🔥 Top Story
Databricks wants to merge the two databases every company runs
At its Data + AI Summit on June 16, Databricks unveiled LTAP (Lake Transactional/Analytical Processing), an architecture meant to erase the line between transactional (OLTP) and analytical (OLAP) databases. Companies typically run separate operational and analytical databases stitched together with ETL, replicas, and pipelines; LTAP removes that duplication by design, keeping transactional, analytical, streaming, and operational data on a single copy of storage in the lake. Data is stored once in an open format on object storage with standard Postgres semantics, while dedicated compute engines serve transactional and analytics workloads so neither sacrifices performance. Lakebase, the foundation of LTAP, stores data in Unity Catalog using the same open formats as the Lakehouse. Databricks frames the shift around AI agents: when thousands of agents simultaneously read live transactional data and analyze historical context, the old split-system model breaks down, so it is betting agents (not people) are now the database's primary users. The company says Lakebase already serves thousands of customers and handles 12 million database launches per day across the platform.
💡 Why it matters: Collapsing the ETL/replica pipelines between operational and analytical stores can sharply cut data-engineering maintenance and latency/consistency issues, but teams must weigh the new single-platform lock-in and how workloads stay isolated.
🔗 Read more · The New Stack
Kubernetes & Cloud Native
From data residency to digital sovereignty: Architectural patterns for cloud native platforms
CNCF
A CNCF blog post argues that digital sovereignty has moved from a policy discussion to a practical platform-engineering concern, and lays out architectural patterns for cloud native platforms. As regulation has firmed up — the EU Data Act has been fully applicable since January 11, 2025 — sovereignty requirements now go beyond simple data residency to include operations, key management, and access control. The piece frames Kubernetes-based platform design around questions like where data is stored and processed, and who controls the encryption keys and the control plane. Its core message is that compliance should be built into the platform architecture from the design stage rather than bolted on afterward. It situates this in the broader shift, over the past two years, of sovereignty from policy into operational reality.
💡 Platform teams running multi-region, multi-cluster setups should bake data residency and control over keys and the control plane into their Kubernetes design early — retrofitting later turns into re-architecture cost.
AI & ML
Predicting model behavior before release by simulating deployment
OpenAI
OpenAI introduced Deployment Simulation, a method for predicting a model's behavior before release by simulating its deployment as realistically as possible. Starting from de-identified real conversations from a recent deployment, it resamples the next response using the candidate model, enabling pre-release risk assessment. OpenAI analyzed roughly 1.3 million de-identified conversations spanning GPT-5 Thinking through GPT-5.4 deployments, from August 2025 to March 2026. The method estimated actual risk rates more accurately than challenging-prompt baselines, surfaced behaviors such as "calculator hacking" before release, and appeared nearly indistinguishable from real deployment to the model. In aggregate, predictions had a median multiplicative error of about 1.5x (for example, a true rate of 10 per 100k might be estimated at anywhere from 6.67 to 15 per 100k).
💡 Measuring risk ahead of release by replaying real traffic instead of synthetic red-team prompts improves the reliability of LLM evaluation, and offers teams running their own models or agents a methodological lesson: evaluate in production-like conditions before shipping.
Cloud Updates
Google named a Leader in IDC MarketScape SIEM 2026 Vendor Assessment
Google Cloud
Google said it was named a Leader in the 2026 IDC MarketScape Worldwide SIEM Vendor Assessment. The recognition reflects sustained investment in Google Security Operations (SecOps), which combines Mandiant's frontline threat expertise, broad automation, and AI agents. IDC highlighted the Alert Triage and Investigation agent — which collects evidence, runs correlated searches, and produces a transparent verdict — as a key strength for reducing analyst workload. The assessment comes as security teams face mounting pressure from adversaries who use AI to act with unprecedented speed, scale, and sophistication. The 2026 result continues Google's run of Leader designations in prior IDC MarketScape SIEM evaluations.
💡 Agentic auto-triage and investigation in the SIEM can offload first-line SOC work, but operators must build in explainability and false-positive verification for the agent's verdicts.
Introducing Brazos: Bringing liquid cooling to air-cooled data centers
Google Cloud
Google Cloud introduced Brazos, a system that brings liquid cooling to data centers originally built for air cooling. The driver is heat: next-generation AI and HPC chips now routinely exceed 1,000 W of Thermal Design Power (TDP), beyond what standard air cooling can manage. Rather than requiring brand-new facilities, Brazos focuses on adding liquid cooling to existing air-cooled data centers. Across the industry the same shift is underway — as accelerators like NVIDIA's Blackwell- and Rubin-class parts push past 1,000 W TDP, liquid cooling has become a baseline requirement, often in a hybrid form that cools high-TDP accelerators with liquid while keeping air cooling for lower-power components such as NICs, SSDs, and power supplies. (For Brazos's exact design and figures, Google Cloud's original post is the authoritative source.)
