Oxx VC on Agentic AI

We sit down with Bob Thomas, Partner at Oxx a B2B SaaS-focused fund, to discuss the hype around Agentic AI, what the future holds and how big businesses can unlock superpowers.

We're on the cusp of a new frontier: Agentic AI. 

Everywhere you look, it’s being applied—from robotics to autonomous vehicles to intelligent personal assistants.

Unlike earlier AI models that rely on predefined rules, or Generative AI focused on content creation, Agentic AI anticipates needs, suggests actions, and makes decisions that align with set objectives. It functions more as a partner than a tool. We wanted to lift the lid on this (equal part exciting, equal part terrifying) technology, so we asked an expert.

Bob Thomas is a Partner at Oxx VC and a finance industry veteran with an operator background. He is a formidable, yet humble investor who casually completed a Master's degree in Computer Science during COVID-19, just to get a better understanding of LLMs. 

Over a call, we discussed the hype behind Agentic AI, his investment in Peak.ai (which uses Agentic AI to optimise inventory and pricing) and what he/Oxx looks for in a founder. 

About Oxx:

Oxx are self-professed B2B SaaS geeks. As a fund, they typically write checks of $5 to $15 million in partnership with B2B software companies in Europe at the early scale-up stage. They invest in businesses that have clearly established Product-Market Fit, with the mission to provide the support, expertise and network to help these businesses scale. 

In particular, Oxx looks for specialised companies, where success comes from going deep. They care about the product value proposition, the tech, and the potential to revolutionise entire categories, even more than they care about financial traction.

The interview:

What was your first experience with Agentic AI?

Bob Thomas: During COVID when everyone needed indoor hobbies, I went back to school and did a Master's Degree in Computer Science. I did my dissertation on breaking LLMs, such as prompting LLMs in malicious ways to give you an unexpected output.

I spent quite a lot of time under the hood of LLMs, trying to understand how you could poison them when they were being set up and how to fine-tune them in ways that made them do bad things. 

The noble answer to why I was interested in the malicious side of LLMs is because it's a tremendous area of opportunity for malicious actors. But really, if you’re a Computer Scientist who's not very good at building things, it's way easier to break. As a result, it was just an easier way into the discipline for me. So, I have familiarity with some of the open-source projects around prompting LLMs at scale.

It’s not about opening Chat GPT, asking a question and waiting for an answer. It’s about sending thousands of permutations with the same prompt to an LLM and then analysing the differences in results. It doesn't sound very much like Agentic, but the tools you use to give LLMs prompts are the basis for what you use to build agents. 

Before I invested in Peak, I was spending a lot of my time working with companies that were solving a big enterprise problem. It’s where I saw huge scope for Agentic AI.

The problem with enterprise companies is that a huge amount of data is collected but siloed. They're not combined in a way that allows a company to make the most intelligent decision. That's what Peak did in Gen One. They helped companies effectively surface and leverage their data to draw insights. 

Why are you excited by Agentic AI? 

I'm super excited about Agentic and I've spent the last 6-12 months at various conferences raving to anyone who will listen about it. I'm excited about two things. If you're a startup, you can build stuff very quickly and cheaply, and then if you're a big business, there are large areas of knowledge work where there's potential to remove costs and improve efficiency. 

The cost advantage it gives early-stage software businesses is huge. If you're building a software business today you can use open-source frameworks of LLMs to do things that historically you would have had to hire people to do.

That can be sales automation, marketing, content generation, design, and coding. It could even be product ideation and product design. A Chat GPT request is not going to design your product for you, however, a team of well-specced agents alongside you is going to massively accelerate what you can do as a product designer. 

Now, if you're in an enterprise, you can't really play with this stuff in the same way, so that's where Peak comes in. That's what's exciting about what they're doing. As I said, Gen One of Peak was to give enterprises “AI superpowers” in bringing all this data together and allowing them to surface the correct strategic decisions. Now you can automate that. Before, someone had to look at a dashboard and make connections within their organisation about changes to activities. Now, you can start to automate the results of the strategic analysis using agents. 

The tools that you had to do this before were really bad. Could you imagine trying to automate your entire business with Zapier if you’re Tesco? It’s unthinkable. Now you can use Peak. If you're a big business, this gives you superpowers. 

Another use case for Agentic AI is with companies that bill by the hour, which are incentivised to take up time. The potential level of automation in human knowledge work is dramatic. 

I have to be slightly careful at this point because I was at a conference earlier this year where the conference organiser said, “Well, obviously you don't think this is going to destroy jobs, do you?” I was like, no, this is going to destroy hundreds of thousands, if not millions of jobs very quickly.

So, I think that as a society, we need to think about things like Universal Basic Income and there need to be protections for people who cannot re-train in a world where this is about to happen. 

