AI Industry
News Digest

Anthropic Acquires Stainless, the SDK and MCP Toolmaker Behind Much of the AI Ecosystem

If you build on the Anthropic, OpenAI, Google or Cloudflare APIs, you have almost certainly used an SDK generated by Stainless without realising it. Anthropic has now acquired the company outright. Stainless generates official SDKs across TypeScript, Python, Go, Java and Kotlin from a single API spec, and powers the command-line tools and MCP servers that let agents talk to APIs. Anthropic has powered every official Claude SDK with Stainless since the earliest days of the API, and reports around the deal value place it above $300 million.

The practical reading for everyone outside Anthropic is that a piece of the developer plumbing the wider ecosystem relies on now sits under a competitor, with hundreds of companies dependent on its output. Expect knock-on changes to how OpenAI, Google and Cloudflare maintain their official SDKs over the next quarter. This was a Direct find from Anthropic's own newsroom, beating mainstream coverage by several hours.

OpenAI Plugs ChatGPT Directly into Live Bank Accounts with a New Personal Finance Experience

If you have been waiting for an AI assistant that can actually see the numbers in your accounts before answering a money question, this is the first mainstream version of that. OpenAI has wired ChatGPT into Plaid (with Intuit coming next) and rolled out a finance preview for Pro subscribers in the US. Once a user links accounts, ChatGPT can pull balances, transactions, investments, subscriptions and upcoming payments into a dashboard, and answer questions grounded in the actual statements rather than generic templates.

It runs on GPT-5.5 Thinking, supports more than 12,000 financial institutions, and is read-only by design: the model cannot move money or see full account numbers. For anyone using ChatGPT for budgeting or financial planning workflows, this is a meaningful upgrade in grounding, though OpenAI is explicit that it is not a substitute for a regulated adviser.

OpenAI and Dell Partner to Push Codex into Hybrid and On-Premises Enterprise Environments

If your engineering organisation has been blocked from rolling out Codex because the data cannot leave your data centre, that constraint is starting to loosen. OpenAI has agreed a partnership with Dell to wire Codex into the Dell AI Data Platform and Dell AI Factory (the on-premises and hybrid stacks many regulated companies already use). Codex is now claimed to be in active use by more than four million developers a week, and bringing it closer to enterprise data should help with code review, test coverage, incident response and reasoning across large repositories without exporting source to the cloud.

Nvidia Hand-Delivers the Vera CPU, Its First Silicon Purpose-Built for Agentic AI, to Top Labs

If you build agent workflows that hit CPU rather than GPU bottlenecks, Nvidia has just shipped the answer it has been promising for a year. The first Vera CPUs arrived at Anthropic, OpenAI and SpaceXAI on Friday, with Oracle Cloud Infrastructure following on Monday. Each unit was hand-delivered by Nvidia's hyperscale and HPC lead, Ian Buck. Vera packs 88 custom Olympus cores using a spatial multithreading scheme that lets each core run two tasks, paired with LPDDR5X memory for 1.2 TB/s of bandwidth. Nvidia claims twice the efficiency and 50% better speed than traditional rack-scale CPUs.

The chip is lined up at Alibaba Cloud, ByteDance, Meta, Oracle, CoreWeave, Lambda, Nebius and Nscale, with Dell, HPE, Lenovo and Supermicro building the server platforms. If you are negotiating compute contracts or planning agentic AI capacity for late 2026, Vera changes the price-performance baseline you should be pricing against.

Nvidia and Dell Stake Out a Deeper Enterprise AI Partnership at Dell Technologies World

Sitting alongside the Vera launch, Jensen Huang used the Dell Technologies World stage to formalise a deeper enterprise partnership and to declare AI demand is going "utterly parabolic". For anyone tracking the channel, the practical takeaway is that Dell becomes a primary route to put Vera-class agentic AI infrastructure inside enterprises that are not buying directly from a hyperscaler. Combined with the same-day OpenAI and Dell Codex partnership, Dell is suddenly the joint-distribution vehicle for both the headline silicon and the headline coding agent of the week.

Musk Pushes a Fresh Grok Update Trained Partly on Colossus 2, with Agent Mode and Lifted Usage Limits

If you have been bumping into Grok rate limits or wondering whether the model would ever ship anything trained on the new supercluster, both questions got answered on the same day. Elon Musk posted on X to announce a new Grok release, confirming it was partially trained on Colossus 2 (xAI's next-generation Blackwell-only cluster), alongside a new Agent Mode capability and confirmed higher usage limits. For anyone using Grok inside an automated workflow, this is the most material xAI update in weeks and is worth retesting against your current Grok 4.3 or 4.20 calls to see whether agent-style tasks improve.

AWS Publishes a Nova 2 Moderation Guide and Adds Custom Code Evaluators to Bedrock AgentCore

If you run user-generated content or compliance-sensitive workflows on Bedrock, both posts from the AWS Machine Learning blog today are practically useful. The Nova 2 piece sets out prompt patterns for moderation tasks and shows where Nova 2 sits versus the smaller variants on policy enforcement. The AgentCore post adds a custom code-based evaluator track, so you can grade agent outputs with your own Python rather than the built-in scorers. Together they reflect AWS filling in the operational gaps that have been holding teams back from putting Bedrock agents into production.