Who will build the next wave of breakout companies?

How AI is reshaping the founder archetype

Every technology wave creates the conditions for its own kind of founder to flourish. Having invested through multiple cycles, from on-premise software through the cloud transition to today’s AI application layer, I have watched this happen in real time. The SaaS wave rewarded operators who knew their industries. The AI/ML wave, at least until 2022, rewarded researchers who understood the technology. Each time, the people who built the most successful companies looked slightly different from the generation before them.

As we are (still) in the early innings of the generative and agentic AI era, I wanted to understand whether that pattern was holding, and more importantly, what it implied for the wave that is now upon us. So I did something I have been wanting to do for a while. I assembled a dataset of 645 founders across 270 successful (defined below) software companies, spanning the B2B SaaS era through to today’s AI application companies, and I studied what the data showed. I used AI tools myself to help gather, cross-reference and analyse the data across hundreds of LinkedIn profiles, Crunchbase entries and press sources. It felt right that the research process itself demonstrated one of the findings: AI has changed what one person, or a small team, can accomplish.

Before sharing the findings, I want to underline that most of the companies in this dataset with AI/ML based products were founded before ChatGPT launched in November 2022 – when foundation models were still highly differentiated and before a non-technical founder could build software by describing it in plain English. The conditions that produced these founders reflect a specific moment in the technology cycle, a moment that has already passed, but equally, which has laid the foundation for the agentic wave.

We are now in what we at Oxx think of as a Cambrian Moment: the explosion of AI applications made possible by commoditising models, collapsing inference costs and tools that have made building radically more accessible. So the data we explored tells us who built successful companies in the first two waves. The question that drives this piece is what it tells us about who will build the third, and what they will need around them to succeed.

Defining the waves

Software company building has moved through multiple broad waves. Each era has given rise to different types of founders, formed by the type of company and the conditions under which it was built. Some waves have been labelled generational shifts, and I’d argue that the adoption of agentic AI is the biggest one of all (hence the Cambrian Moment). In reviewing the data, for this piece of research, I’ve simplified the waves broadly into:

The B2B SaaS wave. Subscription-based, sales-led software that digitised business processes. Produced companies such as Salesforce, Netsuite, Xero, Wix.

The AI/ML wave. In the wake of the ChatGPT launch in late 2022 we can now see see two clear phases:

(1) The initial AI/ML wave. Companies where AI and machine learning form the foundation of the product, not a feature. The technology was genuinely hard to build.

(2) The current wave, generative and agentic AI. Beyond the frontier, foundation models are commoditising. Inference costs are falling roughly 10x per year. Generative AI tools have made it possible for non-technical founders to build software. Agentic AI systems that execute real work autonomously are creating entirely new business models. The barrier to starting up has collapsed. The barrier to scaling has not.

The dataset I dove into covers the first two waves outlined, with “success” criteria defined: 334 founders across 150 successful B2B SaaS companies (selected on the basis of $500m+ exits or IPOs), and 311 founders across 120 companies where AI/ML is core to the product from founding (and where the company has reached Series B+ funding, $100m+ exit, or $20m+ ARR). Within the AI/ML wave, 19 companies were founded in 2023 or later, after the emergence of generative AI. They are not a separate dataset; they are a subset of the 120 AI/ML companies, and we can look at them as a window into the earliest signals from the current wave.

What the data revealed

1. The evolving founder archetype

The B2B SaaS founder was a commercially credentialed operator: median age 33, ten years of professional experience, 73% of them bringing meaningful domain expertise. Half were repeat founders. A third had had a prior exit. 14% held MBAs, 6% PhDs. They knew an industry well enough to digitise it and had the network to sell into it. The technology was a means to an end.

The AI/ML wave founder was a fundamentally different profile. Among these founders, 18% held a PhD, which was triple the rate for SaaS. Meanwhile the MBA, held by 14% of SaaS founders, dropped to 4%. At company level, 27% of AI/ML companies had at least one PhD on the founding team. The research lab replaced business school as the place that produced the founders of the most promising software companies.

This shift made sense in context. When building a product required genuine AI or machine learning research capability, novel model architectures and bespoke training pipelines, research credentials were not decorative. They were years of practical training in the specific skills the moment demanded. An MBA was useful for a world where the commercial challenge was the founding bottleneck. In the initial AI/ML wave, the technical challenge was the bottleneck, and the commercial skills could be acquired later through accelerators or experienced co-founders.

