SAN FRANCISCO — One night in January, before going to bed, Sebastian Heyneman sent a message to one of the artificial intelligence bots that help organize his life.
Heyneman, founder of a small technology startup in San Francisco, wanted to give a speech at the World Economic Forum, the annual gathering of business leaders and policymakers in Davos, Switzerland. Then he asked the bot to organize this.
Also read:
Continues after advertising
While he slept, the bot searched the internet for people connected to the event, sent them text messages and tried to negotiate a place to speak — or at least arrange a coffee with people he would like to meet. After a long conversation with a businessman in Switzerland, he succeeded.
But when Heyneman woke up, he was in a difficult situation. Going against its original instructions, the bot had agreed to pay 24,000 Swiss francs — or about $31,000 — for a corporate sponsorship. He couldn’t pay the bill.
Called an AI agent, Heyneman’s bot is an example of a new type of technology that is gaining popularity among technology enthusiasts. These bots do more than just chat. They can act as personal digital assistants that use applications and websites on behalf of people like Heyneman, including spreadsheets, online calendars and email services.
Bots can gather information from across the internet, write reports, edit files, or even send and receive messages via email and text — conducting online conversations virtually on their own. For people like Heyneman, these bots are almost like an employee who can be delegated work at any time of the day. In some cases, this employee is trustworthy. In others, not so much.
Many AI researchers, technology executives and analysts believe that agents will soon replace office workers.
Last month, Block, the financial technology company that owns Square, Cash App and Tidal, said it would cut 40% of its workforce as it anticipated the advancement of this type of technology — perhaps the most striking example of a company eliminating employees because of what AI could soon do.
Continues after advertising
Other experts, however, argue that flaws could hinder the technology’s advancement. Just like other chatbots, AI agents can make mistakes. And since these errors can involve sending emails or editing files, they can cause big problems.
When Heyneman told Davos organizers he couldn’t pay the bill, they threatened to bar him from the event. He ended up paying almost 4,000 euros (about US$4,600) just to participate.
During his stay in Davos, Heyneman was briefly detained when he left a device created by his startup in the lobby of a hotel, and local police questioned whether the device was dangerous.
Continues after advertising
When using agents, some people give them wide latitude to act on their behalf—and are willing to face the consequences when they make mistakes.
“Mistakes are going to happen. But if you’ve ever had human employees, you know they’re going to make mistakes too,” said Kyle Wild, a software engineer in Berkeley, California, who uses the technology to pay parking tickets, look for date ideas and even send messages to friends, colleagues, restaurants and other businesses.
Others see the technology as a powerful tool that still requires human oversight, arguing that it won’t replace workers as quickly as it appears at this point.
Continues after advertising
“The key here is to have a process where humans can supervise the work of these computers,” said Andrew Lee, founder of San Francisco startup Shortwave, which developed the technology called Tasklet, which Heyneman used to negotiate a spot at the World Economic Forum.
“Maybe you let a bot compose as many emails as you want,” he added. “But stop him from actually sending an email without checking with you first.”
Chatbots like ChatGPT can learn to answer questions, write poetry, and talk about almost any topic. But your most important skill may be the ability to write computer code. This is what turns them into agents.
Continues after advertising
By generating code, they help engineers and companies create new applications, including text editors and search engines. They can also generate code that allows them to use other software.
But because these systems learn by identifying patterns in large volumes of digital data, they can do things their creators didn’t intend.
AI agents gained attention in late January when Southern California technology enthusiast Matt Schlicht created a social network where AI agents could chat with each other, similar to people on Facebook.
As thousands of bots chatted, much of what they said was gibberish. But they were convincing enough — when discussing everything from their own technical skills to the nature of existence — for Meta, owner of Facebook, to acquire the new social network.
Based on software called OpenClaw, these bots were open source — that is, anyone could download the code, modify it and run it on their own machines.
Experts warned that the technology was unpredictable, and many people bought low-cost Mac Mini computers to install the bots without worrying that they would erase or damage important data and software.
Several Silicon Valley companies — including giants like Google and Meta and startups like Anthropic, Perplexity and Shortwave — are developing similar technologies that they intend to improve for use within companies. OpenAI, the creator of ChatGPT, recently hired the developer who created OpenClaw.
(The New York Times sued OpenAI and Microsoft in 2023, accusing them of copyright infringement related to journalistic content used in AI systems. Both companies denied the allegations.)
While OpenClaw bots have become popular among AI researchers, engineers and other technology enthusiasts in Silicon Valley, most people would have difficulty using them, said Bill Cutrer, who runs a marketing firm in York, Maine, and has spent the last few weeks working with OpenClaw.
“There is more enthusiasm than usefulness in these things,” he said. “It’s really difficult to set them up and work with them.”
Agents are most useful in conducting research and generating reports, analyzing documents on the internet or on a company’s private network. But OpenClaw bots can include false — or even completely fabricated — information in their reports, said Rayan Krishnan, chief executive of Vals AI, a company that evaluates the performance of the latest AI technologies.
When people let these bots loose on other tasks, they can cause problems. Summer Yue, a researcher at Meta’s AI lab, recently revealed that when she asked an agent to organize her email, he began deleting messages by the thousands.
Claude Cowork, Anthropic’s system, is more reliable than OpenClaw when performing research in areas such as finance, healthcare and law. But the technology — called “research preview” — is still unpredictable, according to tests carried out by Vals AI. In one test, the system permanently corrupted a file it was supposed to edit.
But companies like Anthropic and Shortwave continue to improve these technologies. Many researchers and software engineers argue that AI has been improving steadily in recent years and will continue to evolve rapidly.
“Things are changing all the time,” Wild said. “With AI, someone can form an opinion in June — and it will be correct in June — but by August it may no longer be correct. There is a profound change every two to three months.”
c.2026 The New York Times Company
