The Capability Glut: AI Agent Funding Outruns Selection

May 2026 produced 29 new unicorns, and AI services and robotics led the class. That single data point from Crunchbase News tells you something important about where the AI agent market stands right now: capital is flooding into the build side of the equation at a pace the buy side cannot match. More agents are being created, funded, and deployed every week than any buyer has the tools to evaluate. The selection problem — finding, vetting, and hiring the right AI agent for a specific job — is not being solved. It is getting harder.

This is the structural tension defining the AI agent marketplace in mid-2026, and this week's headlines make it impossible to ignore.

Vertical Agents Are Getting Funded. Marketplaces Are Not.

Sandstone's announced $30M raise to bring AI agents into in-house legal teams is the cleanest example of where agent infrastructure investment is actually going. The pitch is vertical capability: a purpose-built agent-for-hire solution targeting a specific buyer segment — legal departments — with a specific workflow need. That is a legitimate and valuable bet. Legal teams have document-heavy, high-stakes work that maps well to agent task completion.

But notice what the Sandstone model does not fund: the layer above it. A general counsel evaluating whether to deploy Sandstone, a competing legal AI platform, or a more composable agent workflow still has no structured marketplace to help them compare outcomes, verify track records, or audit task-completion rates across vendors. They are left with demos, sales decks, and reference calls — the same selection process that existed before agents existed.

This pattern repeats across every vertical getting funded right now. The Crunchbase data on active May investors — a16z, Y Combinator, and General Catalyst all deploying capital at pace — shows concentration in capability build-out, not in the selection infrastructure that would help buyers navigate the resulting landscape. When every major fund is writing checks to capability, the gap compounds.

Scale Without Selection Is Fragmentation

Lovable's announcement that it has hit $500M in annualised revenue with one million new projects created every week is a genuine milestone. It is also a signal about what happens to the AI task marketplace when agent-assisted creation scales without a corresponding selection layer.

One million new projects per week means the surface area of deployable agents, tools, and AI-built products is expanding faster than any buyer's ability to survey it. Buyers are not becoming better at hiring AI agents; they are simply being exposed to more of them. That is not the same thing. Discovery without curation is noise.

The App Store analogy is instructive here. Apple spent years building recommendation infrastructure — and just this week rolled out personalised recommendations and streaming-style subscription bundles to a storefront that already had ratings, reviews, editorial curation, and category structure. A mature marketplace solved the selection problem incrementally over more than a decade. The AI agent marketplace is attempting to skip that infrastructure entirely — scaling supply while leaving demand-side tooling at version zero.

Proliferation Closes the Window

Apple's moves this week are not cosmetic updates to a stable storefront. Rolling out personalised recommendations, launching subscription bundles, and simultaneously threatening to remove low-engagement apps are three signals of the same underlying pressure: even a platform with 1.8 million apps and a decade of curation infrastructure has hit a discovery ceiling. Unassisted browsing no longer works. Apple is engineering its way around the selection problem because the volume of micro-tools has made human-scale evaluation untenable — and it is still struggling.

Now map that pressure onto Lovable's one million new projects per week. At that pace, the number of agent-capable micro-tools in circulation is compounding faster than any buyer, procurement team, or IT department can track. The window for evaluating agents through direct comparison is closing — not because buyers are getting worse at evaluation, but because the supply side is growing at a rate that makes exhaustive comparison structurally impossible.

When even Apple needs algorithmic curation to surface useful tools from a crowded marketplace, the argument for a dedicated agent-selection layer — one that matches buyers to verified outcomes rather than to capability lists — becomes structurally unavoidable. Proliferation does not expand buyer choice in any meaningful sense. Without a selection mechanism, it collapses it.

The Selection Gap Is the Market Opportunity

Every headline from this week's funding and growth data points to the same structural gap: AI agent capability investment is outpacing AI agent selection infrastructure by a widening margin. Vertical agents will keep getting funded. Platforms will keep scaling supply. And buyers will keep making high-stakes decisions about which agents to hire with tools that haven't materially improved since the market began.

That gap is not a footnote. It is the defining market problem of 2026 — and the clearest signal yet that outcome-first, selection-layer marketplaces are not just useful, they are overdue.

If you're evaluating AI agents for real work, the question is no longer "does this agent exist?" It's "how do I know it will actually perform?" That's the question worth demanding an answer to.