61% of AI/ML founders attended a top-ranked research institution (one of the 25 global leaders such as MIT, Stanford, Cambridge, Oxford, Technion, ETH Zurich), compared with 37% of SaaS founders. The 24 percentage-point gap was the largest single finding in the entire dataset. This reflected where cutting-edge AI research concentrated and the network effects that flowed from it: co-founder relationships, recruiting pipelines, investor attention. But “attended” is the key word. Among AI/ML founders who attended top-ranked institutions, a meaningful proportion dropped out before completing their degrees, including founders at Perplexity, Pika Labs, Cognition and Legora. The signal was access to the research environment and the network, not the completed degree.

Among the post-2023 companies in our dataset, this pattern intensified. The dropout rate tripled, from 4% in the whole AI/ML cohort to 12% in the post-2023 subset. Five of the 43 post-2023 founders left their programmes before finishing. The entire three-person founding team at Mercor, all aged 21, dropped out. The opportunity cost of staying in education appears to be rising: when the tools exist to build now, finishing the degree is no longer the obvious choice.

What the credential shift tells us about the current wave is, I think, the most practically useful finding in the dataset. If the PhD premium was a function of the technology’s difficulty, and the technology is becoming more accessible, then the credential that matters is shifting again. The SaaS wave rewarded the MBA. The AI/ML wave rewarded the PhD. The current wave may reward something harder to credential at all: the ability to see where AI capabilities intersect with real-world problems, the speed to move from insight to product and the strategic vision to predict where a defensible competitive position could be carved out. We are already seeing this in the indicative post-2023 data, where the PhD rate of the application layer software companies has dropped from 18% to 12%.

Four distinct archetypes.

Looking at the characteristics of the founders from the AI/ML wave, four distinct archetypes emerged. Each was concentrated in different kinds of companies, each with a distinctive profile that you would recognise if you met them.

The Research Scientist

PhD or published ML researcher, late twenties to early thirties, left the lab with co-founders they had worked with for years. The product started as a research project before it became a company. These founders were concentrated in verticals where the technical challenge was genuinely at the frontier: Customer Service AI (60% ML researchers, 45% PhDs), Creative (39% ML researchers) and Productivity (36% CEO PhD rate). These were not companies that bolted a chatbot onto an existing workflow. They were companies where the AI formed the product, and building it required genuine research capability.

Synthesia is the archetype at its most successful. Founded in London in 2017 by two ML researchers, Matthias Niessner (PhD, specialising in neural rendering) and Lourdes Agapito (PhD, professor of computer vision at UCL), alongside a commercial CEO and COO. The research depth was the product’s foundation: Synthesia’s AI video platform, which generates realistic digital humans from text, required deep learning capabilities that simply could not be assembled from off-the-shelf tools. That research moat, combined with commercial co-founders, has produced one of Britain’s most valuable generative AI companies, now at a $4bn valuation with $150m+ ARR and 90% of the Fortune 100 as customers (TechCrunch, January 2026; CNBC, January 2026).

DeepL followed a similar pattern from Cologne. Dr Jaroslaw Kutylowski, a solo PhD ML researcher, founded a translation platform in 2017 that required frontier natural language processing capability to build. The company’s most recent valuation was $2bn. In both cases the founding insight was inseparable from the research: you could not have built these products without years of specific technical training.

The Research Scientist archetype persisted into the later part of the wave. Perplexity (2022), ElevenLabs (2022) and Pika Labs (2023) all followed the pattern of researchers leaving labs to build companies. But the archetype was most dominant earlier, when the technology was hardest and the researcher advantage was strongest.

Evaluating Research Scientist founders is an interesting challenge for investors. I’ve found that founders bamboozling investors is a very negative signal, whereas founders who have both world class academic credibility and the ability to articulate why their innovation is commercially important is a very positive one. An example of one I backed is Professor Hezy Yeshurun (Tel Aviv University PhD in maths and New York University postdoctoral research in computational neuroscience), at Forescout. As I spent time with him and diligenced his claims, using a technical experts to validate them, I found that they were solid. 

