Software Engineer Intern

Runtime: AWS plants a Cedar; JFrog's under attack (again); Dell doubles down. + Anthropic's new Enterprise Frontier Safeguards data-storage service, and METR racks up an unwelcome AI bill. Welcome to Runtime! Today: How companies are using an AWS-built open-source tool to handle access control, OpenAI's hacker agents aren't the only ones on a new Artifactory flaw, 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. First up: Adjudicating agent access. Companies building AI applications quickly find themselves in need of a way to verify the identities of their own AI agents as well as agents connecting to their applications. Traditional role-based access control systems were designed to validate people, not agents, and correctly designating agent permissions is going to be one of the biggest challenges of the post-Hugging Face incident world. In order to solve this problem companies are turning to an open-source project called Cedar that began life inside AWS, Mary Branscomb reported Thursday for The Stack. "Anywhere I need authorization or I need to define an actor, an action and a resource - anything with RBAC or accessed based control for what an actor can do or not do - anywhere you need to make those decisions, Cedar fits really well," said Phil Estes, principal engineer at AWS. Companies like Cloudflare, MongoDB, and Stacklok are using Cedar to set access policies in code without having to define roles across big companies and organizations, which is already hard and will become nearly impossible if agents really take off. "Cedar is perfect for solving the problem where I have no idea what my policies are doing any more because I have more than ten of them, and for finding policy inconsistencies," project maintainer Lucas Käldström told The Stack. AWS earned a rather checkered reputation across open-source communities during the first decade of its existence for building cloud services around open-source projects without contributing much, if anything, back to those communities. That situation has changed over the last few years, and Cedar is a prime example: "We knew how it was valuable inside AWS and the open sourcing [decision] was essentially 'it's hard to believe this couldn't be valuable to others,'" Estes said. The rest of The Stack. Not easy being green: Weeks after JFrog found itself at the center of the OpenAI-Hugging Face hacking incident, hackers (human, The Stack think) are exploiting another vulnerability in its Artifactory repository manager. The critical vulnerability "lets any bad actor with no credentials trivially mint themselves an admin token and thereafter, run riot," The Stack's Ed Targett reported. Chipping in: Broadcom reported a 221% increase in chip revenue Wednesday, as part of a $29.6 billion quarter. And CEO Hock Tan doesn't forsee an end to the AI boom just yet: "We expect [customers'] need for compute infrastructure to inflate even more in 2027 and 2028," he said on a conference call following the release of the results. Gotta serve somebody: Dell's blowout quarter is just another sign that demand for enterprise AI compute is changing everything, a decade after the future of the on-premises server market seemed very much up in the air. The company reported $47 billion in fiscal second-quarter revenue Tuesday, up 58% compared to the same period last year, and raised its guidance for the full year by $25 billion. Self storage: Anthropic customers were not very happy about a data-retention policy the company put into place alongside the release of Claude Fable 5.0, but the launch of Fable 5.1 Tuesday introduced something that appears to meet them halfway. Enterprise Frontier Safeguards still requires customers to make model-usage metadata available to Anthropic, but now they can store that data on their cloud instances and protect it according to their own policies. METR's runnin': The non-profit METR organization is either an important whistle-blower warning of the growing sentience of advanced AI models or a classic case of AI psychosis, depending on which segment of enterprise tech you've been traveling through this week. But critics of its knowledge of modern cybersecurity practices got a boost Monday after the group acknowledged that a flaw in an AI-created internal app allowed anyone to access its AI model account, and somebody (or somebodies) racked up a $600,000 bill before getting caught. Double dip?: The government-to-industry hiring pipeline long predated the AI boom, and will surely outlast it, but Matt Clifford's decision to join Anthropic while keeping his position as the head of the UK's Advanced Research and Invention Agency is raising eyebrows even within those jaded circles. "[Clifford's] intention to remain as Chair of ARIA, which invests taxpayer money in cutting-edge research - much of it AI-related or AI-enabled - creates a clear conflict of interest," Science, Innovation and Technology Committee Chair Chi Onwurah said in a statement. Stack ranking: enterprise moves. Don Dama is the new CEO of Aerospike, joining the NoSQL database company after leadership roles at SingleStore and Mirantis. Dinesh Keswani is the new CTO and partner of Goldman Sachs, joining the financial powerhouse after three years as CTO of Nomura. Tom Bonos is the new chief revenue officer at Sumo Logic, joining the observability company after serving as chief operating officer and chief revenue officer at Applause. Julia Brau Donnelly is the new chief financial officer at Sierra, joining the customer-service agent company after holding finance leadership roles at Pinterest and Wayfair. Juergen Lindner is the new chief marketing officer at Calix, joining the service-provider software company after marketing leadership roles at ServiceNow and Oracle. The Stack is also reading: "Revisiting Joel's Test:" exe.dev's Philip Zeyliger suggests a new update to Joel Sposky's famous advice for evaluating the performance of software teams with the AI era in mind. Microsoft to start disclosing Azure quarterly revenue as company consolidates business units: From the "it's about goddamn time" department, Microsoft will finally report the revenue total of its most important service, which came in at $29.4 billion during its last quarter. Thanks for reading - Runtime is off this weekend for the US holiday, see you Tuesday!

