Skip to main content
On Black Friday 2025, no human typed a product query and got distracted by a notification. No one abandoned a cart because checkout was confusing. The traffic that surged that day came from agents — and it was not a rounding error.
“AI agent traffic surged 805% year-over-year on Black Friday 2025, and AI agents influenced $67 billion in global Cyber Week sales — 20% of all orders.”
805% year-over-year. $67 billion. One in five orders. These are not projections from a consulting deck. They are closed transactions. The Business-to-Agent economy — B2A — did not arrive as a trend to watch. It arrived as a revenue line that already printed.

​
The context

The most durable signal of structural shift is capital allocation. In YC’s Spring 2025 batch, 46% of funded startups — 67 of 144 — were explicitly building agentic AI solutions across 18 different categories. This is not a vertical bet on one flavor of AI. It is a horizontal thesis that the agentic layer cuts across every domain: legal, logistics, devtools, commerce.
“In YC’s Spring 2025 batch, 46% of funded startups (67 of 144) were explicitly building agentic AI solutions across 18 different categories.”
When YC funds 67 companies in a single batch to build the infrastructure and applications of the agent economy, the question is no longer “will B2A matter?” The question is “what do agents need that does not exist yet?” The $67B Black Friday data point answers the urgency. The YC portfolio answers the direction. The uncomfortable implication: if you are still optimizing your growth strategy for human attention, you are already one cycle behind.

​
What I tried / what I saw

Agents break every assumption traditional marketing was built on. Human buyers carry loyalty, loss aversion, brand affinity, and inertia. These are not bugs in consumer behavior — they are the entire surface area that traditional sales and marketing was built to exploit. Agents carry none of them.
“An agent will choose the vendor with the best-structured API response, the lowest price at the moment of query, and the clearest machine-readable capability description.”
There is no relationship to nurture. No brand story to tell. No emotional hook to set. An agent evaluating vendors at query time runs a pure capability and price comparison against machine-readable signals. The vendor with the clearest structured output wins. Full stop. Brand equity accumulated over decades does not transfer to a buyer that has no preferences, fatigue, or social graph. The payment rail problem — and why crypto won by accident. If agents are buyers, they need to pay. Card networks — Visa, Mastercard, Stripe’s card flows — were designed with a human session as a fundamental assumption. 3DS authentication, fraud signals, CVV entry: all of it presupposes a person at a keyboard. An agent has no session in that sense. It has a wallet key and an intent.
“Crypto stablecoin rails are becoming the path of least resistance for agent-native payments, not because of ideology, but because they do not require a human session.”
Companies like Skyfire and Coinbase are not winning agent payments because developers are ideologically committed to decentralization. They are winning because their infrastructure does not assume a human is present — and card networks do. The ideology did not win. The architecture did. Stablecoin transaction volumes reached $710B in early 2025, with 35M unique wallet addresses — up 50% year-over-year. Provenance becomes the new scarcity. In a world where AI can generate unlimited plausible-sounding content, volume is no longer scarce. What becomes scarce is verifiable origin — signal that can be traced back to a real decision, a real build, a real debugging session.
“Developer session recordings are a provenance mechanism. 999x growth is, in this framing, not a content tool — it is a provenance and signal extraction layer for the agent economy.”
A session recording is not a content artifact. It is a timestamped, authenticated chain of decisions made by a real developer building a real system. That is a fundamentally different class of asset than a blog post or a polished editorial take. When agents index the web to make purchasing, integration, or tooling decisions, they will weight for signal density and provenance — not production value.

​
What sticks

The convergence of these data points is a structural shift in who the customer is:
  • $67B in agent-influenced Cyber Week orders proves agents are already transacting at scale.
  • 46% of YC S25 is building agent infrastructure, which means the tooling to deploy more agents is accelerating.
  • Agents select on structured, machine-readable capability signals — not brand, not relationship.
  • Payment rails are adapting to remove the human-session assumption that blocked agent commerce.
  • Provenance — authenticated, traceable signal origin — is becoming the scarcity that content volume cannot replace.
For builders and founders in this environment, the practical question is not “should I think about B2A?” It is: does my current signal — my documentation, my content, my API surface — communicate clearly to a non-human evaluator? Session recordings converted into structured, provenance-tagged knowledge entries are not just useful for your human audience. They are the exact format an agent would weight most heavily when deciding whether your tool is trustworthy, capable, and worth integrating. You are not just building an audience. You are training the next generation of buyers to recognize your signal as authoritative. The B2A economy is not coming. It closed $67 billion in orders last November. The question now is whether your growth infrastructure is built for the buyer that showed up.