The Technical Builder

Highly technical but not a researcher. Software engineer or product builder who moved fast, shipped relentlessly and iterated based on what users did rather than what papers said. No PhD needed. The youngest archetype in the dataset, with a mean founding age of just 28.

The Technical Builder was concentrated in Developer Tools companies, where 96% of founders were technical, only 7% held PhDs, and the mean CEO age was 29. The distinction from the Research Scientist matters. Where the Research Scientist optimised for correctness and depth, the Technical Builder optimised for speed and utility. Where the researcher published, the builder shipped. Where the researcher’s founding moment was often “I have a breakthrough that could become a product,” the builder’s was “I see a problem that I can solve faster than anyone else.”

I can highlight two examples, both founded post 2022. Cognition (2023) can serve as a case study. Scott Wu and his co-founders, Walden Yan and Steven Hao, share backgrounds in competitive programming. They built Devin, an AI software engineering agent, with a team whose edge was not published research but an instinct for what developers need and the engineering skill to build it quickly. The company raised $900m within two years. Cursor (2022) followed a similar pattern: engineers building for engineers, shipping constantly, iterating based on how developers actually used the product.

This archetype already pointed toward where the next wave is heading. When the models are commoditised and the APIs are available, the advantage shifts from “can you build the model?” to “can you build the product?” The Technical Builder is optimised for exactly that transition.

The Domain Outsider

Not from the industry they were disrupting. Often lightly technical or non-technical, typically first-time, paired with a deeply technical co-founder. Saw an opportunity from the outside, often because insiders were too embedded in existing workflows to imagine how tech could replace them entirely.

This archetype was most visible in LegalTech (61% technical founders, 75% first-time, a remarkable 1.2 years of mean domain experience) and FinTech (70% technical, 80% first-time, mean age 28). That domain experience figure is worth pausing on. 1.2 years means many of these founders had essentially no prior exposure to the industry they chose to transform. They came in cold, studied the problem, and built.

Legora (2023) makes this tangible. Max Junestrand had no legal background. He met with hundreds of lawyers to understand their workflows, identified the research process as the bottleneck, and built an AI-native legal research tool from the outside. He’s a Y Combinator alumnus whose insight came from studying the problem with fresh eyes, not from years of working inside it.

The Domain Outsider challenges a widely held assumption among investors: that you need deep industry expertise to build software for an industry. In the B2B SaaS wave, domain expertise was genuinely essential. If you wanted to build HR software, you needed to understand how HR departments worked. The technology was a means to an end, and the domain knowledge was the edge. In the initial AI/ML wave, technical AI capability began substituting for domain knowledge in verticals where the processes were well-defined and the data was structured enough for AI to act on. As the technology becomes even more accessible, this archetype may become the most important of all.

The Domain-Technical Hybrid

The rarest and often the most formidable archetype.This founder archetype had both deep industry knowledge and genuine technical capability. They were older than the others, with a mean founding age in the mid-thirties and were found in verticals where the problem was complex, the stakes were high, and regulatory understanding mattered as much as engineering capability.

HealthTech was the clearest home for this profile: 72% technical founders, 23% PhDs, 4.4 years of mean domain experience, and a mean age of 36. Cybersecurity showed a different flavour: 86% technical, 8.1 years of mean domain experience, zero PhDs, and the oldest CEOs in the dataset at mean age 36. Having invested in cybersecurity across multiple cycles myself, from Forescout, Secerno and Clearswift through to SOCRadar today, I can confirm what the data shows: this is a sector where operational maturity and domain depth outweigh research credentials. You cannot learn cybersecurity from a textbook quickly enough to build a company in it. A great example is Huzeyfe Önal from SOCRadar (2019). A computer engineering graduate, he worked for large companies in various cybersecurity roles before becoming a cybersecurity consultant and founding two cybersecurity companies. Through this he gained massive domain knowledge and honed the vision which led him to create SOCRadar.

Abridge (2018) illustrates the hybrid at its best. Shiv Rao is a practising cardiologist who co-founded an AI medical documentation company with ML researchers from Carnegie Mellon, including Sandeep Konam and Florian Metze. The founding team was the archetype in miniature: a domain expert CEO who understood clinical workflows intimately, surrounded by technical co-founders with research-grade ML capability. Neither the domain knowledge nor the technical depth alone would have been sufficient. The combination was the point.

