Runtime: AWS rolls out a Nitro boost

Welcome to Runtime! Today on Product Saturday: AWS has a new option for sandbox shoppers, Databricks lets the Genie out, and more.

Please forward this email to a friend or colleague! If it was forwarded to you, sign up here to get Runtime for free every week, or level up here.

Ship it

Sand dollars: It's been roughly ten years since enterprise tech decided to embrace the container as the preferred vehicle for their applications. Lightweight and flexible, containers allowed companies to launch and close applications much faster than virtual machines, and their popularity led to all kinds of managed container services at the hyperscalers.

At AWS, customers can choose to run containerized applications on EKS (which gets Kubernetes involved) or ECS, but those applications (or agents) need to share resources with other apps running on a tenant and that might not be a great idea for sensitive workloads. This week the cloud provider shaved five minutes off CEO Matt Garman's upcoming re:Invent keynote speech by announcing AWS Nitro Sandboxes, which can run on either managed container service or the flagship EC2 compute service.

The new service was designed to give "each container the boundary of a hardware-isolated virtual machine (VM). Each sandbox gets the security properties of the Nitro System (strong isolation, zero operator access to customer data, and attestation) while keeping the familiar container experience," AWS said in a blog post. This would allow companies to run sensitive apps or agents on managed container services as if they were running on their own VM, boosting the productivity of their instances.

The strategic importance of sandboxes is pretty clear to most IT organizations working with AI agents at this point, and the new service challenges similar ones from companies like Vercel. But managing costs could be tricky.

Delivery, continued

Shift left: Containers became so popular that companies who wanted or needed to manage their own data centers wanted to get in on the action without having to manage public cloud apps and on-premises apps separately. Red Hat's OpenShift became one of the more popular ways to do that over the years, and this week it introduced OpenShift 5 at KubeCon North America.

The new version "enables IT teams to deploy AI workloads, virtual machines (VMs) and containerized workloads side-by-side through a simple, in-place update from Red Hat OpenShift 4," Red Hat said in a press release. It now supports MCP, and comes with several agents that can help operations teams diagnose and fix issues as they come up.

Wishcasting: If the personal agent is going to catch on across enterprises, it seems more likely that they'll bet on a system that's already managing their corporate data over one managed by Meta. This week Databricks introduced the second generation of its Genie enterprise assistant, which now runs autonomously on behalf of its users.

The idea is that Genie can monitor ongoing projects and take on work as needed, emailing the human user when a decision comes up that requires their input. "Since it runs within a company's existing Databricks environment, Genie is governed by Unity Catalog and Unity Gateway, so every action stays within the controls companies already have in place," Databricks said in a blog post.

Reset button: Even companies with the strongest operations teams and security tools can be hit by catastrophic errors or ransomware attacks that hijack data, which makes having a backup and recovery system absolutely essential. This week Commvault rolled out a new service that promises to recover business context and application logic as well as data in the event something goes wrong.

Commvault Recovery Studio "will orchestrate the recovery of applications, identities, configurations, infrastructure dependencies, and datasets together in an automated, runbook-driven workflow to help organizations resume business operations fast," the company said in a press release. Most companies have backup systems, of course, but the bigger they are the longer it takes, and Commvault promised it will help customers get back up and running in less than the industry standard (according to its data, anyway) 28-day recovery time.

The rest of The Stack

Stop and go: Some sort of networking error forced the US FAA to issue a ground stop to all flights Friday afternoon, which wasn't lifted for several hours. Secretary of Transportation Sean Duffy blamed the issue on the FAA's outdated equipment, but did not provide a lot of details about the root cause in a press conference.

Updated: Air-traffic control networking snafu said to cause nationwide ground stop in US

Some sort of update to the network operated by L3Harris went haywire and forced the FAA to halt all takeoffs across the US Friday afternoon.

The StackTom Krazit

Snow day: Microsoft finally began to recover Friday afternoon from an outage that began with Exchange but wound up spreading to a bunch of different services on Thursday and into Friday. "...the incident was introduced by an issue during an earlier network capacity upgrade that surfaced following a recent network change," Microsoft said on X Friday.

Global Microsoft 365 outage: Engineers try the classic move

Servers started failing Microsoft’s automated health checks

The StackEdward Targett

Downright mythical: Anthropic's Project Glasswing has certainly found a lot of flaws in the world's software, but has it really prevented any serious attacks? New research from VulnCheck found that only two flaws out of the tens of thousands that have been reported have been exploited, and Anthropic isn't providing a lot of fixes for the ones it does find.

Seven months on, Anthropic's Project Glasswing finds just two exploited CVEs and a lot of slop

“I don't know how anyone can read these numbers as anything but damning.”

The StackIain Thomson

Best intentions: AI coding agents are rendering a large amount of conventional wisdom about the best ways to manage software projects obsolete, OpenStack co-founder Rick Clark wrote in the first of a series of essays about the future of software development. The systems we built for humans to manage that process are meaningless to agents, and that will require some adjustment.

In the AI coding agent era, we are building internal combustion horses

Coding agents now write most of the code. The enterprise is bolting them onto a process built for people who wrote it by hand.

The StackRick Clark

Quote of the week

"...old Jensen, the black leather jacket, the Godfather of AI, the rock star viewed as the face of the AI boom, is over and in the past; his value to the company pre and post this week is dramatically different.” — Financial analyst Dan Ives describing the abrupt end to Jensen Huang's perch atop the tech industry.

We're also reading:

How Nvidia’s Jensen Huang – and Trump – jumped the guardrails: Semafor has more details on Nvidia's decision to leverage Huang's close relationship with President Trump to get what it wanted in negotiations with regulators, which might cause him more trouble than any Silicon Valley divorce lawyer.

Thanks for reading — see you Tuesday!