In June, data labeling startup micro1 approached Robby Hogle, CEO of a heating, ventilation and air conditioning company in Troy, Mich., with a proposition: why not learn how to use AI in his business and get paid for it?

Hogle had already been doing some work for micro1 as a side hustle. Hogle, a former investment banker, had been providing feedback to micro1 on how well AI models were able to complete financial modeling tasks. 

Now, micro1 wanted Hogle to test how well different AI models did on tasks involving his day job, such as processing vendor invoices, filling out credit applications or doing payroll for the HVAC firm’s 14 employees. Hogle agreed. He now comes up with tedious tasks he wants to automate and ranks how different AI models handle those tasks based on factors such as speed, efficiency and accuracy. He provides those evaluations to micro1, he said. 

Hogle doesn’t know whose models he is evaluating; they are anonymized, though micro1 likely works with the frontier AI labs, as do other data labeling startups.

Micro1 is doing such work with many other small business owners, in fields such as marketing and professional services, hoping to solve a nagging problem in the AI field: Even as AI models improve rapidly in fields like mathematics or physics, many business customers still complain that they fall short on even relatively simple tasks such as ordering inventory or onboarding new employees.  

To solve that problem, AI developers have sought new types of data to train their models, such as reinforcement learning environments, which simulate common workplace applications like Salesforce or Excel to make models better at tasks involving those apps. Micro1 and other data labeling firms are now gathering data from small businesses like Hogle’s to further expand model capabilities.

Micro1 pays him $1,000 for every labeling task he does, and Hogle said he hopes to complete between 150 and 200 tasks this quarter. That means that the firm could make more than $150,000 this quarter from data labeling alone, he said. 

For Hogle, it’s a no-brainer. The work he does for micro1 to automate back-office processes with AI is work he should be learning to do anyway, he said. And it only adds an extra three to four hours a week to the time he’d already be spending doing tasks like payroll or invoice processing, he said.

Hogle’s HVAC firm is less than a year old and hasn’t yet hired a back-office employee whose job would theoretically be automated as a result of his data labeling work, he said. Many other small, local businesses might have a harder time wrestling with such dilemmas, he said, even if their core work is safe from automation.

He also knows the extra income is likely fleeting, lasting “only last three, six or nine months… it’s a very temporary moment in time for us to be able to do this,” he said. 

In other news…

AI Employees Call for Slowdown (Again)

More than 1,100 senior executives and other employees at leading AI companies signed a letter asking the U.S. government to “support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development.”

If that sounds a little familiar, it might be because Tuesday’s letter isn’t the first one to call for slowing the pace of AI development! In 2023, Elon Musk, Steve Wozniak and other tech figures called for a 6 month pause on training AI systems more powerful than OpenAI’s GPT-4. 

Of course, AI companies kept right on at it. (They probably scraped that letter off the web and used it as training data!) But a lot has changed in the past three years, and Tuesday’s letter has more of a shot of steering the ongoing policy conversation about global governance of powerful AI models.

Calls for coordination on pausing or slowing AI development have been growing in recent months. Anthropic raised the idea last month in response to the possibility that AI could rapidly accelerate the pace of AI development, otherwise known as recursive self-improvement. Such a scenario also motivated Tuesday’s letter: “The world's leading AI companies believe they could be close to automating AI research,” it says. “There is a real risk that capability development rapidly accelerates beyond our ability to understand or control the resulting systems.”

Google DeepMind CEO Demis Hassabis said at Davos earlier this year that he would be open to a détente in the AI race if every AI company agreed to it, though his name does not appear on the letter.

AI employees have signed many statements over the past three years—declaring AI an existential risk, calling for whistleblower protections, endorsing AI safety laws, calling to reinstate an ousted CEO, opposing military deals and defending Anthropic against the Department of Defense. 

One thing is for sure: for an AI system to fully automate an AI researcher’s job, it will have to be able to sign open letters.—Rocket Drew

Here’s what else is going on…

Big Number: 1 Billion

OpenAI’s ChatGPT is on the precipice of having 1 billion weekly active users, a milestone the company was hoping to hit by the end of last year, The Information reported on Tuesday. Though the milestone comes seven months after the company originally expected to hit it, ChatGPT is still one of the fastest, if not the fastest, internet applications to reach that many users less than four years after launching.

Policy Watch

The U.S. government on Tuesday announced a ban on the imports of new foreign-made humanoid robots, which are mostly from China, citing national security risks.

The move reflects Washington’s growing concerns about China’s critical role in the global robotics industry and its supply chain. China currently accounts for the vast majority of the world’s humanoid production.

People on the Move

Robert Hundt, a Google distinguished engineer who was the original software leader for its Tensor Processing Unit chips, has joined Amazon’s chip team under that same title, according to an Amazon spokesperson.

Earnings Report

Data backup and recovery software firm Commvault’s stock slid 16% in the hours following its announcement that its revenue growth slowed to 11% year-over-year, representing $314 million in sales, in the July quarter compared to a year-over-year rate of 13% in the previous quarter.

