Runtime: Cloudflare's "OS" for work; ServiceNow's security surge

Cloudflare released an open-source version of its internal AI agent workspace, Cloudflare OS, which gives every employee an agent and workspace built around their company, enabling search across business-intelligence tools, document creation, and app building without developer help. Runware unveiled the Sonic Inference Pod, a 20-foot shipping container delivering 1 megawatt of compute for AI inference, claiming it is the most cost-effective way to serve inference workloads. ServiceNow launched six new security services at Black Hat, building on its acquisitions of Armis and Veza, to enable autonomous security workflows including the Vulnerability Resolution AI Specialist.

Runtime /tag/runtime/ Welcome to Runtime Today on Product Saturday: Cloudflare releases an open-source version of its internal AI agent workspace, ServiceNow continues to expand its platform with new security tools, 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 . Coffee not included : AI agents are changing a lot of assumptions about how software should be built and consumed https://www.runtime.news/enterprise-software-brings-out-the-guillotine/?ref=thestack.technology , and Cloudflare thinks there's also an opportunity to to interact with the tools they need to do their jobs. This week it introduced an open-source version of a workspace tool that the company developed internally to connect employees with agents. https://www.runtime.news/beyond-the-prompt-why-genai-apps-require-new-design-thinking/?ref=thestack.technology design a new user interface for knowledge workers Cloudflare OS "gives every person an agent and workspace built around their company: how it works, what it knows, and the systems it relies on," it said in a blog post https://blog.cloudflare.com/cloudflare-os/?ref=thestack.technology . Employees can use it to search for data across business-intelligence tools, create documents and presentations, and build apps without having to call in a software developer. Pod save America : Companies have been experimenting with small, modular data centers for years https://www.thestack.technology/container-sized-ai-pods-could-be-the-answer-to-dragging-data-centre-plans-hpe-says/ , but the AI boom is being delivered through more-or-less the same massive data-center complexes that were built for the cloud computing buildout. Runware thinks the time is right for a new approach as opposition to those enormous structures builds throughout the US and edge AI inference becomes more important. The Sonic Inference Pod "delivers 1 megawatt of compute in a 20-foot shipping container, with custom servers, PCBs, and a closed-loop cooling system engineered for inference performance," the company said in a press release https://www.businesswire.com/news/home/20260804942053/en/Runware-Unveils-the-Sonic-Inference-Pod-a-Modular-Data-Center-Built-Specifically-for-AI-Inference? gl=1 1g3lx4k gcl aw R0NMLjE3ODQ4MzI4MzAuQ2owS0NRanc5NGJUQmhEUUFSSXNBTjN2djB4RThTVU5OT0JjU0Y3YWhTZWZPTjhmbEE5c2sycHVkNVF4YlowZGxidl93TV9xbWZuVnRKUWFBaWhyRUFMd193Y0I. gcl au ODcwMTIzMjU1LjE3ODQ3NDI3NzUuMTk2ODA2NjQwNy4xNzg1Mjc1NDYxLjE3ODUyNzU1NTcuMTA2NDYyOTYxLjE3ODUyNzU0NjEuMTc4NTI3NTU1Nw.. ga MTI5ODQzMjIzMy4xNzg0NzQyNzc1 ga ZQWF70T3FK czE3ODU4NTExNTAkbzkkZzEkdDE3ODU4NTIwNjgkajM1JGwwJGgw&ref=thestack.technology . The pod uses the same custom hardware Runware built for its cloud-based inference platform, and the company claims the system is the most cost-effective way to serve inference workloads. SecurityNow, serenity later? : Agentic AI might put the final nail in the coffin of the "best of breed" era of enterprise SaaS, given that agents work much better with direct access to corporate data and other internal agents. This week ServiceNow continued its expansion beyond its IT service management roots with the launch of six new security services at Black Hat that build on its acquisitions of Armis and Veza earlier this year. The new services, described as the company's "autonomous security vision," include tools to help manage vulnerabilities, protect physical assets, and manage identity. "With new AI Specialists that complete security workflows autonomously, including the Vulnerability Resolution AI Specialist, these capabilities enable enterprises to prevent, contain, and remediate risk at machine speed, before threats become breaches," the company said in a press release https://newsroom.servicenow.com/press-releases/details/2026/ServiceNow-delivers-Autonomous-Security-the-industrys-most-complete-security-offering/default.aspx?ref=thestack.technology . PANoramic : Palo Alto Networks is exactly the company ServiceNow is trying to swipe a few dollars from, and it introduced a new version of its flagship