Google promotes k8s for AI; IBM says use a mainframe

Welcome to Runtime! Today on Product Saturday: Google Cloud outlines a new way for Kubernetes users to run inference on their existing clusters, why IBM thinks its new mainframe is an AI engine, and the quote of the week.

(Was this email forwarded to you? Sign up here to get Runtime each week.)

Ship it

Steering the ship: Kubernetes is used inside most enterprise companies that need to run distributed computing clusters these days, and Google Cloud wants to make sure that it can handle AI workloads, too. This week at Cloud Next '25 the company introduced two new features in its GKE managed-Kubernetes service designed to manage inference workloads.

GKE Inference Quickstart helps Kubernetes users deploy AI models more easily, and GKE Inference Gateway helps manage the unique traffic patterns seen across AI applications. "The traditional round-robin load balancing that most teams are accustomed to doesn't actually work with LLM based applications, because LLM applications have really variable response times," Google's Gabe Monroy told Runtime in an interview at Cloud Next.

Coding canvas: Canva is a household name in the world of graphic design, but this week it introduced a new service that will allow users to create their own apps using a no-code builder. Canva Code was built around a large-language model and "a text prompt allows you to create dynamic, custom tools for websites, presentations, docs, social posts, and more, instantly," the company said in a blog post.

Coding has emerged as one of the best applications for LLMs, giving professional developers tools to speed up their workflows and finally delivering on the promise of low-code and no-code tools that allow novices to get up and running. "It’ll be their entrée into coding, because all of a sudden you can see code that’s generated and go in and edit it," Canva CEO Melanie Perkins told Fortune.

New mAInframes?: It's always a little surprising to see how many companies still use mainframes at the heart of their digital infrastructure, although it's next to impossible to imagine anyone using them to build infrastructure from scratch. Still, companies that are heavily invested in mainframe architectures need to refresh their hardware every so often, and this week IBM introduced a mainframe that it said was designed to get those companies into the AI era.

"IBM z17 is a system designed from the ground up to fully integrate into hybrid environments by tightly joining hardware innovations, software capabilities for AI, and rich support for open-standards and tooling," the company said in a press release. Building the new model was a five-year process, IBM's Tina Tarquinio told TechCrunch, and given that five years is an epoch in enterprise tech it's a good thing the company planned the z17 around a new AI accelerator chip.

ID please: If AI agents really take off, companies will have new security challenges authenticating that every AI agent that requests data from their network is legit, an issue that surfaced this week as developers put MCP through its paces. Okta grew into a nearly $18 billion company thanks to its work making it easier to authenticate human users on the network, and this week it introduced a tool for authenticating non-human users.

"The Okta Platform will now bring a unified, end-to-end identity security fabric to organizations for managing and securing all types of identities across their ecosystem, from AI agents to API keys to employees," the company said in a press release. AI agents don't respond to traditional identity-verification techniques like multifactor authentication, and as Okta put it, they have a "high blast radius" if something goes wrong.

Central planning: There are so many companies trying to convince developers and CIOs that their agentic AI development platforms are the way to go that it's getting hard to keep track, but Writer is in an interesting position. The combo model-builder/AI development tools startup this week unveiled AI HQ, its take on the agentic AI development platform.

"With our end-to-end platform, IT and business teams can now co-create agents that are deeply integrated, grounded in data, and designed to deliver real outcomes across every function," the company said in a blog post. And while building agents is hard enough, Writer also introduced an observability service for AI agents that allows users to "enforce policies, monitor performance, and troubleshoot with logs, traces, and step-by-step execution paths," it said.

Stat of the week

Software-development teams have come a long way from the days when they would deploy new code monthly or even quarterly, but there's still a long way to go before companies feel comfortable updating their code on a more frequent basis, which helps prevent security problems and allows them to react to changing business conditions much faster. According to new research from the CNCF, 29% of companies surveyed are deploying code "multiple times a day," as noted by Cloud Native Now.

Quote of the week

"We've been very measured in how we brought our AI message to the market to avoid people feeling like we're overhyping things." — Google Cloud CEO Thomas Kurian, in an interview with Big Technology, during a week when Google hosted a massive event at The Las Vegas Sphere to showcase a breathless behind-the-scenes hype video promoting the role its AI teams played in bringing a reproduction of The Wizard of Oz to the unique venue.

The Runtime roundup

Cybersecurity companies have mostly refused to discuss an executive order from the Trump administration condemning former CISA director Chris Krebs for his refusal to lie about the outcome of the 2020 presidential election and stripping security clearances from his colleagues at SentinelOne, according to Reuters.

Thanks for reading — see you Tuesday!