Enterprise AI spending went from $1.7 billion in 2023 to $37 billion in 2025. That is a 22x increase in two years, and that is procurement data — not a forecast, not a projection, not venture-funded hype. Businesses bought it.
So the question is not whether AI is reshaping enterprise software. It is who is capturing the spend. The Menlo Ventures 2025 State of Generative AI in the Enterprise report gives you a specific answer: AI-native startups now hold 63% of application-layer market share, up from 36% in 2024. That is the largest single-year share shift in the report’s history.
This article is not going to hand you a verdict. The companies generating these results — Cursor, Harvey, ElevenLabs, Lovable — are structurally different from traditional SaaS, and the Jasper AI collapse is a reminder that “AI-native” is not a guarantee. What follows is the data, the company trajectories, the cautionary cases, and a category map of who the AI-native challengers actually are. This article is part of our comprehensive SaaS reckoning guide, which covers the full landscape from the market dislocation to the practical CTO playbook.
What does the data actually say about AI-native growth vs. SaaS incumbent performance?
Quick source note before we get into it: Menlo Ventures is a VC firm with an investment interest in AI-native companies. The survey covers about 500 US enterprise decision-makers who have already adopted AI tools — so it skews toward AI-positive outcomes. With that on the table, the scale of these shifts is difficult to pin on sample bias alone.
Enterprise generative AI spend hit $37 billion in 2025, up from $11.5 billion in 2024. AI-native startups took 63% of the application layer. The mechanism driving this is Product-Led Growth: PLG accounts for 27% of all AI application spend, nearly four times the 7% rate in traditional SaaS. Individual users are driving AI adoption at 4x the rate of traditional software purchasing — reaching enterprise budgets before procurement is even involved. AI deals are converting at 47% versus the traditional SaaS pipeline-to-close rate of 25%. Buyers are moving faster because value is demonstrable earlier.
Category-level data sharpens the picture further. AI-native startups hold 78% of the sales software category, 91% of finance and operations, and 71% of product and engineering. Infrastructure-layer incumbents still hold 56% share where reliability and integration depth matter. But at the user-facing application layer, the shift is documented and it is accelerating.
The companies that prove AI-native is genuinely different — not just better-funded
These are the top performers from a broader population that includes plenty of failures. What matters is not just the growth rate — it is the mechanism behind it.
Cursor (Anysphere): $1 billion ARR in fewer than 24 months, around 300 employees, $29.3 billion valuation at Series D. Revenue per employee: approximately $3.3 million. Salesforce generates approximately $800,000 per employee. Cursor competes against GitHub Copilot — backed by Microsoft — and won by shipping repo-level context and multi-file editing before Copilot did, using a model-agnostic architecture that let developers adopt frontier models immediately.
Harvey AI: $100 million ARR as of August 2025, 500+ enterprise customers, $8 billion valuation — up from $3 billion to $5 billion to $8 billion inside a single year. Harvey raised $760 million in 2025 alone, with ARR largely driven by usage as the tool embeds in daily legal workflows.
ElevenLabs: Over $330 million ARR at end of 2025. $500 million Series D in February 2026 at an $11 billion valuation — more than 3x higher than one year earlier. Enterprise customers include Deutsche Telekom, Square, Revolut, and Meta.
Lovable: Unicorn status in 8 months. $200 million Series A at $1.8 billion valuation, 45 employees, approximately $1.7 million ARR per employee. Investors include the founders of Klarna, Slack, and HubSpot — incumbents betting on the challengers.
A note on valuations: SaaStr data from 3,001 primary rounds shows the top 1% of AI startups command 3–10x normal multiples at equivalent ARR stages. Cursor at $29.3 billion sits above even that — valued as critical AI infrastructure, not a SaaS business. These are winner-take-all investor expectations and not all will materialise. But the revenue efficiency ratios are verifiable and do not depend on investor sentiment to be meaningful.
Jasper AI and the cautionary lesson: why not every AI-native company wins
Jasper is worth reading carefully when you are evaluating AI-native vendors — not because it is a representative outcome, but because the failure mode is specific, identifiable in advance, and absolutely not limited to 2022.
Jasper peaked at approximately $90 million ARR and a $1.5 billion valuation in 2022. By 2024, ARR had declined to an estimated $55–88 million range and both co-founders had stepped down.
The failure mode is what Jasper’s case is really about — the wrapper trap. Jasper’s core product was a text generation interface on top of GPT-3 with no proprietary training data, no fine-tuning, and no workflow lock-in. When ChatGPT launched with comparable capability at lower cost, users could replicate Jasper’s functionality by pasting prompts directly into ChatGPT. Differentiation collapsed.
Compare that to the companies in the previous section. Harvey has a proprietary data flywheel from millions of legal documents. Cursor is embedded in the developer’s daily IDE workflow. ElevenLabs has voice model fine-tuning not available through generic APIs. None of them is a thin interface on a commodity model.
