Call your agents. Or even better, code them—using super simple sentences like this one.
AI assistants that can handle work and everyday personal tasks, all operated by fast English-language commands that require zero coding knowledge, are increasingly defining phase two of the AI boom.
AI tools like Anthropic’s Cloud Code, Cursor, and OpenAI’s Codex can now write and debug software, opening up huge new sources of revenue. This success is pushing their creators toward a bigger ambition: to automate our entire lives.
“ChatGPT started out as something that could answer questions, but the long-term vision has always been to be a super-helper that can really help you get things done,” said Nick Turley, head of ChatGPT at OpenAI.
What started as a way to autocomplete code quickly evolved into semi-autonomous AI bots, or “agents,” that can work for hours with little human supervision. We can ask a bot to create presentations for work, coordinate family schedules, and pick March Madness brackets while it learns our personal preferences, no coding required.
“For me, coding is a kind of new literacy, but fortunately, learning to code now is much easier than learning to read, because you don’t have to practice, the tool just does it for you,” said Boris Cherny, who leads cloud code at Anthropic. “But I don’t want it to end. It’s going to be too disruptive.”
The shift has permanently changed the lives of coders and fueled a selloff in the $1 trillion market as investors and executives consider the technology’s potential to reshape industries including finance, legal and healthcare. AI has already been blamed for thousands of job cuts.
For OpenAI and Anthropic, winning the market for non-coders is the next frontier, especially as both companies are competing initial public offering That could come as soon as the end of this year.
“When you think about the future of knowledge work, it’s a multi-trillion dollar opportunity for companies,” said Dennis Dresser, chief revenue officer of OpenAI and former CEO of Slack. “It’s almost like if you can think it, or describe what you want, you can create it.”
In Silicon Valley, AI agents have already become a way of life.
Venture capitalist Tomasz Tunguz uses them to create charts, blog posts, and presentations. At one point, their spending on AI reached $100,000 per year. This went towards subscriptions to AI tools from Google, Anthropic and OpenAI, as well as fees for accessing their models directly through application programming interfaces or APIs.
“I’ll book travel, I’ll research vacations, I’ll read newspapers, all my email goes through it, my grocery shopping goes through it, music recommendations — I’m not reading magazines anymore — whatever questions I have, anything I want to know goes through the AI,” Tunguz said.
Other users talk about creating a dashboard to track their kid’s baseball stats, or automating the process of signing kids up for camps and daycare. Anthropic’s new features designed for non-technical tasks The target market for these devices is simply “anyone who needs to work on their computer,” said Felix Rieseberg, head of engineering at Cowork.
Rieseberg became a first-time father in January and has used Cowork to track and analyze a large folder of medical records. They’ve also used it to apply for mortgages and compile expense reports.
Theory Ventures founder Tunguz estimates that agents could generate $36 billion in annual consumer revenue in the near future, an amount that would represent a staggering growth rate from about a year ago. But he says this is just the beginning – the real money will come from lucrative enterprise contracts, a much bigger market opportunity than chatbots.
Anthropic and OpenAI charge around $200 monthly for the most expensive tier of their AI tools. In February, Anthropic said that Cloud Code was generating $2.5 billion in annual revenue. OpenAI declined to disclose Codex revenues.
Venture capitalist Tomasz Tunguz uses AI agents to create charts, blog posts, and presentations. At one point, their spending on AI reached $100,000 per year.
Like many power users, Tunguz uses tools to organize the work of multiple agents simultaneously.
“If I use a chatbot, I’m having a one-on-one conversation with the AI, but if I use something like Cowork, I can have 15 to 50 conversations with the AI simultaneously and that means they sell me a lot of inference,” Tunguz said, referring to the AI processing power required to run these systems. “That means more business for these companies.”
A lot of difficulties will need to be overcome to convince millions of new users to adopt agents concerns. People are losing sleep over the idea that using these tools means inadvertently training a model to replace them, or that they represent the beginning of widespread job losses and eventual economic collapse.
There are also concerns about the safety and security of these systems, for example Violence and self-harm People involving chatbots have become more common and stories are circulating about agents going rogue and deleting files. Letting AI complete the work for you often means giving it free rein over your data. And it can feel like taking care of a fleet of agents who are constantly making messes.
The coding war, which set the stage for phase two of the AI boom, has been going on for years behind the scenes, with multiple AI companies competing to win over software developers and influential early adopters.
