GitHub calls HQ; Coding upstarts drop new models

Welcome to Runtime! Today on Product Saturday: GitHub's bid to become the center of agentic coding, Cursor and Cognition roll out new coding-specific AI models, and the quote of the week.

(Please forward this email to a friend or colleague! And if it was forwarded to you, sign up here to get Runtime each week.)

Ship it

Justice League: GitHub hosted its first Universe event this week since Microsoft folded the company into its CoreAI Group, and showed off plans to help enterprises manage their use of major coding agents all in one spot. Agent HQ "[makes] agents native to the GitHub flow," the company said in a blog post, introducing new tools to manage code review and monitor usage.

But the most interesting part of Agent HQ is that GitHub will make AI coding agents from several major AI companies — Anthropic, OpenAI, Google, Cognition, and xAI (but not Cursor, apparently) — available directly within GitHub for paid subscribers. "The future is about giving you the power to orchestrate a fleet of specialized agents to perform complex tasks in parallel, not juggling a patchwork of disconnected tools or relying on a single agent," it said.

Security first?: The wave of AI-generated cyberattacks that many feared with the onset of generative AI has yet to emerge, but like just about everybody else in enterprise software, defenders are jamming the technology into every aspect of their products. This week Palo Alto Networks introduced Cortex AgentiX, the next generation of its security orchestration, automation, and response platform.

AgentiX comes with several security agents built by Palo Alto as well as a no-code tool for building custom agents. Security teams have to wade through a pile of alerts when trying to figure out exactly what has gone wrong, and whoever makes a tool that can accurately automate a lot of that work will draw interest.

Up to code: This week Cursor rolled out the second version of its flagship AI coding tool, which has quickly become a hot item in software development circles. Just like GitHub's Agent HQ, Cursor 2.0 "makes it easy to run many agents in parallel without them interfering with one another, powered by git worktrees or remote machines," it said, and the startup also released Composer, its first AI model built specifically for coding tasks.

Cognition, which snapped up the part of Windsurf that Google Cloud declined to bring aboard earlier this year, also released a new AI coding model this week. "SWE-1.5 reimagines the entire stack — model, inference, and agent harness — as a unified system optimized for both speed and intelligence," the company said, and it is available in Windsurf.

Kind of blue: Nvidia's GPUs are to thank for its mind-boggling $5 trillion valuation, but its data-processing units, or DPUs, have become an important part of next-generation data-center design by taking a lot of the network processing load off the CPU's to-do list. This week it introduced BlueField-4, which will be available next year as part of its Vera Rubin platform.

The new DPU is actually a combination of Nvidia's Grace CPU, which is based on Arm's architecture, and its ConnectX-9 network interface controller. "This means we’re unlocking peak efficiency for AI workloads, from training trillion-parameter models to running massive inference jobs," Nvidia's Dion Harris told CRN.

Alternative rocks: Here's an interesting thread to follow over the coming years: How long will hyperscalers and enterprises put up with Nvidia's domination of the AI market? Right now, they don't have much choice, but Qualcomm took aim at Nvidia this week with the introduction of the AI200 and AI250, designed for AI inference.

The new chips "offer rack-scale performance and superior memory capacity for fast AI inference at high performance per dollar per watt," Qualcomm said in a press release, hoping to tap into the sticker shock that's just part of signing a purchase order for Nvidia's GPUs. They won't be available until next year, however, when Nvidia will also be coming out with a new generation of its flagship GPUs.

Stat of the week

How ready are companies to embrace agentic security products like the one Palo Alto Networks unveiled earlier this week? According to new research from Vanta, 61% of business and IT leaders would feel comfortable allowing an agent to override a human "in some cases," but only 48% acknowledged they "have developed a framework for granting or limiting autonomy in AI systems," which leaves a sizable — and troubling — gap.

Quote of the week

"It is what it is. People are ready to live with it today because, one, they have no choice, and second, because of the productivity benefit." — Zencoder co-founder and CEO Andrew Filev, discussing the uphill climb that AI providers have to match cloud infrastructure computing reliability, even after recent outages.

The Runtime roundup

Microsoft's earnings results showed that OpenAI lost about $11.5 billion last quarter, according to The Register, which even in today's climate is a lot of money to set on fire.

Assuming it can come up with the money, OpenAI announced plans with Oracle to build a 1.4-gigawatt data-center complex in Michigan outside Ann Arbor that will supposedly start construction next year.

Thanks for reading — see you Tuesday!