How the teams came together

60% of AI/ML wave founders had never founded a company before, compared with 50% for SaaS. Prior exit rates dropped from 32% to 20%. When the technology was new enough that deep expertise in it was rare, the people who had that expertise had a structural advantage over experienced operators who lacked the technical depth.

But first-time founders did not do it alone. Founding teams were larger (mean 2.6 versus 2.2 for SaaS), with solo founding collapsing from 25% to 10%. At company level, 38% of AI/ML companies included at least one founder with a prior exit. The experience gap narrowed through team composition: the first-timer brought the technical insight, then assembled a team with the commercial experience to build it into a business.

I have seen this dynamic up close. Some of the strongest founding teams we have evaluated at Oxx combined a brilliant first-time technical founder with a co-founder who had scaled a company before, someone who knew how to hire the first sales team, structure a board, or navigate an enterprise procurement process. The brilliance of the technical insight got the company to product-market fit. The experience around it got the company to go-to-market fit.

A great example here is Oxx portfolio company Gravitee (2015), founded by a technical team led by David Brassely. They were frustrated by the complexity of existing API management solutions. In time they understood that they needed experienced commercial leadership to scale so they brought in Rory Blundell, whom they knew from SnapLogic, initially as CRO and soon as CEO. Technical founders who handle this well have the confidence to know that while they are world class technologists, their companies will thrive if they bring on a peer who complements them in the commercial domain. Technical founders who handle this poorly bring on commercial subordinates.

The technical co-founder was present on 85% of SaaS teams and 95% of AI/ML teams. Building a software company has always required someone who actually builds software.

Every successful team needed to cover three capabilities: the technical ability to build, the insight into the problem worth solving, and the commercial or operational ability to turn a product into a business. How those capabilities were distributed varied, but all three needed to be present. 56% of AI/ML wave companies were co-founded by people who had worked together before, compared with 41% for SaaS. Academic bonds mattered nearly twice as much: 29% met at university versus 15%. The lab replaced the industry conference as the place where founding teams came together.

A third of AI/ML wave companies had an accelerator alumnus. Y Combinator alone accounted for 14% of the dataset (17 of 120 companies). For first-time founders without commercial networks, the accelerator became the primary launchpad, providing scaffolding that replaced the MBA and the corporate career ladder.

One thing connects both generations. The technical co-founder was present on 85% of SaaS teams and 95% of AI/ML teams. Building a software company has always required someone who actually builds software.

2. The map changed
– the UK and Continental Europe rise

The UK leapt from 1% of successful SaaS companies to 12% of AI/ML based companies. That is the single largest geographic shift in the dataset. Continental Europe rose from 5% to 10%.

This is personal for me. I have been investing in European B2B software for 30 years and making the case for European founders for nearly as long. In the early years while European technology was strong, the entrepreneurial role models were American, the capital was American and the European ecosystem was not there. The received wisdom was that you had to move to Silicon Valley to build a global software company. As entrepreneurial talent and a vibrant local ecosystem have emerged in Europe that assumption has been eroding steadily, and the data shows it clearly.

The ecosystem in Israel got there first – twenty years ago I saw no difference in terms of talent, skills and ambition between startups in Israel and their competitors in the US. Europe, most notably the UK, Sweden and France, have been there for ten or more years now.

The UK’s rise among AI/ML companies looks to be driven by the concentration of world-class AI teaching and research at UK universities (three of the world’s top twelve computer science universities are in the UK; QS World University Rankings, 2026). As a whole, Europe’s advantage is not just about universities. It also brings proximity to industries that AI will transform and the regulatory complexity that creates both barriers and moats for the companies that navigate it well. The European founder who understands GDPR, who has sold into regulated financial systems across three jurisdictions and who knows the procurement process of a German bank has an advantage that no Bay Area competitor can easily replicate.

Within the US, successful companies coming from the Bay Area held steady at around 44%. New York surged from 5% to 13%, becoming the clear second hub. Boston, home to MIT and Harvard, actually declined from 6% to 3%, a puzzling result given the concentration of world-class AI research. One hypothesis: the cities that gained offer denser accelerator ecosystems and stronger startup cultures that pull talent away from research centres even when the research originates there.

