AI-First Companies: How AI-Native Firms Are Dismantling Traditional Industries
By Dirk Röthig | CEO, VERDANTIS Impact Capital | 07. March 2026
Cursor hit $1.2 billion in annual recurring revenue in 2025. Harvey, a two-year-old legal AI startup, is now valued at $8 billion and used by 100,000 lawyers. Perplexity multiplied its valuation 40-fold in under two years. These are not flukes. They are the leading edge of a structural shift that is rewriting the rules of every industry these companies touch.
Tags: Artificial Intelligence, Startup, Disruption, Business Strategy, Competitive Advantage
The Distinction That Changes Everything
There is a category error at the centre of most AI strategy discussions today. Executives ask: "How much AI should we adopt?" They install AI tools on top of existing workflows, run pilots, measure time savings, and declare a success. What they fail to see is that the companies threatening their market share are not doing the same thing faster. They are doing something categorically different.
An AI-first company is not a traditional company that uses AI. It is a company that was designed — from its founding architecture, its hiring philosophy, its data strategy, its pricing model, and its organisational structure — around the assumption that intelligence is cheap, scalable, and continuously improving. This distinction, which sounds philosophical, has profoundly practical consequences in every revenue line and every competitive confrontation.
The numbers make the gap brutally clear. AI-native startups are growing at a 100% median annual rate. Traditional SaaS companies — even the successful ones — average 23% (Deepstar Strategic, 2025). That is not a marginal advantage. It is a 4.3-fold compounding divergence that, sustained over even three years, produces outcome differences that cannot be closed by hiring more salespeople or increasing the marketing budget.
Dirk Roethig, CEO of VERDANTIS Impact Capital, has observed this transformation across multiple sectors and investment stages. The pattern is consistent: companies built with AI as their foundational layer outpace traditional competitors by an order of magnitude. This is not because they have superior engineering talent or access to better models. It is because their entire operating system is calibrated to a different assumption about what work AI can perform and how that work should be priced, packaged, and delivered.
Four Companies That Redefined What Is Possible
The fastest way to understand what AI-first actually means is to look at the companies doing it.
Cursor: The Developer That Became a Platform
Cursor, an AI-powered code editor built by Anysphere, went from $100 million to $1.2 billion in annual recurring revenue in 2025 — an 1,100% year-on-year increase that makes it the fastest-growing SaaS product in recorded history (Sacra, 2025). It achieves $3.3 million in revenue per employee, a figure that traditional software companies would consider science fiction. The product works not because it adds AI to an existing editor, but because it was built as an intelligent agent from day one, understanding intent rather than executing commands.
Cursor's disruption target is the entire developer tooling industry — a market dominated for decades by companies like JetBrains, Microsoft, and IBM. Those incumbents have AI features. Cursor has an AI philosophy. The difference is showing up in enterprise adoption and churn rates.
Dirk Röthig has noted that the Cursor case represents a fundamental reordering of how software companies should think about their product-market fit. Rather than building tools that extend human capability by 10 or 20 percent, Cursor builds tools that replace entire categories of human decision-making. This is not an incremental innovation. It is a categorical shift in the value proposition.
Harvey: The Law Firm Without Lawyers
Harvey was founded in 2022 with a single, audacious premise: that large-language models trained on legal knowledge could handle the most expensive work in professional services. By the end of 2025, the startup had reached $195 million in annual recurring revenue, growing 3.9 times from $50 million the year before (TechCrunch, 2025). In December 2025, Harvey raised $160 million at an $8 billion valuation, with technology deployed across some 100,000 lawyers at firms including A&O Shearman and Latham & Watkins.
What makes Harvey AI-first is not that it uses GPT-4 or Claude. It is that the entire business model is premised on replacing billable hours — historically the most protected economic unit in professional services — with AI throughput. Every traditional law firm charges more as it grows. Harvey's cost of service approaches zero as it scales. The competitive math is impossible to resolve in favour of incumbents.
This model, which would have seemed impossible five years ago, now appears inevitable to anyone paying attention. Dirk Röthig has discussed Harvey's model in the context of European professional services, where the resistance to such disruption is particularly high. The cultural attachment to seniority-based billing and the notion that certain professional work cannot be delegated to technology remains a significant barrier to adoption. Yet Harvey's growth demonstrates that this barrier, while real, is not impermeable. The question is not whether AI will handle legal work. The question is how quickly incumbents can reorganise their models around that reality.
Perplexity: Forty Times in Twenty-Four Months
Perplexity AI went from a $500 million valuation in January 2024 to a $20 billion valuation in September 2025 (Leaveit2ai, 2026). It did this with approximately 250 employees and no advertising budget. The product: an AI-native answer engine that treats every search query as a reasoning problem rather than a keyword-matching exercise. Its disruption target is Google — a company with 180,000 employees, decades of search infrastructure, and nearly $280 billion in annual revenue.
