The Agentic AI Cohort - When AI Stops Answering and Starts Acting: The $500 Billion Profit Pool Nobody's Pricing Right

Every major enterprise software company just said the word "agent" on their earnings call. Salesforce counted 3.2 trillion tokens flowing through its LLM gateway. Cloudflare told investors it's building the internet agents will run on. Rubrik launched a product literally called "Agent Cloud" that can undo what autonomous software breaks. That's not a buzzword cycle. That's a profit pool forming in real time. The consensus view says AI kills SaaS. Seats go away, revenue compresses, and the whole category deflates like a punctured tire. I've spent the last several weeks pulling apart the latest earnings releases, contract metrics, and pricing disclosures from 17 companies across the stack, and the picture that emerges is more nuanced and more investable than the bear case assumes. The better framing isn't "AI kills SaaS." It's profit pool migration. Classic software becomes plumbing. The money moves to whoever controls the layers that run, secure, connect, and feed data to autonomous agents. Some of these companies are already showing it in the numbers. Others are faking it until they make it. And a few are watching their retention metrics crack in slow motion. Here's how I'm mapping the opportunity, where the evidence is strongest, and which names carry the most risk relative to what the market's already paying. The Stack: A Highway System for Autonomous Software Think of the agentic AI stack like a highway system. At the bottom you've got the roads and bridges (data infrastructure). In the middle you've got the traffic control and toll booths (security, connectivity, orchestration). And at the top you've got the vehicles (applications, seat-based software). When traffic was light, meaning humans doing tasks manually, the roads didn't need much. But autonomous agents are like adding ten million self-driving trucks to the highway overnight. Suddenly the toll booths, traffic controllers, and bridge inspectors become the most valuable infrastructure in the system. That's what's happening across enterprise software right now. The companies building toll booths and traffic control are compounding. The ones selling parking spots are scrambling to reinvent themselves. The causality chain is straightforward. Agents increase workload intensity: more API calls, more machine identities, more automated changes, more logs. That concentrates value at four control points: orchestration and governance (who approves what agents do), security (who protects against what agents break), connectivity (how agents reach everything), and data (what agents need to know). Meanwhile, seat-based pricing gets mechanically stressed because automation increases output per human, which means charging per person captures less of the value being created. This isn't theoretical. It's showing up in reported metrics right now. The Compounding Layers: Where the Evidence Is Strongest Orchestration: Palantir and ServiceNow Palantir (PLTR) just posted its best quarter as a public company. Q4 2025 revenue hit $1.41 billion, up 70% year over year. U.S. commercial revenue grew 137%. The Rule of 40 score reached 127, which is nearly unprecedented at this scale. FY2026 guidance calls for 61% revenue growth, with adjusted free cash flow projected between $3.9 billion and $4.1 billion. The Palantir thesis is that it's the "trusted action layer." As AI shifts from answering questions to executing decisions, the governance, audit, and integration layer becomes the gating function. Willingness to pay concentrates there. The counterweight is that the market already knows this: at roughly $131 per share, Palantir trades at a valuation that embeds extraordinary expectations. ServiceNow (NOW) is positioning itself as the "AI control tower" for an agentic world. Q4 2025 showed total revenue of $3.57 billion (+20.5% y/y), with remaining performance obligations of $28.2 billion (+26.5% y/y). That RPO number is the real signal. It means enterprises are committing