When you are evaluating an AI-native vendor, ask yourself this: does this company’s value proposition survive if OpenAI, Anthropic, or Google ships a direct version of this product’s core feature tomorrow? If the answer is yes, the risk profile mirrors Jasper’s. If the answer involves proprietary data, deep workflow integration, or category-specific fine-tuning, you are looking at a structurally different beast.
Who are the AI-native challengers in your SaaS categories?
This is a category map, not a buying guide. Its purpose: identify which parts of your stack have active, well-funded AI-native challengers — and which vendor relationships carry meaningful obsolescence risk. Verify the metrics before you make any decisions; this landscape moves faster than any article can track.
Developer tools: Cursor ($1B ARR, $29.3B valuation, ~300 employees) vs. GitHub Copilot (Microsoft). Lovable ($75M+ ARR, $1.8B valuation, 45 employees) targets a different wedge — non-technical users for no-code app building. Two distinct approaches to the same category.
CRM and sales: 78% AI-native share — the highest documented category displacement. AI-native companies like Clay win by attacking research, personalisation, and enrichment workflows that sit outside the Salesforce data model. DayAI is an AI-native CRM replacement in this category. The incumbent’s defence is Salesforce Agentforce, which we get into in the next section.
Customer support: AI-native tools captured $630 million in enterprise spend in 2025. Decagon and Sierra represent distinct approaches to AI-native customer service. The Klarna case below is the most detailed evidence of what aggressive deployment in this category produces at scale.
ERP and finance: 91% startup share — the highest of any category. Rillet is cited directly in the Menlo report as an AI-first ERP challenger competing against QuickBooks and the broader mid-market ERP category. Campfire and Numeric are additional challengers.
Legal: Harvey AI ($100M ARR, 500+ customers, $8B valuation) vs. established legal software platforms. Legal vertical AI captured $650 million in enterprise spend in 2025, growing from near-zero in 2022.
Can incumbents adapt fast enough — the Salesforce Agentforce test case?
Salesforce Agentforce is the most directly relevant incumbent AI adaptation for anyone with CRM exposure — because its success or failure has the clearest read-across if you own Salesforce licences.
The data: per analyst estimates from Q3 FY26 results, Agentforce ARR grew approximately +114% to roughly $1.4 billion. Enterprise deals are closing. The caveat: the +114% figure needs context — the base was small, making the percentage notable while the absolute figure remains modest.
The structural challenge is harder to hand-wave away. Agentforce is built on the Salesforce CRM architecture. There are real benefits — existing customer data, enterprise security, the Salesforce ecosystem — but there is also Salesforce’s path dependence. Every architectural decision made before 2020 constrains what it can do today. As Convequity puts it: “Organisational inertia, path dependence, and diluted talent density leave incumbents structurally disadvantaged compared to startups with small, elite teams.”
Microsoft Copilot has the OpenAI partnership and Office 365 distribution at a scale no AI-native startup can match — but Cursor’s growth against GitHub Copilot suggests distribution does not override architectural advantage where specialisation matters.
The honest position: whether “AI features on a pre-AI CRM architecture” can replicate AI-native performance economics is not yet answered. For a complete overview of the SaaS reckoning — including how the broader market dislocation connects to incumbent adaptation — the pillar article in this series covers the full strategic picture.
What does Klarna’s experience tell us about the limits of aggressive AI replacement?
Klarna is the efficiency case and the cautionary case in enterprise AI adoption — and it is the most thoroughly documented example we have as of early 2026.
The company deployed an AI customer service agent handling the workload of 700–853 agents. Workforce declined approximately 50% through attrition as AI scaled. Revenue per employee grew from approximately $300,000 to $1.3 million — according to Klarna’s own reporting.
Then the reversal. Klarna reported customer satisfaction decline, reversed elements of the strategy, and began rehiring. CEO Sebastian Siemiatkowski acknowledged the overclaim directly: “People were very angry with me for saying that.” Klarna now positions human support as “VIP treatment” — recognising that customers value direct human contact for complex edge cases.
The read-across for anyone not running a global fintech: replacement decisions made without outcome monitoring and a reversal pathway create avoidable risk, regardless of how good the underlying tools are. Klarna had the runway to reverse course. Not every business does.
Both things are true here. The efficiency gains are real. The customer experience degradation at scale is also real.
What should a sceptical CTO conclude from all this evidence?
The evidence does not support a single verdict. Here is what it actually shows.
The market evidence is structural: Enterprise AI spending grew 22x in two years. AI-native startups captured 63% of the application layer. These are procurement decisions, not forecasts. Ignoring this trend is a higher-risk position than engaging with it selectively.
The company-level evidence is structural, not narrative: Cursor’s $3.3 million ARR per employee versus Salesforce’s $800,000 is a business model comparison with direct competitive implications. The mechanism — AI-native architecture enabling different unit economics — matters more than the headline growth rate.
The cautionary evidence is specific and actionable: Jasper AI represents the wrapper trap — thin differentiation plus commodity model dependence — identifiable in advance. Klarna represents replacement risk without outcome monitoring. Both failure modes are specific, not generic.
