Analysis
Venture capital is placing bets across every layer of agentic commerce at once, from multipurpose personal assistants to narrow shopping tools to the payments infrastructure underneath all of them, Newcomer reported in a survey of the category's best-funded startups.
The Multipurpose Agents
- Instinct -- a general-purpose personal agent that has closed a major recent funding round; users report success securing dinner reservations and movie tickets, but the agent is also prone to costly errors -- one customer reported losing roughly $300 when Instinct canceled a flight incorrectly, a concrete illustration of what happens when an autonomous agent's mistake carries real financial consequences.
- Town -- a similarly multipurpose agent with shopping capabilities that has also raised a significant recent round, competing directly with Instinct for the same broad "agent that handles your errands" positioning.
The Focused Shopping Assistants
- Daydream -- founded by Julie Bornstein, previously an executive at Stitch Fix; a personalized AI shopping search engine now embedded directly into retailer websites as of late July, backed by Forerunner, Index Ventures, GV and True Ventures.
- Phia -- co-founded by Phoebe Gates, an AI deal-finding shopping assistant that has raised roughly $43 million total from Kleiner Perkins, Khosla Ventures and Notable Capital; the company has faced allegations of cookie-stuffing in its underlying deal-finding technology, a real reputational and legal risk for a company whose entire pitch is consumer trust.
The Infrastructure Layer
- Catena Labs -- builds digital identification infrastructure specifically for AI agent transactions, solving the "how does a merchant know this agent is actually authorized to buy on this person's behalf" problem.
- Basis Theory -- creates secure virtual cards purpose-built for agent payments, letting a business issue spend-limited, revocable card credentials to an AI agent rather than exposing a real payment method.
- Ramp -- has launched its own AI agents for business procurement and fraud detection, extending agentic commerce into B2B spend management rather than consumer shopping.
Why the Demand Signal Is Real
This isn't speculative capital chasing a narrative with no underlying behavior change: Adobe Analytics found 41% of shoppers used AI for online shopping as of June, and a Retail Dive/Rithum survey found 53% of shoppers trust AI recommenders as much as they trust a brand's own website. Those numbers are the demand-side justification investors are underwriting against, and they're the same data Profound is selling brands tools to respond to from the other side of the same shift.
The Risk Investors Are Pricing Differently
The infrastructure-layer bets -- Catena Labs, Basis Theory -- carry a fundamentally different risk profile than the consumer-facing agents. A payments-and-identity company profits regardless of whether Instinct, Town, Daydream or some future entrant wins the branded-agent race, the same picks-and-shovels logic that made payment processors durable business across every prior wave of e-commerce disruption. The consumer-facing agents, by contrast, are betting their entire user-trust proposition on error rates that are still visibly high enough to produce real financial losses -- Instinct's $300 flight-cancellation mistake is exactly the kind of incident that, repeated at scale, becomes a company-ending trust problem rather than a rounding error.
What to track: whether any of the consumer-facing agents publish real error-rate or dispute-resolution data, since none currently do, and that's the single number that would separate a durable agentic-commerce business from one riding a funding wave ahead of its own reliability.