Whether Perplexity ultimately challenges Google at scale is debatable. That it has forced every major search and information product to rethink its fundamental model is not. That is the pattern: AI-native startups do not need to win to cause disruption. They need only to occupy enough of the future to make the incumbent's present strategy untenable.
Dirk Röthig has observed that Perplexity's valuation rise, while mathematically impressive, understates the strategic damage it is inflicting on information providers across the ecosystem. Google's search advertising model depends on users clicking through to websites. Perplexity's model depends on users staying within the application and accepting the AI-generated answer. This is not a competitive difference. It is a competitive extinction event for every business model built on the assumption that Google's search dominance is permanent.
Klarna: When the Incumbent Goes AI-First
Klarna is a case study in an existing company choosing the AI-first path aggressively — and experiencing both its benefits and its complications. The Swedish fintech used AI to reduce its workforce from approximately 5,000 to 3,000 employees while simultaneously doubling its revenue (CNBC, 2025). CEO Sebastian Siemiatkowski's prediction: fewer than 2,000 employees within four years.
The Klarna case is instructive precisely because it is not straightforward. When Klarna initially automated customer service at scale, some quality metrics declined, and the company brought human agents back for certain tasks (Fortune, 2025). The lesson is not that AI-first does not work — it is that the transition from legacy operations to AI-native operations involves real friction, and managing that friction requires strategic intentionality rather than headline-driven optimism.
Dirk Röthig has highlighted Klarna's experience as evidence that the AI-first transition is not frictionless for incumbents, even when they commit seriously to it. The challenge is not technical. It is organisational. An incumbent with 5,000 employees operating on a legacy cost structure cannot simply replace 60 percent of its headcount with AI and expect the organisation to function seamlessly. The culture, the training systems, the management structures, and the incentive models are all built around a labour-intensive operation. Reorienting those systems toward an AI-native model requires time and carries real risks. Klarna is managing those risks better than most, but even Klarna is not immune to them.
The Economics of AI-Native Business Models
The financial profile of AI-native companies differs from traditional technology firms in three structurally significant ways.
First, revenue per employee is radically higher. AI-native startups average $3.48 million in revenue per employee, compared with roughly $580,000 for traditional SaaS companies (Deepstar Strategic, 2025). Midjourney, the AI image generation company, operates with a small team and generates approximately $2 million per employee (Dealroom.co, 2025). OpenAI runs at approximately $1.5 million per employee against $3.7 billion in ARR. These figures represent not just operational efficiency but a fundamental rethinking of what the relationship between headcount and output can be.
This economic advantage is not temporary. It is not the result of selecting a favourable market moment or having exceptional founders. It is structural. When the product is intelligence, and intelligence is scalable without proportional increases in headcount, then revenue per employee necessarily diverges from the historical norm. This divergence compounds. A company generating $3.48 million per employee can invest in R&D, marketing, and sales with a fundamentally different cost basis than a company generating $580,000 per employee. Over three years, that difference becomes unbridgeable.
Second, time to scale is compressed by orders of magnitude. Traditional SaaS companies needed 5 to 7 years and 200+ person teams to reach $100 million in annual recurring revenue. AI-native companies are reaching the same milestone in 12 to 18 months with fewer than 20 employees (Menlo Ventures, 2025). This compression destroys the incumbent advantage of experience and market entrenchment. By the time a legacy player recognises the threat, the challenger has already crossed the threshold where momentum becomes self-sustaining.
Dirk Röthig has emphasised that this time compression is perhaps the most strategically significant aspect of AI-native business models. In traditional competitive dynamics, incumbents have time to respond. They can hire consultants, launch strategic initiatives, and gradually shift their product roadmaps toward new threats. The 5 to 7 year timeline of traditional SaaS growth meant that incumbents generally had 18 to 36 months to recognise a threat and begin a response. With AI-native companies compressing that timeline to 12 to 18 months, and with much of that time consumed by fundraising and hiring, incumbents often do not fully perceive the threat until the challenger is already beyond the point where a catch-up effort makes economic sense.
Third, the value proposition is categorically different. Traditional enterprise software optimises workflows. AI-native companies own decisions. Harvey does not make legal work faster — it replaces the need for junior lawyers to do first drafts, research, and contract review. Cursor does not make coding faster — it replaces the entire cognitive loop of a mid-level developer for a significant fraction of tasks. This distinction means that the value captured is not proportional to time saved but proportional to decisions replaced.
This distinction has profound implications for pricing, for customer acquisition, and for the way that value cascades through organisations. When you optimise a workflow, you capture perhaps 20 to 30 percent of the value you create — the rest accrues to the customer. When you replace a decision or a worker category, you capture far more of the value you create, because you have eliminated the need for expensive human judgment. This is why Harvey can command such high prices despite being a two-year-old startup. It is not charging for a feature. It is charging for the elimination of an entire layer of professional services labour.