It's becoming increasingly difficult to distinguish what is a chatbot and what is a human in a way that I think makes it quite deployable in people's workflows. You've got Agentic AI businesses like 11X and Salesforge which act as sales reps and Harvey, which augments legal workflows. In Sweden, where I'm based, there's a business called Leya that's now raised two massive rounds. They bring legal information into high firms and into general councils. Then there are all sorts of different coding applications. 

If you look at things like Aider, a truly Agentic coding assistant on GitHub, you can build a web application in minutes. Maybe you’d then want to share it with a developer and they can improve it, but you save tens of hours at the beginning of a project when you can get prototypes incredibly quickly. There's a business called QA.tech here in Stockholm, that uses agents to test web applications. Again, this was something that we used to pay people to do. With agents, you can truly simulate human behaviour in microseconds. 

As the barriers to building products disappear with the democratisation of software, what will separate the wheat from the chaff?

I think it's a combination of three things; the team, the angle on a market and funding. 

On the angle of the market, people find it quite difficult to sell tools to knowledge workers that will replace their knowledge work (for obvious reasons!) If I came to you and said I'm going to sell you the tool that does all of your job, you would be very nervous about buying that tool. It's a hard sales channel for this reason. Cracking this will be the route to win.

On the funding side, unfortunately, the ability to raise capital is a huge reason for a startup’s success. Look at Harvey, who I just mentioned. They’re a two-year-old business that’s raised $206 million in total.

If you're a UK legal tech business looking to automate professional services in the same way that they are, you've kind of got two options. Outrace them, which feels pretty unlikely, or have a different market segment, different route, or different angle on the opportunity.

I think it’s possible to compete with them. I don't think Harvey wins this entire space, but you just have to start at a different place.

What do you look for in a founder? 

Domain expertise is super important. People with high levels of knowledge in a field typically find Go-To-Market fit faster. If you're selling to the insurance industry, having never worked in the insurance industry, you're likely going to have a bad time.

We also often look for founders who have done the founder journey a few times before and have seen the negative consequence of the aggressive, premature application of venture capital. 

I remember in my previous business, we raised tens of millions in total and we were just obsessed with raising high amounts of capital at a high valuation. Actually, if you're running a business where you're really focused on your customer and your unit economics and truly build value for yourself as the founder, that's not the calculus you're doing. You're thinking what's really the right amount of capital to take in now for the size of the opportunity you’ve identified? 

We'll size an opportunity with founders and we're very collaborative and flexible about things like valuation and terms. When we have conviction around a business in a space, we get very aggressive on those things, which is very founder-friendly.

Where do you tend to find the best founders?

It’s a total mix. I think a lot of the time, we spend quite a lot of time within our fund ecosystem and more broadly with people who are technical and commercial experts from enterprise software. People who have careers at Oracle, SAP, Alphabet, or even smaller, scaling businesses. They'll say, hey, we just contracted with this business, it's a game-changer. This is really interesting technology. You should check it out. I think that network is incredibly helpful, but we’re also very open to people who come via inbound out of the blue.

What does Oxx offer founders other than capital?

We make sure that we have experts in our network who can unlock game-changing opportunities for startups. But more importantly, we truly have a partnership dynamic.

We've known most of the founders we work with for two or three years before we invest, so we’re the people they'll pick up the phone to when they're raising. For example, we first met Richard Potter, founder of Peak, in late 2018. We had just made our investment in Funnel, a marketing data enrichment business, and were really obsessed with the idea of gathering enterprise data and enriching it. 

I did market mapping and found Peak, found Richard, and went out through my network to 20 different people asking how I could get in touch with him. Eventually, I went up to Manchester to meet him and we started to get to know each other didn't invest until about two years later. 

So, I think investment is a partnership. If that partnership dynamic isn't there, it's quite hard to fabricate in a day. It takes time.

Why do you focus on scale-ups that have established Go-To-Market strategies?

We've seen quite a few investors work to apply capital to businesses that were still experimenting with their Go-To-Market model. In an area like SaaS, which has a global sales footprint and is quite competitive, that's a recipe to waste quite a lot of money quite quickly.

It's a niche strategy, targeting a business which has a niche approach to scale, but if you're a business with under 100 million revenue run rate, having a broad approach is a highway to inefficiency. 

This strategy has been really successful and fund one is doing very well. We're partway through deploying fund two now and we'll raise a third fund at some point in the next couple of years. The strategy for fund two will be exactly the same as fund one, as will fund three. 

What changes are you seeing in the VC world right now? 

I do think the recent economic woes in Europe have impacted early-stage funds. Some of them are finding it hard to raise and they're being a bit more cautious on deployment. Also, the ambition of some of the teams to build a global business is lower. I think that's what we want to work to try and solve at Oxx. We want to be there for the founders who don't necessarily want to build a local European business, but who want to build a global business. We can give them access to the capital, support and the support on the road to do that.

---

We matchmake founders, investors and executives and it’s the quality of our network that sets us apart.

Want to join this exceptional network? Want us to help matchmake you?

Join the network

Get in touch: hello@swiftscale.co