Two years ago, investors backed AI companies on the strength of a team and a thesis. That playbook is dead, according to Socheat Chhay, Managing Director of Sopra Steria Ventures, the corporate venture arm of one of Europe’s largest IT services groups.

In an interview published today by Tech Nation, Chhay laid out the new criteria: proprietary data, defensible algorithms, and deep domain expertise. “It’s not enough to build a vertical wrapper product on top of an existing model,” Chhay told Tech Nation. “The question is: do you have proprietary data which will produce outputs that nobody else can replicate?”

The shift is happening against a backdrop of record capital deployment. Crunchbase data shows Q1 2026 set an all-time record with $300 billion invested across roughly 6,000 startups globally, with 80% of that capital flowing to AI. The mega-rounds tell the story at the top: OpenAI’s $122 billion at $852 billion valuation, Anthropic’s $30 billion Series G, Mistral AI’s $14 billion empire.

The Wrapper Problem

Chhay’s sharpest critique targets AI wrapper companies that bolt LLM integrations onto existing products and claim an AI-native valuation premium.

“Every founder now sprinkles AI as a powder over their company to enter into AI investment thesis and gain valuation premium,” Chhay told Tech Nation. Sopra Steria distinguishes between three categories: a SaaS company augmented by AI, an AI wrapper, and a company with a genuine AI-first moat. The third category is what gets funded.

Chhay pointed to a specific example (unnamed) of a well-funded legal tech company built largely on top of a foundational model without a unique data layer. “With a new free Claude legal plug-in released in the market, we don’t know what’s going to happen to them,” he said. That reference lands differently this week, after Anthropic launched a full Claude-powered legal software suite with 20+ MCP connectors and 12 practice-area plugins for law firms.

Services Over Software

The investor shift Chhay describes aligns with a thesis that went viral last month: Sequoia Capital partner Julien Bek’s argument that the next trillion-dollar company won’t sell software, it will sell outcomes. As Fortune reported, Bek’s framework plots services on two axes: intelligence versus judgment, and outsourced versus in-house. The sweet spot for AI-native firms is high-intelligence, low-judgment work that companies already outsource: insurance brokerage, tax advisory, payroll, compliance, simple legal services.

Bek calls these startups “autopilots,” borrowing from aviation. A human still monitors the system and handles the hardest decisions, but the process is largely automated. The distinction from “copilots” is operational: autopilot firms deliver the entire service, not just a tool that assists a human worker.

Chhay confirmed the same pattern from the investment side. “Anything that a services company does that can be replaced without human intelligence is the opportunity,” he told Tech Nation. The verticals he is actively investing in: manufacturing and supply chain operations, fintech infrastructure, legal and compliance workflow automation, and DevOps engineering services.

Agent-Native Economics

The economics have changed in ways that favor agent-native companies. Chhay noted that AI-native startups can reach unicorn status in roughly five years with limited headcount and high profitability, compared to the ten-year SaaS playbook. Founders are requesting smaller rounds because they are “replacing people who can have their tasks automated with tokens which outputs more than it could cost them in the future,” according to Tech Nation.

For agent builders specifically, the bar is defensible automation IP plus proven use-case depth. An agent framework or orchestration layer alone is not defensible. The proprietary value comes from the data and workflows the agents operate on: the customer-specific context, the domain-specific decision trees, the accumulated operational knowledge that competitors cannot replicate by plugging into the same foundation model.

The Implication for Agent Startups

The funding criteria Chhay and Bek describe create a specific filter for the agent ecosystem. Agent infrastructure plays (orchestration, deployment, observability) compete on execution speed and developer experience, but face commoditization pressure as cloud providers add native agent support. The defensible position is vertical: agents that accumulate proprietary operational data within a specific domain, where switching costs grow with every automated workflow.

The companies that will raise in 2026 are not the ones with the best prompt engineering. They are the ones with proprietary data flywheel that makes their agents measurably better at a specific job than any general-purpose alternative. That is a harder thing to build than a wrapper, which is exactly the point.