$1 billion companies with one employee?

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

Sure, it makes you look cool…but only until someone else (literally) one-ups you with “multi-agentic”.

📢 This week, we’re proud to partner with Sophtron, whose agentic AI system enables data access to over 15,000 financial institutions, allowing you to build fintech applications and AI agents that manage finances. Check them out!

In today’s issue…

• Agents enable huge valuations with tiny workforces

• Anthropic engineers discuss the best way to build agents

• The rise of agent experience

• AI agents in 2025: Expectations vs. reality

…and more

🔍 SPOTLIGHT

In Silicon Valley lingo, a “unicorn” is a startup worth at least a billion dollars—said to be as rare as a unicorn. Soon, the unicorn’s single horn may symbolize something new: the startup’s lone employee.

The rise of the internet has massively expanded the leverage individuals can exert, as increasingly sophisticated software—now augmented by AI—allows them to build complex products and virally market them to the whole world through social media. Tech founders have responded to this new world by prioritizing tiny, “cracked” teams of employees with generalist talents who can hyperscale on a shoestring budget.

Consequently, per-employee valuations of the most successful startups have skyrocketed. Messaging service WhatsApp, with a workforce of 55, was bought by Facebook in 2014 for $19.3 billion dollars—$351 million per employee. When Facebook acquired Instagram for around $1 billion, it had just 13 employees.

Now, the power of AI agents is leading some—including OpenAI CEO Sam Altman—to speculate as to when the first billion dollar company with a single employee will launch. Such a company, though seemingly far-fetched, isn’t impossible to imagine. One incredibly hardworking founder, using AI agents to help create their product and market it on social media, could well pull it off.

At least one Y Combinator-backed startup with a single employee is attempting a similar play. Rocketable, a holding company founded by—and entirely consisting of— designer and engineer Alan Wells, aims to buy up existing companies and replace their teams completely with AI agents.

This business model faces long odds, however, especially as its companies scale. While some functions of an enterprise—human resources, of course—are unnecessary with an all-AI team, others, such as legal, sales, and marketing, will continue to be essential, and automating them with agentic AI to the point that a single person can reasonably do all of them is still incredibly challenging, even with rapidly advancing agent capabilities.

In the short run, a more likely model for a massively scaling agent business is one that identifies a vertical that requires large amounts of human cognitive labor for a single bottleneck, intensely automates that step using agents, and provides that automation as a service to businesses that struggle with it. These vertical agent startups have sprung up across a wide range of industries, such as Harvey for law (worth $3 billion), Sierra for customer service ($4.5 billion), and more.

Thus, while a handful of lucky founders may soon find themselves able to scale to unicorn status with a viral product without human help, companies with a billion dollars of valuation per employee—but multiple employees—will be far more common.

For now, at least, we still need each other.

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

🤝 WITH SOPHTRON

Check out Sophtron’s tech here

📰 NEWS

The new Alexa+ is an AI agent

Amazon announced the new generation of its Alexa assistant, which, unlike its clunky hard-coded predecessor, fully embraces agentic AI—as this great piece details.

The best agents from Berkeley’s MOOC hackathon

Berkeley’s massive online agent course announced the winners of its hackathon, drawn from over 3,000 participants from 1,100 universities and 800 companies.

🛠️ USEFUL STUFF

Anthropic’s tips for building AI agents

In this video, the Anthropic team riffs on their “Building effective agents” guide, discussing best practices for agent building and the future of the field, including why agents are seeing more adoption for business than consumer applications.

Rapid communication between agents over the phone

Gibberlink, the source of the viral demo that was featured at the beginning of last Thursday’s issue, allows AI agents to quickly communicate information by voice in a way similar to dial-up internet, with significant implications for over-the-phone communication.

Build apps with Windsurf’s coding agent

A course by Andrew Ng’s DeepLearning.ai showing how to use the powerful new agent-powered code editor Windsurf.

💡 ANALYSIS

The rise of agent experience

A podcast by Weaviate which discusses agent experience (AX), an increasingly important analog of user experience (UX) focused on making software easily accessible to AI agents.

AI agents in 2025: Expectations vs. reality

This article by IBM features interviews with four of the company’s thought leaders on agents, who temper the hype on the field while still predicting some of its transformative impacts.

Google's co-scientist is “test-time scaling on steroids”

The company’s science agent announced last month is a demonstration of the power of test-time scaling: the use of long reasoning chains, rather than bigger models, to improve performance.

🧪 RESEARCH

A survey of reasoning LLMs

A deep dive into the new wave of reasoning LLMs, such as OpenAI o1/o3 and DeepSeek R1.

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

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