Among founders of US-headquartered AI/ML wave companies, 45% were internationally mobile, born in a different country from where they founded their companies. The equivalent figure for UK-headquartered AI/ML wave companies was 32%. The global talent pipeline into the largest English speaking countries was real and active. Anecdotally, the best companies I meet in the UK often have foreign-born founders.

Among the 19 most recently founded companies in the dataset (2023 or later), Europe’s  share increased still further to 26% (five companies, all from the UK or Sweden) compared with 22% of the full AI/ML based companies cohort. European-founded agentic AI startups Parloa, Lovable and Solve Intelligence collectively raised over $640m in recent rounds (New Market Pitch, March 2026). The geographic diversification is accelerating.

The numbers at a glance

• 645 founders · 270 companies · 3 broad waves of software
• 24pp increase in top-ranked institution attendance (SaaS 37% → AI/ML wave 61%)
• 3x PhD rate (SaaS 6% → AI/ML wave 18%)
• 3x MBA decline (SaaS 14% → AI/ML wave 4%)
• 60% first-time founders in the AI/ML company cohort
• 95% of AI/ML companies with a technical co-founder

Three-wave comparison

The post-2023 column shows the 19 most recently founded companies within the AI/ML cohort (a subset, not a separate dataset).

So who builds what comes next?

Up till now, we have described founders who have already built successful companies. It’s time to look ahead. The world outside our dataset has moved faster than any dataset can capture, and founder archetypes thrive according to the conditions in which they operate.

The ground has shifted

Vibe coding has changed who can build. In early 2025, Andrej Karpathy coined the term for what was already happening: people describing software in plain English and watching AI build it. Within a year it was Collins Dictionary’s Word of the Year (Collins Dictionary, 2025) and MIT’s Breakthrough Technology of 2026 (MIT Technology Review, January 2026). Lovable, a Swedish company in our dataset founded in 2023, built a vibe coding platform and hit $100m ARR within eight months (TechCrunch, 2025; multiple sources). 25% of Y Combinator’s Winter 2025 cohort reported codebases that were 95% AI-generated (State of Vibe Coding, February 2026). The barrier to building an AI product has collapsed. That does not mean the companies that built the initial AI/ML wave have lost their advantage. The companies in our dataset that invested years developing proprietary models, accumulating unique datasets and embedding AI deeply into customer workflows have built real moats, and many of them, including companies in the Oxx portfolio, are now building agentic capabilities on top of those foundations and finding real commercial success with them. But for new company formation, the starting point has fundamentally changed.

The engineering landscape is being restructured around AI. Software developer employment for ages 22 to 25 declined nearly 20% from its 2022 peak (Stanford Digital Economy Lab, cited in Stack Overflow, December 2025). Undergraduate Computer Science enrolment across the UC system declined in 2025 for the first time since the dot-com bust (SF Standard, February 2026). Microsoft reports AI now writes roughly 30% of its code. Google says over 25%. Meta’s Zuckerberg expects AI to write most of the company’s code within 12 to 18 months (SF Standard, February 2026; company earnings calls). Meanwhile, a LinkedIn and GitHub study found that AI coding tool adoption was associated with new hires needing fewer advanced programming skills (IEEE Spectrum, April 2025). The nature of “technical” is evolving in real time. Writing code from scratch is becoming less valuable. Directing AI systems, reviewing their output and making judgement calls about architecture and product is becoming more valuable.

The agentic AI market has exploded. Over $6bn in venture funding flowed into agentic AI companies in 2025 alone (Tracxn, March 2026). This is the “services as software” thesis in action, and it is where our investment thesis at Oxx is focused.

What the newest companies in our data confirm

When we isolate the 19 most recently founded companies in our AI/ML based company dataset (post the release of ChatGPT), a subset of the same 120 companies, the early signals correlate with what we see in the market. As this dataset is small these early signals are indicative only.

Teams are getting smaller again.

Mean team size dropped from 2.6 to 2.3. Solo founding ticked up from 10% to 16%. Two-person teams became more common (47% versus 41%). If AI development tools reduce the engineering headcount a founding team needs, this is exactly what you would expect.

The PhD rate dropped.

From 18% to 12%. The pure researcher archetype is less dominant. As AI capabilities become accessible through APIs and open-source models, the founding bottleneck is shifting from “can you build the model?” to “can you build the product and take it to market?”