The incumbent evidence is genuinely unresolved: Salesforce Agentforce has commercial momentum, but whether incumbent architecture can match AI-native performance economics remains open. The evidence as of early 2026 supports neither “incumbents are doomed” nor “incumbents will catch up easily.”
The practical question to ask yourself: in which specific categories of your stack is AI-native architecture producing demonstrably superior outcomes, and which of your current vendors is in a structural position to adapt? The evidence here builds the case for taking that seriously. The Bain four-scenario framework for evaluating vendor survival is in the vendor framework article. The CTO audit playbook is in the downstream article. The broader strategic picture of what CTOs should do in response to the AI-native challenger surge is in the SaaS reckoning overview for this cluster.
Frequently Asked Questions
What is an AI-native company and how is it different from a SaaS company that adds AI features?
An AI-native company builds its core product architecture around AI from inception. A traditional SaaS company adds AI features on top of existing architecture designed before AI capabilities were available. Cursor vs. GitHub Copilot makes this concrete — Copilot adds AI to GitHub’s existing platform; Cursor was designed from the start as an AI-first development environment. That architectural difference produces structural differences in cost structures, distribution models, and revenue-per-employee ratios. It is not just a branding distinction.
Why are AI-native startups winning market share at the application layer so fast?
Three structural mechanisms. Product-Led Growth — 27% of AI app spend vs. 7% in traditional software — meaning individual users create bottom-up enterprise demand before procurement is involved. Superior unit economics enabling competitive pricing. And AI-native architecture delivering outcomes incumbents cannot match at equivalent cost. PLG is the most counterintuitive one. Users adopt it, it spreads, procurement follows. It is the reverse of how enterprise software has always worked.
Is the AI startup boom real or another bubble?
The enterprise spending data reflects actual procurement decisions — CFOs are not buying $37 billion in AI software speculatively. Market demand is real. Specific valuations contain speculative elements — some will compress. The honest answer distinguishes between “is enterprise AI real?” (yes) and “are all current valuations justified?” (some are not). Both can be true at the same time.
Why does Cursor have a $29 billion valuation with only 300 employees?
$1 billion ARR, $3.3 million revenue per employee, and investor expectations of dominant market share in developer tools. Cursor is valued as critical AI infrastructure — comparable to Databricks at approximately 40x ARR — not a conventional SaaS business. Whether those expectations materialise is a different question.
What happened to Jasper AI and could the same thing happen to other AI startups?
Jasper built a text generation product on top of GPT-3 without proprietary data, fine-tuned models, or workflow lock-in. When ChatGPT launched with comparable capability at lower cost, differentiation collapsed. Peak: approximately $90 million ARR, $1.5 billion valuation in 2022. Decline: estimated $55–88 million ARR in 2024; both co-founders departed. The same risk applies to any company whose core value proposition is entirely deliverable by foundation model improvements. Companies with proprietary data flywheels (Harvey), deep workflow integration (Cursor), or category-specific fine-tuning (ElevenLabs) have a structurally different risk profile.
Is Salesforce Agentforce genuinely competing with AI-native CRM alternatives?
Analyst estimates show Agentforce ARR growth of approximately +114% to roughly $1.4 billion — commercial momentum is real. But Agentforce inherits both the benefits and constraints of Salesforce’s pre-AI architecture. Whether “AI features on existing CRM architecture” can match AI-native CRM performance economics is not yet answered. Keep a close eye on this one.
How should I evaluate whether an AI-native startup will survive long enough to replace my current vendor?
Four signals: proprietary data or model assets — not just API access; depth of workflow integration creating switching costs; ARR trajectory and revenue-per-employee as proxies for durability; and commoditisation risk if foundation model capability improves. Apply the wrapper test: if OpenAI or Anthropic shipped this product’s core feature tomorrow, would the value proposition survive? If it would not, you know what you are dealing with.
Are AI-native startups really better than established SaaS platforms?
It depends on the category. For developer tools, legal AI, and voice AI — the evidence supports AI-native superiority for those specific workflows. For ERP and core financial systems — the evidence for displacement is less developed. Rillet and other ERP challengers are early-stage relative to Harvey or Cursor. “Better” is a category-specific question, not a universal one.
Should I replace my SaaS tools with AI-native alternatives right now?
Not wholesale and not immediately. The Klarna case shows aggressive AI replacement without outcome monitoring creates customer experience risk even when the efficiency gains are real. The evidence-based approach: identify which categories in your stack have AI-native challengers with documented performance advantages, evaluate those tools, and build in outcome monitoring and a reversal pathway before full migration. The methodology is in the stack audit playbook.
What does revenue per employee mean and why does it matter for evaluating AI-native startups?
Revenue per employee is annual recurring revenue divided by full-time headcount. AI-native: Cursor approximately $3.3 million, Lovable approximately $1.7 million. Established SaaS: Salesforce approximately $800,000, most SaaS companies $200,000–$400,000. The gap reflects AI-native architecture enabling automation of work that traditional SaaS companies handle with more headcount. It is a proxy for durability — a company with high revenue per employee can price competitively and survive downturns better than a headcount-heavy incumbent.