Open intelligence, Open Data: SingleStore delivers zero-copy compute on Apache Iceberg. Jul 2, 2026 Organizations today are making a fundamental shift in how they think about data. Instead of locking information inside proprietary systems, enterprises are increasingly standardizing on open data formats that provide flexibility, interoperability, and long-term ownership of their data assets. At SingleStore, SingleStore, Inc. believe open data is the foundation for the next generation of agentic applications. That's why SingleStore, Inc. is announcing its third generation Cluster architecture, running on zero-copy data fabric with direct support for Apache Iceberg. Together, these innovations enable organizations to build and deploy Agentic Applications using SingleStore's HTAP engine, without duplicating Open Data across workloads or applications. The rise of Open Data. Data is the fuel for modern intelligent applications and AI, but too often organizations face a difficult tradeoff: either move data into specialized systems to gain performance and generate insights, or leave data in open storage and sacrifice operational capabilities. Apache Iceberg has emerged as the leading open table format because it solves many of the challenges of managing large-scale data lakes. Organizations gain a standardized, open way to store and govern data while maintaining freedom of choice across tools and platforms. Agentic applications need more than readable data. Agents have to discover information, run transactions, generate insights, and keep pace with data as it changes - which means the underlying platform needs to actively participate, not just serve files. Reading isn't enough; the platform has to read, write, and reason. HTAP engine for agentic apps. SingleStore's unique HTAP (Hybrid Transactional and Analytical Processing) architecture enables organizations to unify operational workloads, analytics, and AI on a single engine. With SingleStore, applications and AI agents can: * Read data from open Iceberg tables * Write updates and operational changes * Perform real-time analytics * Execute vector search and retrieval * Combine structured and unstructured data * Generate insights and automate workflows using agentic workflows This ability to read, write, and reason on data makes SingleStore the ideal engine for agentic applications. Rather than moving data between transactional systems, warehouses, lakes, and AI infrastructure, organizations can build intelligent applications directly where data lives. Third generation clusters using zero-copy data fabric. Today SingleStore is introducing its third generation cluster architecture, designed to deliver extreme scalability and isolation while sharing data across hundreds or thousands of applications. SingleStore's Zero-Copy data fabric enables compute clusters to attach any database without copying, duplicating, or relocating data. This reduces organizational complexity, and removes the challenge of managing disparate silos of data between operational and analytic database engines. Organizations can create clusters, scale workloads, and deploy AI-powered applications directly against their data while maintaining a single source of truth. The best part is this is an automatic upgrade for existing customers, and no migration is needed to move from legacy compute "workspaces" to third generation clusters. The result is faster time to insight, lower costs, and dramatically simplified data operations. Smart Read Replicas. Agentic workloads often compete with operational systems for resources, creating bottlenecks and unpredictable performance. Smart Read Replicas let teams attach dedicated clusters to a production database without copying the underlying data, so applications, AI inference, and analytics can scale independently while still reading from the same governed, up-to-date source. Database Branching. Just as software teams branch code, data should work the same way. Database Branching lets teams spin up full database environments in seconds without duplicating the underlying data, so developers can test features, data engineers can validate pipeline changes, and AI teams can experiment with models and prompts - all against real production data, and without the traditional costs of duplication. Smart DR. Traditional disaster recovery means paying for replicated infrastructure that sits idle most of the time. Smart DR uses the same Zero-Copy data fabric to let organizations deploy hot or cold clusters across cloud regions - cutting recovery time and operational complexity while optimizing total cost. Open Data meets open intelligence. Apache Iceberg solves the problem of where data lives and who can access it. SingleStore solves what you can do with it: transacting, analyzing, searching, and reasoning in real-time, without moving data to yet another system. Together, they close the gap between open data and intelligent applications. The architecture described here - Gen3 HTAP clusters, Zero-Copy data fabric, read replicas, branching, smart disaster recovery - isn't just an operational convenience. It's what makes agentic applications practical at enterprise scale. Get started today. Ready to experience the power of open data, Apache Iceberg, and Zero-Copy fabric? Try SingleStore Helios and discover how a unified HTAP platform can help you build the next generation of data, AI, and agentic applications - directly on open data. Start building with SingleStore Helios today. On this page. Start building now. Get started with SingleStore Helios today and receive $500 in credits.