Welcome to�AI Decoded,�Fast Companys weekly newsletter that breaks down the most important news in the world of AI. You can sign up to receive this newsletter every week�here.
Coming soon: The one-person, billion-dollar startup
Were beginning to see a new kind of lean startup company, enabled in large part by new AI agents. Since these companies rely far less on people power, some of themthe ones addressing real market needsare achieving extraordinary revenue-per-employee numbers.
AI coding tools could become major enablers of these lightly staffed AI startups, simply because theyre starting to automate software development tasks that once required a human designer or engineer. The makers of these coding tools provide early examples of such startups.
We are seeing the rise of the AI-first company, where tech companies use AI and agents to complete tasks before hiring employees, says Jeremiah Owyang, general partner at Blitzscaling Ventures.
A few notable examples:
Anysphere, the company behind the Cursor AI coding tool, has only about 20 employees. By the end of 2024, its annual recurring revenue (ARR) had reached $100 millionroughly $5 million per employee. Now, TechCrunch reports that its ARR has surged to $300 million in 2025, bringing its per-employee revenue to $15 million. (Anysphere was named one of Fast Companys Most Innovative Companies in the Applied AI category.) The company has reportedly turned down several buyout offers, including one from OpenAI, to remain independent.
Another AI coding tool maker, Windsurf (formerly known as Codeium) has reached $40 million in ARR, up from $12 million at the end of 2024. With around 170 employees, this implies a revenue-per-employee figure of approximately $235,000. OpenAI is reportedly in advanced talks to acquire Windsurf for around $3 billion, or about $17 million per employee.
A South Korean startup called Nari Labs, with just two employees, recently unveiled a new text-to-speech model called Dia that may outperform category leaders ElevenLabs and Sesame in terms of voice authenticity. Voice samples posted to X by the company appear to support this claim. Nari Labs has shared its model on GitHub and could potentially build a business around licensing future models. The company did not immediately respond to a request for comment on Tuesday.
Midjourney, the creator of the well-known text-to-image AI tool, has kept a lean profile, with only about 10 employees. PitchBook estimates that by the end of 2024, the company was generating approximately $200 million in annual recurring revenuearound $20 million per employee.
A lesser-known AI company, Nexad, which develops and deploys AI-native advertising within AI applications, has just six employees but has already reached 30 million users through its chat app partnerships. Founded in 2024, the startup has secured $6 million in seed funding in a round co-led by Andreessen Horowitzs Speedrun accelerator and Prosus Ventures.
This trend will only continue as companies realize they can gain efficiencies by using software for repeatable tasks, says Blitzscaling Ventures Owyang, while reserving human talent for strategy, innovation, creativity, leadership, and community.
These AI-first companies are becoming more viable as AI models improve. The generative AI boom began with models that could string words together in useful ways, but only recently have models gained the ability to reason independently and work through processes with a degree of autonomy and agency.
The place where this ability is having the greatest impact today is in AI coding assistants, but many expect AI agents to take over other tasks, like invoicing and customer support, that were formerly the sole province of humans. In the future, when Company A wants to buy from Company B, it could be a matter of two AI agents working together to open the business relationship.
OpenAI CEO Sam Altman speculated in early 2024 that because of AI agents, a billion-dollar company employing one individual might be created. In fact, Altman said he has a running bet with some of his peers on when such a unicorn might appear. That day might be coming sooner than we think.
It might even go further than Altman envisions, Owyang says: In a future that once seemed like science fictionbut may be just a few years awaywe could see companies comprised entirely of AI agents, with no clear indication of whether any humans are at the helm.
Why OpenAI buying Chrome could face antitrust headwinds
A federal court ruled last August that Google holds a monopoly on internet search. This week, the court is working to determine a list of remedies to address that antitrust issue. Federal antitrust officials are urging Judge Amit Mehta to order Google to sell its Chrome browser, which acts as a major funnel of user traffic to Google Search. Google earns the majority of its revenue by selling ads around search results and referring users to brands when they search for products.
Depending on the buyer, selling Chrome might fix one monopoly only to create another a few years down the line. OpenAIs head of ChatGPT, Nick Turley, testified that his company would be willing to purchase the popular browser, which could fetch as much as $20 billion, according to Bloomberg. That would place the browser in the hands of a company currently building its own internet search business. OpenAI licenses Bing search data and is developing its own search index. The company reportedly tried to license Googles search datathe most complete inventory of the webs contentsbut was denied. The Justice Department has also proposed requiring Google to license its search index to other search competitors.
“To grow further, OpenAI needs to move beyond supplying models and start owning the customer connection,” says Info-Tech principal research director Brian Jackson. “Gaining control over a major browser like Chrome would expand its reach and create new data opportunities, while helping it better compete with Google and its Gemini platform.”
OpenAI is among the first companies to offer an alternative to the classic Google search thats become almost reflexive for many webusers. Instead of returning a list of most relevant links, ChatGPT, Perplexity and other chatbots return a direct answer to a users question, in narrative form (Google has its own AI search format called AI Overviews). Instead of doing searches from Chrome (by either using the URL bar or right-clicking on search terms) Ive found that using a chatbot desktop app, which typically display a handy little prompt window triggered by a keyboard shortcut, is just as easy and often yields more useful results, depending on the type of search. A ChatGPT-optimized Chrome could provide another easy entry pointone that far more people would likely use. In that experience you might highlight something in the browser, right-click, and see a search with ChatGPT menu item where search with Google used to be.
Of course, that would create a major advantage for OpenAI in search, while other AI search providers such as Google, Perplexity, and Anthropic could be put at a disadvantage. If AI search continues to grow in popularity, theres a real chance that it becomes the dominant way of searching the web at some point in the future. Does that make OpenAI the next search giant?
Meta rolls out new AI features in its Ray-Ban AR glasses
Im excited about Metas Ray-Ban smart glasses, both because of their stylish design (theyre not bulky or awkward), and their potential to integrate useful AI features. Meta has chosen a great feature to lead with: On Wednesday, the company began rolling out its live translation capability to smart glasses users in all markets. (The feature was first teased back in October.)
In effect, people will be able to travel abroad and hold reasonably smoothif occasionally clunkyconversations with speakers of different languages. When the microphones on the glasses hear a different language, the words are sent to an AI server in the cloud, which then sends the translated words back down and through the glasses earphones. And, in a nod to on-device AI, Meta allows users to download language packs directly to the glasses, enabling offline translation without a network connection. For now, live translation supports English, French, Italian, and Spanish, with more languages on the way.
Meta also seemed to move up the timeline for another AI featureLive AI, in which the AI continually watches the live view from the devices cameras in order to assist the user in things they may be doing. If it sees food preparation, the Meta assistant might offer recipes from the web or substitutes for missing ingredients, or if the user is exploring a new neighborhood it might supply mapping or navigation features. Meta now says Live AI is coming soon to general availability in the U.S. and Canada. Live AI gives a feel for how Meta has hoped AI would enhance the user experience in its smart glasses.
More AI coverage from Fast Company:
Microsoft thinks AI colleagues are coming soon
Trump is reportedly drafting an executive order to integrate AI into public schools
Broadcom is betting big on ethernet to disrupt AI workloads and data centers
This startup wants to reprogram the mind of AIand just got $50 million to do it
Want exclusive reporting and trend analysis on technology, business innovation, future of work, and design? Sign up for Fast Company Premium.
