The Sweet Science: Why The AI Era Belongs To Middleweights

A $60B Vote of Confidence

The market tends to assume that big companies will capture the biggest gains from AI. However, AI is tough to get right at any size. Look at Klarna, valued at $6 billion in 2024, which made headlines claiming its OpenAI-powered chatbot could handle millions of conversations and do the work of 700 customer service employees. Customers hated the rollout, and by 2025, Klarna was hiring back humans.

Why Cursor?

Each year we speak with thousands of operators and founders, and one pattern is clear: The biggest long-term gains from AI will not flow to heavyweight incumbents or many AI-native startups but to scrappy middle-market technology companies, the middleweights. The next phase of AI disruption will be challenging, but middleweights can gain serious ground.

The Road Ahead

One objection is that hyperscaler companies will go after certain verticals. If Copilot inside Microsoft 365 or Salesforce 1 agents can run a workflow, how does a middleweight company survive? The answer depends on what constitutes durable advantage. Horizontal platforms are built for generalized work, not the messy, regulation-heavy, category-specific workflows of the real world. Middleweights can win by making their software the system of record that AI calls into instead of software that AI replaces. The odds for making big, impactful gains with AI right now favor the middle market, where proven growth companies can use customer trust, domain expertise, capital structure and speed to transform their businesses, taking market share from slower incumbents. With three-quarters of AI's economic gains now being captured by just 20% of companies, per PwC, entrepreneurs who stand still may already be losing the round.

What makes for a winning middleweight company? The best middleweight technology companies share the five traits below, all working together as a system. Disciplined self-assessment. Middleweights are designed to act quickly on honest feedback, and their boards help them test where AI generates value versus where it merely consumes engineering capacity and budget. Seat-based pricing is one area for brutal assessment. When autonomous agents do the work, the revenue model should reflect outcomes, not users. In 2023, customer service platform Intercom made a bold switch, pricing its AI agent Fin at 99 cents per resolved conversation. That agent became the company's core offering, and it recently sold to Salesforce for $3.6 billion. Outcome-based pricing might seem painful at first (and reorganize your GTM team and their incentives), but it anticipates an agentic future.

Agility. Enterprise companies are weighed down by technical debt and legacy infrastructure. Middleweights have enough scale and proprietary data but not so much organizational mass that every experiment needs 10 layers of approval. Their agility is as much cultural as structural. These are ambitious, scaling companies growing 20% or more with strong unit economics, and a tech-first mindset runs through the entire business, not just the engineering org. Take the restaurant software Toast, where early AI gains came from product leads using AI to cut documentation and process work; those product teams then built a flywheel connecting new product features to external communications, with LLMs continuously editing and improving instructions for AI agents.

Workflow ownership. In the AI era, the strongest moat is owning a complex workflow. Middleweights have spent years gaining this position — integrating into customer systems, accumulating exception-level data, learning operational nuances that take a claims process from 95% accurate to 99.5%. (The last 4.5 points are the moat.) An FTV Capital company, Agiloft, doesn't just apply AI to contracts; its moat is absorbing the decision workflow around each contract. As a contract moves through approvals, negotiations and redlines, the important part is learning from the history of why internal teams decided the way they did. Well-positioned companies will hold the institutional memory that AI agents need to query to do their jobs.

Technical capacity. Most large companies are stuck in AI pilot purgatory, and the market still underestimates how operationally demanding AI deployment is. Middleweights have something most AI-native startups lack: years of working with real customers. ReliaQuest, another FTV company, started in 2007 as a service-heavy cybersecurity business that has learned deep detection logic from operating in more than 1,000 customer environments, including some of the largest global enterprises. As the company saw rapid automation from machine learning, then more sophisticated AI, it moved in-house SOC analysts into higher-value product development roles, allowing engineers with deep expertise to focus on the most critical tasks.