One of the most important companies among them is Cursor, a small startup that played a big role in starting the coding war in 2023 by launching a tool that made it faster and easier to code. The cursor works with most AI models, meaning a user can tap Anthropic one day and OpenAI the next, or other models being produced around the world.
Insiders Cursors became a Silicon Valley darling by ushering in the “vibe-coding” era – basically allowing non-engineers to build software, apps, or websites without knowing how to write a line of C or Python. Nvidia CEO Jensen Huang has called it his “favorite” AI tool.
Last Value $29.3 billionCursor has grown to nearly 400 people and recently surpassed $2 billion in annual revenue, doubling in a three-month period. Despite recent
The reason there remains hope is that both OpenAI and Anthropic are subsidizing the cost of use on their platforms. They’re charging far less than the cost of running the AI, just as Uber and Lyft once did when they offered rides for just a few dollars in a race to drive each other out of the market.
Working with an AI agent, especially one that can work for hours non-stop, can be very expensive. Cursor charges through subscription tiers with usage limits. To lure customers, Anthropic and OpenAI are offering power users higher token limits like a $1000 price inside a $200 per month plan.
Recently, Cursor has expanded from a focus on coding more broadly to “helping developers create a factory that makes their own software,” said Michael Truell, the startup’s young, red-haired leader. Written on X.
A Cursor spokesman declined to comment.
Not long ago, Boris Cherny was living a very quiet life in Nara, Japan, where he worked remotely for Meta and loved writing code from Starbucks while watching the city’s famous deer wander outside.
Cloud Code began as his side project in late 2024 after joining Anthropic. “I started making products for myself,” Cherney said.
Cloud Code was released in early 2025 and developers praised it, claiming that it is more capable and reliable than cursors, and is capable of running autonomous agents rather than serving as a coding co-pilot that requires constant human input.
The tool went viral in November when, just before Thanksgiving, Anthropic released an updated model of the cloud that made the tool even more powerful. In December, it spread beyond software engineers and began growing rapidly.
Some users said they were “Cloud-pilledOr on “Cloud Benders”. They wore hats embroidered with the product’s crab mascot and stopped Cherney on the street for selfies or autographs. Suddenly, developers weren’t writing any code, they were writing Management of multiple agents simultaneously.
Demand for OpenAI’s codecs increased this winter when the company introduced its latest model, which has far better coding capabilities than its predecessors. Over the past 30 days, OpenAI has seen a “fundamental shift” in the way businesses use AI, said Dresser, OpenAI’s revenue chief, who called it a “wildfire moment.”
Codex has more than two million weekly active users, OpenAI said. The Journal said the company is finalizing a strategy to shift focus toward coding and business users, leaving behind projects that pull unnecessary focus. informed.
Traffic has increased eightfold in the past two months, requiring ever-more computing power to be brought online.
“Development is challenging,” said Thibault Sotiaux, head of Codex, who helped create the product, a day after an unexpected surge in users in mid-March caused technical problems for Codex. “We have to set up new data centers. We’re investing in making things more efficient and enhancing the underlying infrastructure.”
To keep top users happy, AI companies are treating them like VIPs, responding to their complaints faster on social media, or even treating them to alcohol.
When OpenAI was debating how to make its models’ personality more “fun”, Codex product lead Alexander Embiricos saw a power user arguing on X that the models should really be more concise and concise. Feedback helped the company decide to give users the option of personas, he said.
Anthropic co-founder Ben Mann and others at the company have organized events for power users. Over Chinese food in Seoul’s Gangnam district in December, coders shared ideas with Anthropic executives about how to make agents more trustworthy.
Sigrid Jin, an AI startup activist who attended the Seoul dinner, said she spent 25 billion Cloud Code tokens last year alone. At the time, usage limits were loose, allowing early enthusiasts access to tens of billions of tokens at very low costs.
Despite his countless hours with cloud code, Jin is not loyal to any AI lab. The tools available have different strengths and weaknesses, he said. Codex is better at reasoning, while cloud code produces cleaner, more shareable code.
Jin flew to San Francisco for Cloud Code’s first birthday party in February, where attendees waited in line to compare notes with Cherney. The crowd included a Belgian cardiologist who built an app to help patients get care, and a California lawyer who built a tool to automate building permit approvals using cloud code.
“It was basically like a shared party,” Jin said. “There were lawyers, doctors, dentists. They didn’t have software engineering backgrounds.”
Write to Kate Clark here kate.clark@wsj.com