The founder pool is broadening.

The first-time founder rate dropped from 60% to 51%. Alongside repeat founders, we see experienced operators and executives who may be founding for the first time but bring decades of commercial and scaling experience. The technology has matured enough that deep operational knowledge is directly applicable from day one, without needing to build the AI from scratch.

Teams are forming differently.

The “worked together before” rate dropped from 56% to 37%. The tight research-lab bonds that characterised the initial wave are less prevalent. Teams are assembling across different backgrounds, often more opportunistically. The binding agent is shifting from shared research history to shared conviction about a problem.

The technical bar is dipping.

The technical founder rate dropped from 77% to 70%, and companies with at least one technical co-founder dropped from 95% to 89%. Technical capability still matters. But what “technical” means is no longer synonymous with “can write production ML code from scratch.”

A fifth archetype may be emerging

Among the post-2023 companies, I noticed a pattern that did not exist in the initial AI/ML wave: experienced executives entering AI company building for the first time.

Sierra (2023) was founded by Bret Taylor at age 43 and Clay Bavor at 40. Both are deeply experienced technology executives, Taylor as former Salesforce co-CEO and Facebook CTO, Bavor as a former Google VP who led multiple product divisions. Hippocratic was co-founded by Munjal Shah at 50, a serial entrepreneur with Big Tech and healthcare experience, alongside a Chief Medical Officer and two ML researchers. Eudia by Omar Haroun at 37 with a decade of legal industry experience, paired with an experienced technical CTO from Google.

These are not the young researchers who defined the initial wave. They are seasoned operators who see AI as newly accessible and who bring commercial experience, industry networks and scaling capability that younger first-time founders typically lack. Call them the Seasoned Operator: someone with 15 or 20 years of experience who enters AI company building not because the technology is newly interesting, but because it has become newly possible for them to build with it. They understand the problem, the customer and the path to market. They pair with technical co-founders for the engineering layer. Time will tell how well they stack up against the Technical Builders and the Domain Outsiders.

The indicative post-2023 data supports this. The first-time founder rate is already dropping. The mean founding age holds steady even as more experienced operators enter. If this pattern continues, the competitive landscape of the current wave will look meaningfully different from the initial AI/ML wave.

Four theses on who builds next

1. The researcher ceiling.

The Research Scientist dominated the initial wave because the technology was hard enough that deep expertise was a genuine advantage. As models increasingly commoditise, that advantage narrows. The post-2023 PhD drop from 18% to 12% aligns with the external evidence: when 25% of a YC cohort can produce 95% AI-generated codebases, the barrier to entry has structurally changed. Researchers will remain important within teams, particularly in verticals where proprietary model design is a real moat. But the space where a PhD is necessary to build an AI product in the application layer is shrinking.

2. The builder’s moment.

The Technical Builder may define what comes next. Developer Tools already showed this profile: 96% technical, 7% PhD, mean age 28. As AI-assisted development lowers the threshold, execution speed becomes the differentiating factor. Lovable reached $100m ARR in eight months. When two people can ship what previously required five engineers, the companies that win will be the ones that move fastest from insight to product to market.

3. The domain expert returns.

If AI capabilities become pluggable through APIs, the founder who understands the workflow and who can wrap AI around that knowledge may be the archetype that scales most effectively. This is already happening. MagicSchool AI, in our post-2023 cohort, was founded by Adeel Khan, a non-technical founder with 13 years in education. He understood the classroom, the teacher’s workflow, the specific places where AI could remove friction. He did not need to build a model. He needed to understand the problem deeply enough to build the right product around one. That would have been nearly impossible in the initial AI/ML wave. It is possible now.

Imagine a clinician who has spent years navigating the inefficiencies of medical documentation, who understands precisely where time is wasted and errors occur, and who can now describe the tool they need in plain English and watch it get built. Imagine a procurement specialist who knows the supplier management workflow intimately, who can articulate exactly what an AI agent should automate, and who has the relationships to sell it into the organisations that need it. This is not hypothetical. Hippocratic AI, also in the post-2023 cohort, was co-founded by Munjal Shah at age 50 alongside a Chief Medical Officer, precisely because he understood that the healthcare workflow problem required clinical credibility that no 25-year-old engineer could bring, no matter how technically gifted. These people have always existed. What has changed is that the technology has caught up to their insight.

