Fluidcloud makes a multicloud model; Galileo opens up agent controls

Welcome to Runtime! Today on Product Saturday: Fluidcloud releases what it calls a "large infrastructure model," Galileo releases an open-source way to manage agents, 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

Movin' out: Thanks to modern infrastructure-as-code tools, it's never been easier to move workloads between cloud providers, but that in no way implies that it's easy. A startup called FluidCloud released a new AI model this week that it said paves the way for switching cloud providers or moving between regions on a particular cloud provider.

The Large Infrastructure Model "generates production-grade infrastructure as code (IaC) with 99%+ accuracy and is purpose-built for multi-cloud architecture reasoning, cross-cloud Terraform migration, and operational continuity," the company said in a press release. It can generate new Terraform files based on older ones that can then be used to stand up infrastructure on a new provider without having to manually reconfigure all those settings, which can differ quite a bit from provider to provider.

ID, please: Enterprise AI agents are going to have huge ramifications on the methods companies use to manage identity on their networks, given how quickly they can be created and how dangerous malware disguised as an agent could be. Sailpoint introduced new features for its flagship platform this week that help users assess agent identity and understand what that agent is permitted to access.

A lot of older identity-management tools "[were] simply not designed for an environment where a developer can spin up a new AI agent, grant it access to sensitive data, and decommission it in a matter of minutes," the company said in a blog post. Sailpoint Platform can now extend or deny privileges to identities on a network in real time, and can manage agents created by several big enterprise software companies, including The Big Three cloud providers as well as Salesforce, ServiceNow, and Snowflake.

King of insight: Identity management won't be the only challenge when it comes to herding enterprise agents; if companies manage to deploy them at scale they'll need some kind of orchestration tool to make sure they're all pointed in the right direction. As Runtime has covered, every enterprise software company with a pulse is trying to convince customers to use their agent-management product and keep them on their platforms, but a startup called Galileo released an open-source control plane that could give businesses a little more flexibility.

"Agent Control lets developers define guardrails once and apply them everywhere," co-founder and CEO Vikram Chatterji told The New Stack. It was released under the Apache 2.0 license, and also allows users to make changes to their agents without having to take them offline, the company said in a press release.

Search for tomorrow: Cribl's observability tools help companies understand what's happening on their networks through telemetry data, and its Cribl Search product allows them to dig deeper when looking for something specific. This week the company said it had added "agentic telemetry" to a new version of Cribl Search that will allow users to keep up with the activity generated by AI agents.

"The new Search experience helps organizations cut log management costs, resolve security and IT incidents faster, and streamline tooling," Cribl said in a press release. It can automatically tag new telemetry data so that administrators can quickly find what they need, and the company said performance improvements to the new Search product will also make it "more affordable" to use it with agents, which generate a lot more activity than traditional applications.

No idling: One of the reasons companies like Anthropic are struggling with AI reliability is that GPUs are relatively inefficient processing engines compared to CPUs, but one person's trash is another person's treasure. This week AI inference cloud FriendliAI introduced a new service called InferenceSense that lets customers rent those bored GPU cycles to other customers and split the proceeds with the company.

"What we are providing is that instead of letting GPUs be idle, by running inferences they can monetize those idle GPUs," founder and CEO Byung-Gon Chun told VentureBeat. The company likened it to "AdSense for GPUs" in a blog post, but also noted that if the primary customer realized they need that excess capacity, "InferenceSense preempts immediately—your jobs always come first."

Stat of the week

The examples are really starting to pile up at this point; companies that take the extra steps to prepare their data for AI applications have a much easier time putting those applications into production than companies that don't. New research from DataHub found that while 90% of survey respondents said their data was "AI ready," a lack of trusted data delayed AI rollouts for 61% of those surveyed.

Quote of the week

"If Claude outages push people back to ChatGPT twice a week, benchmark scores are secondary. Reliability is the product now." — JustPaid head of growth Harshith Vaddiparthy, pointing out that model availability is becoming more important than model performance as enterprise AI applications start to take off.

The Runtime roundup

AWS and Cerebras announced a joint AI inference service that will pair chips from both companies to handle different parts of inference workloads in the hopes of improving performance.

Companies spent $125.3 billion on servers in the fourth quarter and $444.1 billion during 2025, an 80.4% jump compared to 2024, according to IDC.

Thanks for reading — see you Tuesday!