Which Industries Are Most Exposed?
The disruption pressure is not evenly distributed. The industries most vulnerable are those characterised by high information intensity, large volumes of repetitive expert judgment, and pricing models built on time and headcount rather than outcomes.
Legal services, financial analysis, medical diagnosis, and software development sit at the highest exposure end of the spectrum. Healthcare is particularly striking: AI spending in that sector hit $1.4 billion in 2025, nearly tripling 2024 figures, with AI-native companies beginning to displace incumbents in imaging, diagnostics, and clinical documentation (Menlo Ventures, 2025). The medical imaging market, worth approximately $8 billion globally, is now experiencing direct competition from AI companies that can read radiographs, CT scans, and MRI images at speeds and accuracies that equal or exceed human radiologists.
Industries with heavy physical infrastructure — logistics, manufacturing, construction — are more protected in the short term, though even there, AI-native companies are targeting the planning, optimisation, and decision layers that sit on top of physical operations. A logistics company cannot eliminate its trucks with AI, but an AI-native logistics company can eliminate 40 to 60 percent of its route planners, demand forecasters, and operations coordinators — precisely the path that Klarna has pursued.
The pattern I have observed, both as an investor and as someone who speaks regularly with operators in these sectors, is that the disruption almost always begins at the margin — a task that incumbents consider too small, too messy, or too low-margin to defend — and then expands inward until it threatens the core.
Dirk Röthig has documented this pattern across multiple European industries. In German manufacturing, for example, the first applications of AI-native approaches appeared in quality control and predictive maintenance — areas where incumbents had not built particularly strong product positions and where the costs of getting it wrong were relatively contained. Within 18 months, those same AI-native companies had expanded into production scheduling, supply chain optimisation, and ultimately into the core engineering and design processes that had traditionally been the domain of highly skilled German engineers. The speed of that expansion caught many incumbents unprepared.
What Incumbents Get Wrong
The most common mistake incumbents make is treating AI-first disruption as a technology problem that can be solved by technology procurement. They buy a licence to an AI tool, integrate it into existing systems, and measure the outcome against existing KPIs. This is the wrong frame.
The threat from AI-native companies is not that they have better tools. It is that they have built their entire value chain — including pricing, hiring, data collection, product development, and go-to-market motion — around a model of the world in which AI capability is the central operating assumption rather than a feature addition. Responding to that with tool procurement is like responding to the emergence of e-commerce by adding a website to a mail-order catalogue business.
The second mistake is underestimating how quickly AI-native business models scale once they establish product-market fit. Incumbents benchmark their response timelines against their own historical experience: a traditional enterprise software company might expect 24 to 36 months to develop a credible AI-augmented product and begin customer deployment. In that 24 to 36 month window, an AI-native startup can go from concept to market dominance in a vertical. By the time the incumbent's AI feature ships, the challenger is already consolidating market share and erecting switching costs that make displacement economically irrational.
Dirk Röthig has written about the broader urgency of AI adoption in the European context — particularly for the German Mittelstand, where the cultural and structural resistance to AI-first thinking is most acute. The challenge is not knowledge. Eighty-eight percent of senior leaders report using AI in at least one business function (McKinsey, 2025). The challenge is organisational identity: who is willing to redesign their company around the assumption that AI will handle most of the cognitive work? For many Mittelstand companies, particularly those built on the reputation and relationships of their founders or long-tenured leaders, this identity shift is psychologically difficult. It requires relinquishing the notion that human judgment, experience, and intuition are irreplaceable. For companies that have built their brands and their customer relationships on precisely that foundation, accepting that AI can replicate or exceed that value is existentially threatening.
The Road Ahead
Global AI spending is projected to reach $2 trillion in 2026 (Insight Global, 2026). That figure represents not just investment but the scale of the competitive reordering now underway. The companies that have built AI-native models are not approaching this reordering as a threat to manage. They are approaching it as the opportunity to define the next industrial era.
For incumbents in every sector, the relevant question is not whether AI-native competitors will attack their market. They already are. The question is whether the response — cultural, structural, and strategic — comes quickly enough to be relevant. In industry after industry, the evidence from 2024 and 2025 suggests that the window for a meaningful response is narrower than most boards currently believe.
The competitive pressure from AI
Über den Autor: Dirk Röthig ist CEO von VERDANTIS Impact Capital, einer Impact-Investment-Plattform für Carbon Credits, Agroforstry und Nature-Based Solutions mit Sitz in Zug, Schweiz. Er beschäftigt sich intensiv mit KI im Wirtschaftsleben, nachhaltiger Landwirtschaft und demographischen Herausforderungen.
Kontakt und weitere Artikel: verdantiscapital.com | LinkedIn
Top comments (0)