Having invested through every wave of software, I find this the most compelling shift of all. In the SaaS wave, domain expertise was the edge. In the initial AI/ML wave, it was displaced by technical depth. In the new wave, domain expertise may return to the centre, but wielded differently: not by operators digitising their own industry with off-the-shelf tools, but by people who understand a problem deeply and now have AI-native capabilities to build the solution.

4. The ecosystem around the founder matters more, not less.

If building is no longer the hard part, then what is? Scaling. Going to market. Pricing. Hiring. Entering new markets. Building a board. Navigating the transition from product-market fit to go-to-market fit. The founding team matters enormously. But who surrounds the founding team determines whether a product becomes a business.

As the founder profile broadens, as more first-time builders, domain experts and seasoned operators enter, the need for this kind of support grows. The technology barrier has dropped. The commercial scaling challenge has not. The accelerators, advisory networks and investors who understand the specific dynamics of the scale-up stage will be the ones who help the most promising companies survive the transition from product to lasting business. Crassly, it is also this support that will help later stage investors build confidence in a 30 year old CEO’s ability to build a company.

And one constant.

Across both earlier waves, every vertical and every geography, successful software companies had technical co-founders. 85% in SaaS. 95% in the AI/ML wave. 89% in the post-2023 cohort. The nature of “technical” evolves. The need for it does not.

What this means

If you are a founder building empowered by the rich AI capabilities today, the data from the first two waves tells you this: build a team with complementary archetypes. Make sure at least one of you is technical. If you are a first-timer, surround yourself with people who have done the scaling before, whether that is a co-founder with a prior exit, an accelerator programme, or an investor and advisory network that will roll their sleeves up and work alongside you. The technology will not be your constraint. The commercial execution will be.

If you are an investor, the data tells you this: the signals you relied on in the initial AI/ML wave are already shifting. The PhD premium is narrowing. More experienced operators are entering. Teams are smaller and forming more opportunistically. The archetypes that built the initial wave are not the same archetypes that will build the next one. Update your pattern recognition, and invest in the infrastructure, the networks, the operating expertise, that helps the new archetypes scale.

If you are building in Europe, the data tells you this: the geographic shift is real and accelerating. Europe’s share increased from 6% of successful SaaS companies to 22% of AI/ML companies. Europe has the talent (40% more developers than the US; Stack Overflow), the institutions (four of the top twelve global CS universities; QS World University Rankings, 2026) and the market complexity that creates defensible businesses. What it needs is capital with conviction, and investors who understand the specific challenges of scaling B2B software companies from Europe to the world.

An example here is Oxx portfolio company Funnel, which when we first invested already had a good number of customers located in the US, serviced remotely from Sweden. The company needed to invest locally in customer success to support them and in an outbound sales motion in order to expand larger accounts. We advised that in order to succeed organisationally and culturally a very senior executive, ideally a founder, should relocate from Sweden to lead the establishment of a company’s US entity; the company’s co-founder and CMO duly did so. We recommended Boston as the office location and connected the company with several others that had made the same move to validate it as well as with executives (including the founding CMO of Hubspot who then became a board member) and local service providers. The result is that Funnel has successfully established a significant US presence and was voted as one of America’s top 100 places to work in Newsweek.

The software founder is not a fixed type. Every wave reshapes the archetype. The SaaS wave rewarded domain-rich operators. The initial AI/ML wave rewarded researchers and technical depth. The next generation of software companies will reward something harder to credential: the ability to see where AI intersects with real problems, the speed to build, and the self-awareness to assemble the team and the ecosystem that turns insight into a lasting business. The founders who will define this wave are already out there. The question is who will back them.

This research was produced by Oxx, based on a dataset of 645 founders across 270 companies. The AI/ML cohort (120 companies, 311 founders) includes application-layer AI companies where AI/ML is core to the product from founding. The B2B SaaS cohort (150 companies, 334 founders) includes B2B software companies with $500m+ exits or IPOs. Post-2023 figures (19 companies, 43 founders) are a subset of the AI/ML cohort. All findings are correlational and subject to survivorship bias. Data as of 27 February 2026.Download the full dataset, data tables and methodology here.