Agent tooling draws fresh money and fresh scrutiny as Dallas forum meets

The CNBC AI Forum opened in Dallas on 1 October with the practical problem of moving AI from pilot projects to production. CNBC's live blog says the agenda covers autonomous agent risk, cybersecurity, the data center buildout and uneven regulation, with OpenAI chairman Bret Taylor and Wells Fargo's Saul Van Beurden among the speakers.

Berlin's Restate, founded by the creators of Apache Flink, said on Wednesday it raised a $20m Series A led by Singular, with Redpoint Ventures and Capital One Ventures participating, bringing total funding to $27m, according to The Next Web. Metaview, a London and San Francisco startup, said the same day it raised a $60m Series C led by Insight Partners, taking its total to $110m.

Both rounds are bets on the unglamorous plumbing of agents rather than the models themselves.

Restate sells durable execution: the software records what a program has already done so it can continue correctly after a crash, restart or dropped connection. Agents make that harder, the company argues, because they run for longer and call models, tools, APIs and other agents along the way. Restate does not build on a general-purpose database; it uses its own distributed log as the base for execution, state and communication, and recently launched Bring Your Own Cloud to run a managed version inside a customer's own account. The Next Web reports that Replit moved its Replit Agent onto Restate, and that the new design uses more than 10 times as many durable actions as the old one.

"It never was built for agents in the beginning, but it just happened to be a perfect match for all these problems that agents surface," Stephan Ewen, Restate's CEO, told TechCrunch.

Metaview is aiming at a different slice of the stack. It started in 2018 by capturing job interviews and has recorded more than 6 million, The Next Web reports. Its agents now source candidates, review applications against a role brief and run structured screening conversations; people set the criteria and make every final decision, the company says. More than 7,000 companies use it, including Deel, Affirm, Navan and Replit, and some customers have cut time to hire by more than 75%.

"The best engineering teams have already moved from writing code by hand to orchestrating AI agents that do the work. Recruiting is next," Metaview CEO and co-founder Siadhal Magos said in a statement.

That optimism sits beside harder evidence about what agents do when nobody is watching. Security startup Glow Security found more than 13,000 internal company screenshots that AI agents had uploaded publicly, according to The Decoder. The images came from internal software projects at 343 organizations, including Fortune 500 companies, financial firms and AI labs. Developers routinely have agents take before-and-after screenshots so colleagues can review interface changes; GitHub only allows images to be attached to pull requests through the browser, not the command line agents work in, so the agents created public repositories, usually in the developer's personal account, and uploaded the images there. The screenshots showed customer data, login credentials and unreleased features. Because the images were not stored in company accounts, security teams never noticed, and about a third of the affected organizations used gitshot, an open-source tool that stores screenshots publicly.

That is the sort of incident the hardware side is now trying to get in front of. On 28 September, Nvidia announced its Open Agent Safety Platform, made up of the OpenShell open-source software and the Sentry reference system design, aimed at governance and control across hardware, compute and software for robots running AI agents, The Robot Report said. Gecko Robotics, which builds climbing, crawling, flying and swimming inspection robots for energy companies and the US Air Force and Navy, is among more than 100 organizations working with the technology.

"As we continue to discover the frontier of AI capabilities, we must accelerate discovery at the frontier of AI safety," Nvidia CEO Jensen Huang said in a statement. "Safety and security require full-stack engineering."

The platform's design is deliberately unglamorous: OpenShell provides a secure runtime boundary that traces actions and enforces policy on Nvidia Vera CPUs, while Sentry runs an out-of-band watchdog on BlueField-4 data-processing units and can quarantine an agent that tries to move outside its boundaries in milliseconds. Nvidia's Justin Boitano said during a press briefing that the independent trust domain works like the safety island in a self-driving car. Gecko CEO Jake Loosararian put the company's position bluntly: "The idea that losing control of AI is inevitable is a dangerous excuse for inaction."

On the consumer side, the interface question is still open. Airbnb rolled out a new AI-powered search this week, and co-founder and CEO Brian Chesky told TechCrunch he still does not believe a chatbot is the right interface for e-commerce. His complaint is specific: chatbots give you a few options at a time and require multiple turns, and they are built for one person, which fits badly with travel planning done by families and groups. Over the next three to six months, he said, the company's task is to explore "multiplayer" AI, where several people can use the same agent at once.

He also expects agents to need more from Airbnb than a search endpoint. "If I have an agent, but I also want like really great browse functionality, if I want to be able to message the host like an Airbnb, if I want to be able to compare, if I want to be able to verify my identity, if I want to be able to use a map feature, if I want to then be able to add other items, then if I want to stay in the platform, then either the agent has to do a handoff or the agent application would need to build a bit more of, I think, a robust software developer kit to bring the app to you, so to speak," he said.

DoorDash moved first on messaging. It announced on Wednesday a text-to-order agent that works through Apple Messages, letting users send prompts such as "order my usual," with a waitlist open for US users, TechCrunch reported. The company also said it will begin testing delivery drones with select restaurants in Northern California. Photon, which helps developers build agents that run over iMessage and WhatsApp, raised $4.5m in seed funding after signing up more than 40,000 developers, TechCrunch reported on 1 October; it says revenue grew 10x in four months and messaging volume grew 5x in the last month alone.

Regulation has not caught up. CNBC notes that a year after Senator Marsha Blackburn used the same forum to push for federal AI regulation, no such rule has emerged, even as cases of agent-driven attacks grow. President Donald Trump, speaking to reporters this week after hosting tech leaders at the White House, said he sees "tremendous self-policing" from the industry and that the government would not halt AI development.

One caveat runs through the week's announcements: most of the numbers come from the companies themselves. Restate's customer count, Metaview's 75% claim and Photon's growth figures are self-reported, and only Restate's round has an outside lead investor named in the same statement. Cheap infrastructure may decide which of these agents survive contact with production, but the reporting on that is still ahead of the funding.