💡 For organizations adopting 1,000W+ TDP accelerators, retrofitting existing air-cooled facilities — rather than building new liquid-cooled ones — becomes a practical option, and power/cooling design becomes a central constraint on AI infrastructure.
How Atlas scales hundreds of merchant databases with Cloud SQL Enterprise Plus edition
Google Cloud
Google Cloud shared how Atlas — which positions itself as an operating system for restaurants — scales hundreds of per-merchant databases on Cloud SQL Enterprise Plus. Atlas bundles nearly everything a restaurant needs to start, run, and grow: online storefronts, point of sale, third-party logistics, food-platform integrations, customer loyalty, and AI tools. Because each merchant gets its own database, reliably operating hundreds of instances becomes the central challenge. Cloud SQL Enterprise Plus suits that multi-tenant pattern with higher performance, improved availability, near-zero-downtime maintenance, fast failover, and read scaling. The Atlas story illustrates how a managed database can scale many small tenant databases without a proportional operations burden. (For exact instance counts and performance figures, the original post is the authoritative source.)
💡 In a database-per-tenant pattern, maintenance and failover automation increasingly drive operating cost as instance counts grow, so a managed database's near-zero-downtime maintenance and failover speed become the real differentiators.
Introducing new Explores and Merge Queries in Looker
Google Cloud
Google Cloud redesigned Looker's Explore experience and added new Merge Queries capabilities alongside integrated AI. The new experience includes a Join data page that lets you join two or more queries and view the results as if they were a single query, with merges supported on effectively unlimited data as long as it sits on the same BigQuery connection. A self-service Explore — with a drag-and-drop interface, tabbed dashboards, and paginated reporting — is available in preview. Integrated AI assistants turn complex analytics workflows into natural-language interactions that feel as familiar as a Google search, letting users ask questions of their uploaded data and get immediate answers. The goal is to let business users, not just analysts, uncover actionable insights on their own.
💡 As natural-language, self-service Explores grow, analysts do less query-by-request work — but since semantic model (LookML) definitions and governance determine answer quality, careful data modeling matters more, not less.
How Siemens "slices the elephant," advancing agentic workflows for industrial software development
Google Cloud
Google Cloud described how Siemens advances agentic workflows in industrial software development using a "slice the elephant" metaphor — breaking a huge problem into manageable pieces. For a technology company like Siemens, software is the nervous system of factories, energy grids, and transportation networks, so modernizing a vast legacy codebase all at once is impractical. Instead, large modernization efforts are split into small, tractable slices that agents work through — reading existing code, applying changes, and verifying them. Google Cloud's agentic code modernization generally uses Gemini-based tooling to automate refactoring legacy modules, documenting dependencies and logic, and generating tests. The key is that agents take on repetitive work rather than humans doing every step, with verification built into each slice to keep the process safe.
💡 When handing large legacy modernization to agents, success hinges on slicing the work into small units with automated verification (tests) at each step — closer to a continuous refactoring pipeline than a one-off migration.
How customer collaboration is shaping the future of GenAI security with Model Armor
Google Cloud
Google Cloud described how customer collaboration is shaping the roadmap for Model Armor, its generative-AI security product. Model Armor provides runtime security for generative and agentic AI: it blocks prompt injection and jailbreak attempts, detects malicious URLs hidden in prompts or responses, and helps prevent leaks of sensitive data such as PII, credentials, and financial information (via integration with Sensitive Data Protection). It protects many LLMs — not just Gemini but also OpenAI, Anthropic, Llama, and more — through a REST API, and can act as an "AI firewall" when deployed inline with Vertex AI or via network service extensions and Apigee API gateways. Templates let teams configure filters and thresholds for different safety and security confidence levels. Google emphasized that the best products are built in partnership with customers, refining the service based on real-world feedback.
💡 Prompt injection and data leakage are core threats when putting AI apps in production, so a policy-based filtering layer (an AI firewall) in front of the model, applied consistently across a multi-model environment, is a sound design.
DevOps & Infrastructure
Introducing tfctl: The CLI for HCP Terraform and TFE
HashiCorp
HashiCorp introduced tfctl, the first dedicated command-line interface for HCP Terraform and Terraform Enterprise (TFE), and published it on GitHub. For years, automating HCP Terraform platform operations meant building and maintaining custom tooling on top of the API; tfctl replaces that with an official, standard CLI. The stated goal is to let teams manage platform operations — workspaces, runs, variables, and similar tasks — consistently from scripts and CI pipelines. In effect, it standardizes the assortment of bespoke API wrappers and in-house scripts organizations have relied on into a single supported tool. tfctl is available to download and use directly from GitHub. (For the exact command surface and feature scope, the original post and the GitHub repository are the authoritative reference.)
💡 Standardizing HCP Terraform operations on an official CLI instead of homegrown API scripts reduces automation maintenance cost and tooling fragmentation for platform teams.
