Jason Lemkin says if growth isn't accelerating, you're not an AI company — and PE has abandoned B2B SaaS

Summary

Jason Lemkin, founder of SaaStr, argues that AI adoption has become binary. If a company's growth isn't accelerating, it's not meaningfully deploying AI. The metric separates real implementation from performative announcements across both public and private companies.

Lemkin applies this test to major public software vendors. Meta accelerated growth through AI-powered ad matching, making it a genuine AI play. Microsoft missed on the software side despite AI momentum in other divisions, signaling that agentic products alone don't drive value if revenue doesn't follow. ServiceNow and Salesforce have time to reaccelerate, but the clock is running. He's grown impatient with companies that talk AI without showing revenue lift.

The startup landscape has fractured into extremes. Vibe-coded products proliferate so rapidly that differentiation has collapsed. A successful seed investor recently told Lemkin: "I'm giving up because everyone can vibe code something. I can't even tell the difference." Yet the competitive bar has simultaneously climbed. Investors now expect companies to grow from $1M to $100M ARR within a year, nearly unprecedented velocity. That same velocity advantage means bootstrapped operators can survive without venture capital if they can assemble a lean, elite team. But the pace of software development is now so fast that a founder without four or five exceptional engineers will get crushed by YC cohorts, funded or not. Atlassian co-founder Michael Cannon Brooks once told Lemkin: "I was lucky I had five extra years." That window doesn't exist anymore.

The middle market is hollowing out, echoing patterns already visible in media. At the bottom, vibe coding enables niche software for micro-markets like web broadcasters, dental practices, and specialized workflows that previously had no software options. At the top, distribution and platform control matter: Salesforce, Google, Figma. A company between $50M and $800M ARR faces an existential problem. Private equity has effectively abandoned B2B SaaS. Lemkin says it plainly: PE has just said goodbye to B2B. A $140M-ARR company he advises found itself unable to exit, even on attractive terms, because larger PE-backed platforms aren't interested in acquisitions at that scale. Companies that would have commanded $20M to $30M in revenue exits five years ago are now on the block with no buyers. The playbook of reaching $20M ARR, showing efficiency, and getting acquired at 5x to 10x revenue has died.

At SaaStr, Lemkin is running 20 agents that replaced eight people. He uses AgentForce alongside YC companies Artisan and Qualified, plus Clay, which recently raised at a $5B valuation. AgentForce handles reactivations by scoring lapsed customers and sending outreach with a 70% open rate. Critically, the agent closes deals on Saturday nights, a task no human sales rep wants. One agent closed a six-figure deal on a weekend. The concrete payoff is compression: eight sales people down to one human plus agents, freeing high-touch sellers to hunt larger deals.

But even within SaaStr's own agent stack, vendor fragmentation persists. Different agents excel at different use cases, and it's unclear whether dominance is possible yet. No single platform owns the agentic layer.

AI discoverability is reshaping vendor selection in ways most companies haven't internalized. Lemkin asked Replit what the best CRM for him to use was, and it said HubSpot. When he built a web app in Lovable and Replit, agents steered him toward Resend over SendGrid because SendGrid's free tier had been throttled into uselessness. He's never switched back. WorkOS, an authentication vendor, went years struggling, then suddenly exploded because every agent recommends it. Vendors like Clay, which powers agent selection logic, are benefiting from this discovery shift. Agents, not Google, are now the initial touchpoint.

GEO tools are attempting to monetize this discovery layer, but Lemkin dismisses most as scammy because they demand credit cards upfront before trial. SaaS companies that let agents try products free and recommend them based on actual functionality are seeing explosive adoption.

On niche enterprise AI, Lemkin is skeptical of unit economics alone. A visual AI system for optimizing farm yields is interesting, but if the ROI and pricing haven't shifted beyond legacy software benchmarks, it won't compound to something meaningful. The bar is steep: the AI agent must be so powerful and so ROI-positive that a company can charge 4x to 10x what it charged before. Older-generation SDR software like SalesLoft exited to Vista at $2.5B. Modern AI SDRs like Artisan, Qualified, and Clay cost around $100 per user per month, far higher than legacy tools, but the math works because they deliver outsized ROI. Without that leverage, scale stays limited.

Growth acceleration is the only metric that matters in AI. Revenue must follow, otherwise companies are shipping theater. The middle market is doomed unless they own the agentic layer. At the bottom, vibe coding has opened tiny markets that were never worth addressing before, a thousand flowers blooming fast.