The rise of Business-to-Agent (B2A) companies

Plus: how to get an agent startup to $10 million revenue, billion-dollar companies with one employee, and more

Author

Michael Cunningham
February 03, 2025

In partnership with

Welcome back to Building AI Agents, your biweekly guide to everything new in the AI agent field!

For those of you expecting to become parents soon, congratulations! If you were were having trouble coming up with baby names, look no further.

In today’s issue…

  • Business-to-agent (B2A) joins B2B and B2C

  • OpenAI releases deep research agent

  • How to get an agent startup to $10 million revenue

  • Billion-dollar companies with one employee

…and more

🔍 SPOTLIGHT

Source: Created by BAIA staff

The tech world is saying its ABCs, with B2C and B2B now being joined by B2A.

While AI agents are seeing extensive use to automate companies’ internal operations, they also constitute an increasingly important part of the internet. Simple, hard-coded web scrapers and other bots are already a significant proportion of web traffic, but now more complex AI agents are beginning to make their presence felt. Agent builders are deploying systems that book vacations, shop for goods and services, perform research, aggregate news, and more. Consequently, for many merchants and websites, AI agents represent an important and growing segment of their customer base.

Companies’ business models are commonly divided into two main categories: business-to-business (B2B) and business-to-consumer (B2C). As their names suggest, in the former, a company’s customers are other businesses, while in the latter, they are individual consumers. The two are understood to each have their own unique advantages and challenges. Now, the rise of the agentic internet has led startup incubator Y Combinator to announce the birth a new business model: business-to-agent (B2A). In their most recent “request for startups”—a list of hypothetical new businesses the firm is interested in backing—YC expressed a desire to invest in companies which exist to provide services directly to AI agents, rather than to the humans behind them.

Although this may seem far-fetched at first glance, there is precedent for similar markets in which humans and computer bots transact seamlessly with each other; for instance, on stock exchanges, sophisticated algorithms already perform a majority of trading. To facilitate a similar ecosystem for AI agents, companies and researchers are already working on standardized communication protocols that agents can use to interact with websites, taming the wild jungle of graphical interfaces and non-standard APIs that constitute the current internet. Most notable of these is Anthropic’s Model Context Protocol (MCP), an attempt to create a common interface by which agents can access tools, databases, and other functions that they can use to interact with the outside world.

Others are developing payment platforms that allow agents to financially transact with humans. The multi-billion dollar payments company Stripe recently launched Stripe Agent Toolkit, which enables agents to pay humans directly, as well as businesses such as hotels and airlines. A host of smaller startups have sprung up to compete, such as Payman, Skyfire, and Protegee, some of which faciliate payments in cryptocurrency, while others stick to the more traditional financial system. Industry insiders have begun positing that this will lead to entire marketplaces driven by agents, which run the gamut from those where agents act as assistants to human shoppers to ones where humans are not present at all. Agents’ increasing ability to spend money has led at least one influential tech CEO to speculate that they could become the target of ads intended to persuade them to buy certain products, just as humans now are—in a fulfillment of YC’s vision of business-to-agent commerce.

The rise of this new economy will have profound implications for both agent builders and the coming wave of B2A startups. Regardless of which side of the ledger you are on, the future is clear: human beings are no longer the only ones who can sell and be sold to.

🗳️ POLL

Welcome to our first Building AI Agents poll, where we identify the most important trends in the industry in real-time by hearing directly from the people working in it.

Let us know what agent framework is the most important piece of your stack! Results of this first poll will be released this week’s Thursday issue.

Which is your favorite agent framework?

Login or Subscribe to participate in polls.

If you find Building AI Agents valuable, forward this email to a friend or colleague!

🤝 WITH THE AI REPORT

There’s a reason 400,000 professionals read this daily.

Join The AI Report, trusted by 400,000+ professionals at Google, Microsoft, and OpenAI. Get daily insights, tools, and strategies to master practical AI skills that drive results.

📰 NEWS

Source: OpenAI

Yesterday, the company announced the launch of deep research, a new capability for ChatGPT powered by o3 which leverages its thinking and web browsing abilities to deliver detailed research on a topic. This closely follows the release of o3-mini to ChatGPT users on Friday.

The Twitter founder’s latest project is Goose, an agent framework designed specifically to operate with Model Context Protocol, which was developed in collaboration with Dorsey’s company Block.

The tech and consulting giants announced a collaboration to help enterprises deploy AI agents, leveraging the former’s technological prowess and the latter’s implementation expertise.

🛠️ USEFUL STUFF

Source: Created by the author using Dall-E 3

Serial entrepreneur Greg Isenberg shares a roadmap for building a vertical agent startup and growing it to $10 million in annual recurring revenue (ARR).

A detailed tutorial video showing how to use one of the leading agent frameworks to build workflows and agents, and explaining the difference between the two.

This post walks through the essential layers of the modern AI agent stack, from model serving to observability, and the key software libraries that enable each of them.

The author of this piece explains some important principles for intelligently building teams of agents, informed by lessons from the functioning of human teams.

💡 ANALYSIS

Created by the author using Dall-E 3

This piece discusses the high likelihood that some startups powered by AI agents will soon begin to hit $1 billion valuations with only a single employee, and the potential implications for society.

A new version of the VC firm’s original thesis on the exploding field of voice agents, reflecting the enormous changes the field has gone through since the original report was released in May of last year.

The Meta CEO states that agents capable of replicating the functions of the company’s engineers will arrive in 2025, though the true effects may not be felt until the following year.

An interview with the founders of leading agent orchestration startup Agency (maker of AgentOps) in which they discuss some of the ways agents tend to fail, and how their tech is addressing them.

🧪 RESEARCH

Conseca architecture | Source: arXiv

Agents often determine whether a given action is safe using hard-coded rules, but this fails to account for the complexities of context. Contextual security for agents (Conseca) is a new framework designed to intelligently identify actions as safe or unsafe using LLM-powered reasoning.

Thanks for reading! Until next time, keep learning and building!

What did you think of today's issue?

Login or Subscribe to participate in polls.

If you have any specific feedback, just reply to this email—we’d love to hear from you

Follow us on X (Twitter), LinkedIn, and Instagram