Why AI Isn’t Killing SaaS Yet

Ara Karazian, lead economist at Ramp, uses data from 50,000 businesses and $100 billion in annual spend to challenge the dominant 'SaaSpocalypse' narrative — the idea that AI will wipe out SaaS companies, collapse seat-based pricing, and consolidate everything around a handful of frontier model providers. He breaks SaaSpocalypse into two testable claims: (1) that businesses are shifting away from traditional SaaS to AI model companies, and (2) that the way businesses buy software is fundamentally changing toward token or agentic pricing models. He argues neither is supported by the data. Seat-based contracts still represent 65-75% of spend, flat platform fees account for 20-30%, and token-based pricing uptake — even at companies like HubSpot and Adobe that offer it — is under half a percent of spend. Companies like Figma continue to grow despite competition from AI-native alternatives like Anthropic's design product.

On the AI model market itself, Karazian discusses the Ramp AI Index, which tracks business adoption of AI models. Anthropic has recently overtaken OpenAI as the most-used model among businesses on Ramp's platform. However, he cautions against interpreting this as a winner-take-all dynamic, noting that early adopters are increasingly using multiple models simultaneously. A significant trend is rising cost-consciousness: for high-intensity token spenders, AI token costs have increased 13x over the past year, which Karazian calls unsustainable. This is driving some businesses toward routing platforms like OpenRouter, which allow access to cheaper or open-source models. DeepSeek, despite initial hype, never exceeded 1% adoption on Ramp's platform and faces ongoing security perception issues. Google's Gemini is underrepresented in paid adoption data because it is bundled into Google Workspace for free.

Karazian identifies several areas of genuine growth that often go unreported in mainstream AI discourse. Answer Engine Optimization (AEO) — software that helps firms track how they appear in AI model outputs — is a fast-growing category that didn't exist before and is being led by new entrants like Profound, not legacy SEO players. AI-native CRMs like Adio are growing quickly, though they are far from unseating Salesforce. He also notes that many of the fastest-growing vendors on Ramp's platform are infrastructure, workflow, and application-layer companies, not the frontier model labs themselves.

On the question of AI's impact on employment, Karazian is cautiously optimistic. He observes a decoupling of revenue growth from headcount growth at software companies, but also notes that firms actively adopting AI tend to be fast-growing and still have substantial work for employees. He is skeptical of the 'AI will destroy all jobs' framing promoted by model companies, suggesting it functions more as TAM-expansion rhetoric aimed at investors and capital allocators than as a genuine economic forecast. He also discusses how legacy players like Deloitte are taking a more conservative stance on AI adoption compared to competitors, and how sectors like journalism may be quietly experimenting with AI in ways that complement rather than replace human reporting work.