Seven years ago, the category that now dominates enterprise technology investing did not exist. Agent Development — companies building the tools that make AI agents work — went from zero percent of the Enterprise Tech 30 in 2019, according to an analysis by CrewAI citing the ET30, to 43 percent of the entire list in 2026, according to the eighth annual ET30 published by Wing Venture Capital. The vote that produced that result came from 98 venture capitalists across 85 firms managing a combined $2.6 trillion in assets under management. They were not betting on a vision. They were confirming one.
The ET30 is not an editorial list or a pay-to-play ranking. It is a structured annual vote by investors who fund the companies they are evaluating. That methodology is what makes the category movement significant: when a category grows from zero to 43 percent of a $2.6-trillion investor vote in seven years, the signal is not a company's self-assessment — it is where capital is actually moving.
"The theme for 2026 is agents moving from demo to production," said Peter Wagner, a founding partner at Wing Venture Capital and the creator of the ET30. "Last year, we saw the rise of agentic applications. This year, those agents are being deployed and doing real work across legal, accounting, insurance, IT, and customer support."
What the ET30 data shows structurally is a market organizing itself around a dependency chain. The Giga tier — companies with more than $1 billion in capital raised, including Anthropic, OpenAI, and Databricks — represents infrastructure. The Late tier — companies between $150 million and $1 billion — represents agents as the product, companies selling AI systems that do work. The Early tier, where CrewAI sits alongside LlamaIndex, E2B, and seven other companies, is building the agentic toolbox: orchestration frameworks, sandboxed execution environments, evaluation tooling, and tool-calling primitives.
That early-tier focus on primitives is worth examining. "The stack is still early," the ET30 report states — not as a hedged observation but as a data-driven conclusion drawn from where seed-stage capital is concentrated. Sixteen ET30 companies are now building AI agents in some form, spanning customer service (Decagon, Sierra), go-to-market workflows (Clay, Gong), finance (Ramp), and enterprise search (Glean). But the tooling underneath those agents is being built by a separate, earlier cohort.
The AI-native signal is also stark. Seventy-two percent of the 2026 ET30 companies are AI-native — meaning AI is their core product or delivery mechanism — up from 3 percent in 2022. In four years, the share of AI-native companies on the list increased 24-fold. The companies that are not AI-native are increasingly the exception, not the baseline.
CrewAI, which posted about its second consecutive ET30 selection last week, is using the list placement to make a narrower claim: that multi-agent orchestration has won the architectural debate. "Last year, the vote was a bet on the vision," the company said in a blog post. "This year it's a bet on the proof." The proof it cites is 2 billion agentic executions processed and Fortune 500 production deployments. Those are real numbers, but they are self-reported, and the blog post is a company celebrating itself — which is exactly what a company blog post should do. The more durable evidence is the investor vote.
There are structural cautions worth naming. The ET30 is a list of private companies selected by venture capitalists, which means it reflects the views of investors who tend to fund venture-backed companies. It systematically underweights sectors where venture-backed disruption is structurally difficult — regulated financial services, defense, healthcare — and it captures a phase of the market that may not reflect what is happening at the earliest seed stage or at the largest enterprise tier. The list is a window into a specific slice of the market, not the whole picture.
Open source is another cross-cutting signal. The ET30 report notes that open source appears at higher rates across every stage compared to prior years and is "becoming a default distribution model for AI infrastructure and developer tools." For agent infrastructure specifically, that matters: if the tooling layer is open source, the competitive moat for any single vendor in the orchestration or primitives space is thinner than it appears from a company blog post.
The clearest structural read from this year's ET30 is that the agentic stack is splitting into two layers. At the top, vertical agents — purpose-built for legal, accounting, customer support, and GTM workflows — are raising later-stage capital and winning enterprise contracts. At the bottom, the infrastructure primitives that make those agents work are still being built and battle-tested. The investors who voted this year funded both layers simultaneously, which is a coherent bet: you need the tools before you can widely deploy the products.
What to watch next is simpler: whether the early-tier tooling companies can convert their presence on lists like this into enterprise contracts before the vertical agent companies consume all the available budget. The vote is a signal. The enterprise sales cycle is the test.
† The proof it cites is 2 billion agentic executions processed and Fortune 500 production deployments. [Source-reported; not independently verified.]
