First off, check out Jyoti’s Wednesday scoop about Meta’s latest AI model efforts. In sum, Meta is telling its employees “we’re back in the game!” and setting high expectations for the company’s next open-weight large language model, codenamed Avocado.

Yes, Meta AI leaders did the same last year with Llama 4 and we all know how poorly that turned out! But given what happened then, we imagine that this time will be different.

On to today’s column…

As viral tools like OpenClaw bring AI agents back into the tech-industry mainstream, startups are increasingly developing specialized software for the agents themselves to use to actually do the job of human coders and other white-collar workers.

The latest such startup of note is Daytona. It sells personal virtual computers, known as sandboxes, for agents to run and test code; analyze data; or use enterprise apps like Salesforce or Excel. The sandboxed environments ensure that the agents won’t accidentally damage the computer of their human masters or disturb their company’s active codebase.

The two-year-old Daytona, which is based in New York City, has raised a $24 million Series A round led by FirstMark Capital, CEO Ivan Burazin told The Information exclusively.

Enterprise software firms Datadog and Figma also participated in the funding round, he said. Such firms are intently watching the agent field and are surely wondering what the future looks like when bots rather than humans access their services. Uncertainty about that future is driving the SaaSpocalypse selloff in software stocks, including Datadog and Figma.