Software is dying. Not the code, the business model. For twenty years, the tech industry has been a giant game of "counting heads," a lucrative racket where companies charge $150 a month for every human being sitting in a chair clicking a mouse. But humans are slow, expensive, and they need vacations.
Databricks CEO Ali Ghodsi thinks the party is about to end. In his view, the "AI agent" isn't just another feature to be bolted onto a side menu. It’s an extinction-level event for the SaaS giants we’ve spent two decades enriching.
The logic is brutally simple. If an AI agent can log in, scrape the data, run the analysis, and send the email, why do you need five junior analysts sitting in Salesforce seats? You don’t. You need one person to watch the machine, and maybe not even that. This creates a massive, gaping hole in the revenue of companies like ServiceNow, Workday, or any other vendor that bills by the "user." If the user isn't a human, the pricing model collapses.
"The old world was: how many employees do you have? Give me $100 per employee," Ghodsi recently remarked. It was a predictable, steady stream of cash. But agents don't have heartbeats. They don't need a login or a sleek UI. They just need an API and enough compute to keep the lights on.
The friction here isn't just about efficiency; it's about a fundamental clash of interests. SaaS companies are currently sprinting to rebrand themselves as "AI-first." Salesforce launched Agentforce. HubSpot is talking about "Agentic CRM." They’re trying to pivot, but it’s like a landlord trying to charge rent to the ghosts haunting the building. If their software actually works—if the agents truly take over the labor—the seat-based revenue disappears. They are effectively building the very products that will cannibalize their bank accounts.
Look at the math. A standard enterprise license for a top-tier CRM might run a company $3,000 per user, per year. Multiply that by ten thousand employees. That’s a $30 million check written every year just for the privilege of letting humans enter data into boxes. Now, replace those humans with a fleet of autonomous agents that cost a fraction of that in tokens and server time. The CFO isn't going to keep paying for ten thousand seats out of nostalgia. They’re going to cut the check to whoever provides the intelligence, not the interface.
This is where the cynicism gets real. Ghodsi isn't just a disinterested observer; he’s the guy selling the pickaxes for this specific gold rush. Databricks wants your data to live in their "Lakehouse" so you can build those very agents. He’s essentially telling the world that the "skin" of software—the buttons, the tabs, the colorful dashboards—is worthless. All that matters is the data and the model that chews through it.
The industry is currently obsessed with "hallucinations" and whether an AI can accurately summarize a meeting. That’s a distraction. The real fight is over who owns the workflow. For a decade, SaaS companies won because they owned the "System of Record." They were the digital file cabinets of the corporate world. But if agents can bypass the file cabinet and go straight to the raw data, the cabinet becomes an expensive piece of furniture.
We’ve seen this movie before. In the 90s, we bought software in boxes. In the 2000s, we rented it through a browser. Now, the browser itself is becoming a bottleneck. Why navigate a clunky UI when you can just tell a bot to "fix the supply chain leak in the Midwest"?
The transition will be messy. Companies will fight to keep their "moats" by locking down their APIs or charging "agent fees" that mirror the old seat prices. It’ll be a desperate, litigious crawl to maintain margins. We’re already seeing the tension. Salesforce is trying to charge $2 per "agent conversation." It feels like a stopgap, a way to tax the future before it arrives.
If Ghodsi is right, the tech "landscape"—a word I hate but one that fits this scorched-earth scenario—will look unrecognizable in five years. The giants that dominated the cloud era are suddenly looking like the hardware vendors of the 80s: necessary, perhaps, but ultimately commoditized.
Software used to be a tool for people. Now, software is becoming the worker. And workers don't pay for subscriptions.
Which leaves us with a singular, uncomfortable thought: If the value of software is no longer tied to the people using it, what exactly are we paying for?
