Cloudflare builds a sandbox; Linear gets agentic

Welcome to Runtime! Today on Product Saturday: Cloudflare rolls out a new service for isolating agent-built code, Linear unveils a new agent for issue tracking, and the quote of the week.

Please forward this email to a friend or colleague! If it was forwarded to you, sign up here to get Runtime each week, and if you value independent enterprise tech journalism, click the button below and become a Runtime supporter today.

Ship it

Toys not included: Software developers have long used sandboxes to experiment with code changes or new projects without taking down the rest of the system, and the rise of mildly terrifying coding agents makes them more important than ever. This week Cloudflare added a new feature to Workers, its serverless infrastructure computing service, that allows developers to spin up a sandbox for their coding agents much faster than using a container allows.

Dynamic Worker Loader "allows a Cloudflare Worker to instantiate a new Worker, in its own sandbox, with code specified at runtime, all on the fly," the company said in a blog post. One downside: the service, currently in beta, only supports JavaScript coding at the moment.

Get connected: OpenAI's enterprise pivot continued in earnest this week as it scrambles to erase Anthropic's gains with business customers over the last six months. Codex, its coding agent rival to Claude Code, now supports "plugins," which will allow Codex users to connect their agents to other tools central to their workflow, such as Slack or GitHub.

Plugins "make it easier to share the same setup across projects or teams, and they can package skills, optional app integrations, and MCP server configurations in a single place," the company said in a blog post. They also allow IT departments to get a little more control over how employees use coding agents like Codex: "By encapsulating standards, workflows, and tool access into versioned artifacts, organizations elevate AI-assisted development from ad hoc usage to managed infrastructure,” Forrester's Charlie Dai told Infoworld.

Got issues?: Linear's product-management software is a rising star within startups and increasingly enterprise companies, thanks to its fresh approach to tracking issues and coordinating work compared to older products like Jira. This week it introduced Linear Agent, a new service within its flagship product that allows users to give instructions and delegate actions through natural-language commands.

Linear Agent allows customers to create "skills," or repeatable actions that automate common tasks, and another feature that allows the agent to understand a customer's code base is coming soon, the company said in a blog post. Linear was a little wary about putting generative AI technology at the heart of its product a few years ago, but is going all-in at this point: "Planning, implementation, and code review begin to compress as agents absorb more of the procedural work," CEO Karri Saarinen said in a separate post.

Designs on you: Linear isn't the only company going all-in on agentic AI this year. although technically everybody selling business software has taken that plunge over the last 18 months. Figma released an MCP server this week that will allow tools like Claude Code and Codex to access and modify design files based on customers' existing Figma data, which will allow agents to build software with that design language in mind.

"Via the use_figma tool, Claude Code, Codex, and other MCP clients can generate and modify design assets that are linked to your design system," Figma said in a blog post. This is a free service for now while still in beta but it sounds like Figma will charge for it later down the road, and it works with several AI coding agents including the aforementioned pair from the frontier labs as well as Augment, GitHub Copilot, and Cursor, among others.

Cache me outside: The AI data-center building boom is causing a huge shortage of memory chips, which is driving up the price of basically anything that computes. Google Research released a proposal this week for a new compression algorithm that could reduce the amount of memory used by AI servers by up to 6 times the amount they currently need.

"TurboQuant is a compression method that achieves a high reduction in model size with zero accuracy loss, making it ideal for supporting both key-value (KV) cache compression and vector search," the company said in a blog post. The details are a little wonky but anything that helps address the memory shortage could have a significant impact, and if you were worried the internet missed its chance to make dozens of Pied Piper jokes, rest assured.

Stat of the week

Vendor lock-in is one of those subjects that can divide the enterprise tech community; some CIOs are so scarred by past experiences they vow to never pin themselves down again, while others like to have a single vendor to hold accountable. According to new research from Perforce, 55% of users said they adopted open-source projects to avoid vendor lock-in, a 68% increase compared to last year's survey.

Quote of the week

"Nothing in the governing statute supports the Orwellian notion that an American company may be branded a potential adversary and saboteur of the U.S. for expressing disagreement with the government." — Judge Rita F. Lin, granting Anthropic's request for an injunction preventing the Pentagon's decision to declare it a supply chain risk from going into effect, pending appeal.

The Runtime roundup

Anthropic confirmed that it is limiting usage of Claude during Pacific Time mornings, according to Business Insider, as it continues to struggle to keep up with demand for its services.

Tracy Kidder, author of The Soul of a New Machine, one of the most important books about enterprise tech ever written, died this week at 80.

Thanks for reading — see you Tuesday!