security service this week at Black Hat in hopes of keeping customers in the fold. The PAN-OS 12.2 Ceres release "brings to market 55+ innovations Editor's note: that's 56, right? , including three flagship, core capabilities, designed to shift the balance of power back to defenders," the company said in a press release https://www.paloaltonetworks.com/blog/2026/08/redefining-network-security-for-the-frontier-ai-era/?ref=thestack.technology . Those core capabilities include something called Advanced Virtual Patching, which promises to reduce "the exposure window from the industry-average 55 days it takes to deploy a traditional patch down into a near-zero window of exposure," and Advanced IP Defense, which takes advantage of PAN's large installed base to share suspicious IP addresses across its customer base. The company said it was using "Frontier AI" to provide these capabilities, which implies that a founding member of Anthropic's Project Glasswing https://www.anthropic.com/glasswing?ref=thestack.technology is tapping into the capabilities provided by Claude Mythos 5 that non-participants can't access. Transfer station : Data silos are an almost universal problem across companies and organizations, as our profile of the Scottish Government's Chief Data Officer this week shows, and getting rid of those silos is all but mandatory to participate in the AI boom. This week Google Cloud announced new features for its BigQuery Data Transfer Service that helps customers ingest data from various sources into the popular data warehouse. Customers with data stored in the Iceberg format on cloud storage services like AWS's S3 will soon be able to directly transfer that data into BigQuery, and BigQuery DTS will also soon support transfers from MCP servers, Google said in a blog post https://cloud.google.com/blog/products/data-analytics/new-bigquery-data-transfer-service-capabilities/?ref=thestack.technology . It can also now take in data directly from Snowflake, and will soon extend that capability to SaaS services like HubSpot and Shopify. Join peers following The Stack on LinkedIn Join peers following The Stack on LinkedIn Measuring up : Tokenmaxxing was over almost before it started once the first bills came due, but new attempts to manage AI costs, known as "tokenomics," might not make much sense for most enterprises either, argues former OpenStack co-founder Rick Clark in a new contribution to The Stack . "Technology should be measured with the same metrics as the business. Not adjacent metrics, not proxy metrics, the same ones," Clark wrote. Paying off : Uber was patient zero for the tokenmaxxing epidemic, but this week it revealed that its engineering teams are starting to figure out how to effectively use AI coding agents. Coding output it's not clear what the metric was, of course doubled during the last quarter without blowing costs out of the water, the company said during its quarterly earnings call. No true Scotsman : As noted above, eliminating data silos is a big part of Scotland's Chief Data Officer Tom Wilkinson's job as the steward of public data for the Scottish Government. But even when those silos come down, encouraging public servants to share data across organizations is harder than it sounds, because a lot of those workers are "hoarding it purely out of that sense of obligation to protect it," he said. One to rule them all : One reason why AI inference from the model providers can be so rickety https://www.runtime.news/as-ai-adoption-surges-ai-uptime-remains-a-big-problem/?ref=thestack.technology is because running those models across multiple types of chips is much harder than it sounds. This week AMD acquired a startup that could help them tackle that problem from the other side, by designing bespoke chips for a given model. We're also reading: OpenAI posted another one of those blogs where it spun the delay of its next model, Astra, as a good thing because apparently the model is too good at hacking and OpenAI is t aking safety more seriously of late for some reason https://www.thestack.technology/openai-we-hacked-hugging-face/ . Meanwhile, OpenAI engineers gave a talk at Black Hat revealing more details https://www.youtube.com/watch?v=87DyyMV0kCY&ref=thestack.technology behind the incident that led to that newfound respect for AI safety. We're convening in London on November 4-5, for a series of exclusive workshops and fireside conversations on the rise of BYOC as a favoured enterprise SaaS deployment model; how many CDOs are consolidating their data estates with Apache Iceberg; how CISOs at FTSE 100 scale are handling supply chain risk with GSK's CISO and more. Ticket applications are subject to pre-vetting. Get in touch with ed@thestack.technology if you want to be in the room.