AI & Tech Brief: Dispatch from AI+ Expo
Post AI+ Expo: enterprise AI adoption accelerates, standards debate intensifies amid vendor lawsuits, Washington, says report.
Image: GlobalBeat / 2026
AI news brief: Google and OpenAI unveil competing agents at AI+ Expo
SAN FRANCISCO — Google and OpenAI revealed rival AI agents Monday that can book travel, code websites and control web browsers without human oversight.
The announcements at the AI+ Expo mark the first public demonstration of AI systems that can chain together multiple online tasks, not just answer questions.
Both companies claim their agents represent a leap toward artificial general intelligence, though neither allowed independent testing. The demo sets up a direct clash between Alphabet and Microsoft-backed OpenAI for control of the next wave of AI tools.
Google CEO Sundar Pichai showed Project Mariner navigating airline sites to find round-trip flights from New York to Tokyo, then emailing the itinerary to a mock employee. The system completed the 12-step process in 34 seconds. Pichai told the crowd of 2,000 developers the agent “reasons across tabs, handles dropdowns and captchas, and knows when to pause for your credit card.”
Minutes later OpenAI president Greg Brockman premiered Operator, a chatbot upgrade that can open browser windows, fill forms and purchase items. Brockman typed “order a dozen bagels for tomorrow morning” and watched as the AI selected a local bakery, picked flavors and paid with a prepaid card. He said 10,000 wait-list users will get access this week.
The dueling demos triggered instant criticism from safety researchers. Anthropic co-founder Daniela Amodei, whose company is building similar technology, warned that agents “remove the human from sensitive loops like payments and medical bookings.” She called for mandatory audit trails before public release.
Consumer advocates raised louder alarms. John Davisson of the Electronic Privacy Information Center said agents that can spend money “shift liability to users when things go wrong.” He cited a Stanford study released Sunday showing AI agents hallucinate 18% of the time when given multi-step shopping tasks.
Google and OpenAI both responded that users must approve each purchase and can set spending caps. Pichai said Mariner logs every click and allows replay. Brockman promised Operator will “refuse to buy weapons, drugs or financial products.”
The agent race has already redrawn alliances inside Silicon Valley. Nvidia stock jumped 5% Monday on hopes the new tools will drive demand for more GPUs. Salesforce announced it will embed Operator into its customer-service software, while Uber said it is exploring Mariner for corporate travel booking.
Investors poured $400 million into agent startups last quarter, according to PitchBook. Several young firms set up booths outside the main hall hoping to be acquired. “Big Tech wants the pickaxes, not the miners,” said Yanda Erving, founder of 6-month-old startup ChainMind.
Regulators watched from the sidelines. FTC chair Rebecca Slaughter toured the expo floor but gave no speech. A staff member told reporters the commission is “gathering facts” on how agents handle personal data. The EU’s AI Act, which takes full effect in August, classifies autonomous systems that can cause financial harm as “high risk,” requiring audit logs and human oversight.
Neither company gave a public release date. Google said Mariner will stay in small beta “through 2026.” OpenAI listed Operator as a “research preview.” Behind the scenes employees admitted the technology still fails on roughly 1 in 20 complex tasks, usually when websites change layouts or require two-factor authentication.
Background
AI agents have been a dream since the 1990s when researchers at MIT tried to build shopping bots. Early efforts flopped because websites blocked automated traffic and natural-language understanding was too brittle. The field revived after large language models mastered code generation in 2022, letting agents interpret HTML and JavaScript like a human developer.
Google first teased Project Mariner in December when it published a paper showing an earlier version buying laptops on eBay. OpenAI followed in January with a technical report on WebGPT, a similar system that browsed the web to answer questions. Monday’s expo marks the first live public duel between the two approaches.
What’s Next
Both companies plan closed beta tests with corporate partners this summer. Google will charge $30 per user per month for Mariner when it reaches general availability, according to two employees. OpenAI has not set pricing for Operator but told investors it expects the tool to add $500 million in annual revenue within two years. Regulators in Brussels and Washington plan hearings on autonomous AI in June.
The close timing of the reveals signals that agents, not bigger chatbots, will dominate AI marketing for the rest of the year. Whether users trust software to spend their money remains the open question that could decide the race.
Technology & Science Editor
Sarah Mills is GlobalBeat’s technology and science editor, covering artificial intelligence, cybersecurity, public health, and climate research. Before joining GlobalBeat, she reported for technology desks across Europe and North America. She holds a degree in Computer Science and Journalism.