Deals and Debuts

See The Information’s Generative AI Database for an exclusive list of private companies and their investors.

Cybersecurity company Cyera agreed to acquire Oasis Security for approximately $1 billion, paid mostly in cash with the remainder in Cyera shares. 

Spur Intelligence, a bot-detection startup, raised $200 million in funding led by Insight Partners.

Dwelly, a startup whose AI software runs the day-to-day of renting out homes, raised $170 million in a Series B funding round, made up of $95 million in equity plus a $75 million debt facility from Trinity Capital. 

Recursive Superintelligence, a neolab founded by Richard Socher, signed a multi-year, $410 million agreement with Amazon Web Services to run its automated AI-research system and co-develop purpose-built infrastructure. 

Coursera made a $100 million strategic equity investment in LearnVector, a new AI-native learning company founded in 2026 by Andrew Ng.

Freehand, a San Francisco-based startup that offers AI agents to manage supply-chain spend for companies, raised $75 million in seed funding led by Battery Ventures.

Fish Audio, which develops AI voice models for voice cloning and text-to-speech, raised $52 million in a seed funding round led by Coreline Ventures and Capital Today. 

Act Security, a startup whose cloud security software shrinks the access surface across a company's cloud, raised $60 million in total funding, including a $20 million seed funding round led by Team8 and Bessemer Venture Partners. 

Mate Security, an AI cybersecurity startup, raised $35 million in a Series A funding round led by Canaan Partners. 

Harmony, a New York City-based startup that offers enterprise AI agents, raised $34 million in funding led by Lightspeed Venture Partners.

COR, which aims to help people and AI agents work better together, received a $30 million investment from FTV Capital. 

Hush Security, a startup that aims to secure a company's non-human workforce, raised $30 million in a Series A funding round backed by Akamai Technologies, Battery Ventures and YL Ventures. 

Greyparrot, a startup that offers cameras to sit above the sorting belts at recycling facilities and identify each discarded item's material, product and brand in real time, raised $27 million in a Series B funding round led by Omar Mir.

Epitel, a company whose AI-driven system monitors brain activity remotely, raised $26 million in a Series B funding round led by Catalyst Health Ventures and Genoa Ventures. 

telli, a startup whose AI-native platform automates customer conversations for consumer brands, raised $15 million in a seed funding round led by redalpine. 

Weave, a San Francisco-based startup that tracks engineering work and measures how much of it is done by AI versus human developers, raised $13.5 million in Series A funding led by Standard Capital.

Credible Data, which provides truthful data to better ground AI agents, raised $10 million in a seed funding round backed by Gradient, SignalFire, K5 Global, Godard Abel, Wes McKinney, Alex Dean and SV Angel. 

Pangram, which detects AI-generated content, raised $9 million led by Menlo Ventures.

Antares Labs, a startup that turns a real-estate firm's own institutional knowledge, proprietary data and workflows into custom AI systems, raised $7.25 million in a seed funding round backed by Fifth Wall, Base10 Partners, Bloomberg Beta and Sandwith Ventures. 

Fincart, a startup whose AI-powered operating system helps online merchants run their businesses, raised $2.8 million in a seed funding round led by Launch Africa and Antler MENAP. 

AI legal startup Legora on Wednesday announced that it has acquired Wexler, a London-based startup that uses AI to search for facts across large documents. Legora said that Wexler‘s team will become the founding team of Legora’s new London engineering hub.

Conversica, an enterprise AI software company that builds AI agents for customer conversations, announced a majority recapitalization led by a fund managed by Morgan Stanley Expansion Capital, which will take a majority stake. The deal size was not disclosed.

GrubMarket, an AI-powered food supply-chain company that connects wholesalers and distributors to grocers and other food buyers, confidentially filed for a U.S. IPO. It was valued at $4.5 billion in February. 

Ionic Digital, a bitcoin miner and AI-infrastructure company formed from Celsius Mining's former assets, surged more than 25% to nearly $63 in its Nasdaq debut, giving it an implied valuation of about $2.75 billion.

The Agentic AI Foundation shipped a major update to the Model Context Protocol, the open standard that connects AI agents to outside software, moving it to a fully stateless architecture with a hardened authentication model and a formal 12-month deprecation policy for future changes.

Core Scientific and AMD announced a deal under which AMD will secure more than 500 megawatts of U.S. data-center capacity from Core Scientific starting in 2027, with the ability to scale up to 2.5 gigawatts.

New From Our Reporters

Exclusive

OpenRouter Financials Suggest Steep Price For Possible Acquirer Stripe

By Amir Efrati and Stephanie Palazzolo

Exclusive

OpenAI’s ChatGPT Nears 1 Billion Weekly Active Users Seven Months After Target

By Stephanie Palazzolo and Erin Woo

Artificial Intelligence

Anthropic’s Claude Code Reigns Despite Rising Interest in Codex, Open-Source Models

By Laura Bratton

Thank you for reading the AI Agenda Newsletter! I’d love your feedback, ideas and tips: stephanie@theinformation.com.

If you think someone else might enjoy this newsletter, please pass it forward or they can sign up here.