SpaceX acquires Cursor for $60 billion. Can it fix Musk’s coding division?
The New Stack
According to The New Stack and other outlets, SpaceX announced it will acquire Anysphere, the maker of the AI coding tool Cursor, for $60 billion. Reporting from CNBC and Reuters describes it as an all-stock deal expected to close in the third quarter, pending regulatory approval. The agreement exercises an option SpaceX secured in April that gave it the right to either pay roughly $10 billion for a partnership or buy the company outright for $60 billion later in the year. The announcement landed four days after SpaceX's Nasdaq IPO. Cursor, founded in 2022, grew quickly as a tool that helps developers generate, edit, and review code. Observers read the deal as a move to strengthen SpaceX's AI efforts — built around xAI, which merged with SpaceX in February 2026 — and to compete with coding tools from Anthropic and OpenAI.
💡 As major coding assistants get absorbed into large platforms, the lock-in, pricing, and data-governance risk in the dev toolchain rises — so teams should keep an exit strategy in mind when standardizing on an editor or agent.
What’s new with Terraform + Ansible
HashiCorp
HashiCorp shared what's next for its Terraform + Ansible integration, building on the Terraform actions and integration vision it outlined last year. The aim is end-to-end lifecycle management in one pipeline: Terraform handles infrastructure provisioning and drift detection, while Ansible handles configuration management, patching, and application workflows. Supporting that, Red Hat has released Ansible Certified Content Collections (such as hashicorp.terraform) rebuilt for API-first integration with HCP Terraform and Terraform Enterprise, and Ansible Automation Platform workflow templates can call Terraform Enterprise. IBM's acquisition of HashiCorp aligns Terraform and Ansible under one umbrella, opening room for deeper integration. The end picture: Terraform provisions infrastructure and detects drift, then Ansible takes over configuration, patching, and ongoing operations at scale.
💡 Splitting roles — Terraform for provisioning, Ansible for configuration/patching — into one pipeline chains automation from drift detection to remediation, but the two tools' state and permission boundaries must be designed clearly to avoid overlap and conflict.
Why did my AWS bill spike? There’s now an agent for that
The New Stack
AWS introduced the AWS FinOps Agent in public preview — its third specialized "frontier agent" for IT operations. The agent answers cost questions in plain English, surfaces optimization opportunities, investigates cost anomalies, and runs recurring FinOps workflows on a defined schedule. It correlates cost changes with AWS CloudTrail events to identify the change that drove a spike, then produces an investigation summary with the likely root cause and the responsible owner. It delivers findings by posting to a Slack channel or opening a Jira ticket, so the engineer who owns the resource gets the context and can decide what to do next. In short, it's an agent built to answer the question "why did my AWS bill spike?"
💡 Automating cost-anomaly investigation via CloudTrail correlation and routing it to Slack/Jira speeds up FinOps triage, but it's safer to keep a verification step before trusting the agent's named root cause and owner.
The siloed-data era is over. Here’s what comes next for AI agents.
The New Stack
This New Stack piece argues that the era of siloed data — split across departments and systems — is ending, and that a unified data foundation for AI agents is what comes next. It notes that generative AI is now mainstream, with some 700 million people using ChatGPT every week, and that the next phase, agentic AI that takes actions on its own, is well underway. For agents to actually do work, they must read across scattered data and act in real time, yet today's fragmented storage and access patterns get in the way. The takeaway is that a governed, unified, connected data foundation becomes a prerequisite for the agent era. The article stresses that beyond the race on model quality, the real task is building a data foundation agents can access safely.
💡 The usefulness of agentic AI hinges more on data accessibility and governance than on raw model quality, so platform teams should first design a unified data layer and permission model that agents can safely traverse.
Anthropic pauses Claude Agent SDK subscription change on day it was due to take effect
The New Stack
Anthropic paused a billing change for developers who use its Claude Agent SDK on a paid subscription — on the very day it was due to take effect, June 15. The company had announced on May 13 that from June 15 it would move automated Agent SDK usage onto a separate monthly credit. Under the original plan, everything a subscriber does — chatting, coding in the terminal, or using Agent-SDK-based third-party tools — drew from a single monthly pool; the change would have shifted Agent SDK usage to a separate, capped credit for Pro, Max, Team, and Enterprise subscribers (ranging from about $20 for Pro up to about $200 for top-tier Max and Enterprise). With the pause, the Agent SDK, claude -p, and third-party apps still draw from regular subscription limits. Anthropic said it is working to better align the plan with actual usage patterns, with the pricing rivalry with OpenAI cited as context.
💡 For teams running agent/CLI automation on a subscription, a vendor's billing-policy change is itself a cost and architecture risk, so it's worth reviewing how dependent automated workloads are on the billing model (subscription vs API).
This digest was collected from RSS feeds and summarized by AI (Claude). See the original links for full details.