to go and install rig for them help them run the migration and they use rig out the back so i think we're the beginning of finding these sort of verticalized paths to market which our core infrastructure can serve yeah and while you're building this understanding of what the customer needs how are you adapting your own product and roadmap. So how much do you let customers inform? And how much do you take an opinion and kind of drive it yourself knowing a bit better, should I say?

19:30Toby Mather:I think with Lingumi, I had this very clear vision for what a solution looked like. And with my co-founder, did Y Combinator with his last company, YC has this very, so let the market pull the solution out of views approach. I think that combination has been really helpful. And I've dropped all my 22 year old self's kind of vision of the future. I don't really have a strong sense of exactly what rigs going to be in the future. AI is changing very fast anyway. So we're really just thinking of ourselves as kind of riding the boat on the ocean and seeing where that leads us. And so we tend to bias towards customer needs and customer demand, just what our customers tell us they need.

20:02Toby Mather:And that's what led to us building an ingestion platform. It's what led to us running managed warehousing. It's what led to us adding write capabilities to tools, not just read capabilities. You can then push data out to places. And that's what made Rig a full stack platform. But there have been a couple of things where I just had an idea based on my experience. and put it in the product. And a couple of those things have become really big features. Like we have a feature called virtual data models, EDMs. And basically my idea was that when a data engineer is to take lots of data together and summarize it or enrich it, like you do in clay, where you can sort of enrich data in rows, it requires sort of custom data engineering.