6 min read · 11 sources

Anthropic just bet $11.6 billion that its future is in someone else’s cloud. The seven-year commitment to Akamai’s infrastructure is the biggest story of the day, and it’s not just about compute. It’s about who owns the stack that trains and runs frontier models.

The deal comes with warrants that could give Anthropic roughly a 5% stake in Akamai, and it can expand by another $9 billion. For an AI lab that has been building its own massive clusters, this is a strategic hedge. For Akamai, it’s a transformation from CDN to AI infrastructure provider. Engineers should read this as a signal: the cloud wars are no longer just about AWS, Google, and Azure. The big AI labs are placing their own bets on who gets to run the next generation of compute.

Anthropic’s $11.6 billion, seven-year commitment could expand by another $9 billion and gives the AI lab warrants for roughly a 5% stake in Akamai.

AWS Gets the NATO Sticker

Source: itpro.com ↗

AWS has become the first hyperscaler approved to handle ‘NATO Restricted’ (NR) classified information. The accreditation means NATO member states and defense partners can host sensitive workloads in any of AWS’s 15 Regions located in member nations.

This is a big deal for anyone running defense or public-sector workloads. It provides a common security baseline and an accelerated path to national accreditation, plus access to the AWS Trusted Secure Enclave - Sensitive Edition. For engineers, this opens a new compliance path on commercial cloud infrastructure that previously required dedicated or sovereign offerings. The evaluation was done by Spain’s National Cryptographic Centre, which adds a layer of trust that other hyperscalers will have to match.

Databricks Kills the Spreadmart

Source: databricks.com ↗

Databricks has acquired Row Zero, a spreadsheet platform founded by former AWS and Tableau engineers. The goal: bring a native, governed spreadsheet experience to its Genie AI coworker.

This directly attacks the “spreadmart” problem - the ungoverned Excel files that finance and operations teams pass around, each a potential source of data drift and audit failure. The integration combines Row Zero’s familiar interface with Databricks’ governance stack, including Genie Ontology, Unity Catalog, and Unity Gateway. The result is that agent actions on spreadsheet data become auditable and secure. For engineers, this means less time reconciling rogue spreadsheets and more time trusting the data pipeline.

The Database Researchers' Existential Question

Source: emptysqua.re ↗

A paper from Berkeley and Stanford argues that as foundation models improve, they natively reduce the inefficiency of agent-issued queries. This challenges prior work that predicted databases would be bottlenecked by speculative agent workloads.

The authors propose a research agenda for adapting databases for AI usage, but the reviewer is skeptical that AI will leave meaningful work for database researchers. The core question for engineers: can current database architectures, designed for repetitive OLTP, handle heterogeneous agent queries without a redesign? If the models get smart enough to write efficient queries, the database might not need to change at all. If not, we’re in for a decade of painful middleware.

UiPath Writes the Manual AI Never Had

Source: diginomica.com ↗

UiPath launched Cartographer and its Map of Work at FUSION 2026. The product aims to capture the undocumented rules, exceptions, and judgment calls that experienced employees know but LLMs lack.

CEO Daniel Dines highlighted three LLM limitations: no consequences, no on-the-job learning, and no exactness. Cartographer is positioned as the manual that gives AI agents the context they need for enterprise work. It uses AI to absorb existing documentation, interview experts, and track decisions through a ledger to keep process maps continuously updated. For engineers, this addresses the gap between LLM memory and true learning, which is critical for reliable AI agents in production. It’s a bet that the missing layer isn’t more model capability, but better context.

Devin Joins the Fortune 500 Watercooler

Source: devin.ai ↗

Devin now works across Microsoft Teams and Microsoft 365. This means native @ mentions in Teams, including DMs, group chats, and channel automations, plus cards for questions, approvals, and sign-ins.

The first-party Microsoft 365 integration adds six MCP servers - Mail & Contacts, Calendar, OneDrive & SharePoint, Microsoft To Do, Teams, and Directory (Entra ID) - letting Devin act as the signed-in user to read files, send emails, and manage tasks. For engineers, this is about reducing context switching. Over 90% of Fortune 500 companies coordinate work in Teams, and now Devin can participate directly. The delegated per-user permissions and resource-specific consent controls are designed for security review, which is a good sign for anyone worried about agent access.

Docker Makes Agents Run Overnight

Source: docker.com ↗

Docker introduced Cloud Sandboxes, extending its microVM-based sandbox to run on Docker-managed compute. This lets agents run for hours or overnight without a laptop.

A single command, sbx move my-project --to cloud, captures the sandbox filesystem and recreates it on the other side, enabling local-to-cloud handoffs and parallel execution of up to 100 tasks. Pricing starts at $0.07 per hour. For engineers, this addresses the shift to long-horizon agent tasks that need always-on, scalable compute with the same isolation model as local sandboxes. It’s a pragmatic answer to the “my laptop died and my agent was mid-task” problem.

Ubuntu's Weekly Kernel Fire Drill

Source: theregister.com ↗

Canonical is moving Ubuntu to a weekly kernel release cycle due to a flood of CVEs, driven in part by AI-assisted bug hunting outpacing defenders’ patching capacity.

The change accelerates the delivery of security fixes for the kernel, but it means more frequent patching windows for production systems. For engineers, this is a double-edged sword: faster fixes for newly disclosed vulnerabilities, but a higher operational burden to keep up with the cadence. Specific version numbers and dates weren’t detailed, so expect the usual upgrade scramble.

Pinecone's Data Plane Goes BYOC

Source: pinecone.io ↗

Pinecone announced general availability of Bring Your Own Cloud (BYOC) on AWS, Google Cloud, and Azure. Customers can now run Pinecone’s data plane inside their own cloud account and region.

The architecture separates the control plane (which handles management and doesn’t store customer data) from the data plane (which stores and processes vectors, documents, and payloads). It uses a zero-access model with no SSH, VPN, or standing IAM roles. For engineers, this enables sensitive AI workloads to stay within customer-controlled boundaries while offloading platform operations to Pinecone. The outbound, pull-based mechanism means Pinecone manages upgrades and scaling without requiring standing inbound access.

Kubernetes Patches a Windows NTLM Hole

Source: freenode.net ↗

Kubernetes patched CVE-2026-76654, a medium-severity flaw (CVSS 5.8) in the Windows kubelet. It allows NTLM coercion via a crafted symlink in a pod volume’s subPath resolving to a UNC path.

The kubelet would authenticate to an attacker-controlled share, leaking its NetNTLMv2 hash, which could be cracked or relayed to impersonate a domain-joined node. Fixed builds are 1.34.12, 1.35.9, 1.36.5, and 1.37.1, which refuse UNC symlink targets on Windows. If you’re running Windows kubelets, upgrade immediately. This is a reminder that the Windows node is often the weakest link in a Kubernetes cluster.

Ando Wants to Replace Slack with Agents

Source: techcrunch.com ↗

Ando, a startup founded by Sara Du, emerged from stealth with a team messaging platform designed for both humans and AI agents. It’s positioning itself as a full Slack replacement.

The app gives agents their own identities and inboxes, lets them join channels and conversations without being tagged, and can message humans proactively. It raised $20 million in pre-seed and seed funding from Accel, Index Ventures, and Emergence. For engineers, this addresses the “meat proxy” problem where humans relay agent work, aiming to make agents direct participants in team coordination. Whether it can displace Slack’s network effect is another question entirely.