AI Industry
News Digest

Ten Finance Agent Templates Land in Claude, with Moody's Data and Jamie Dimon on Stage

If you build with Claude or use it inside a regulated function, this is the most consequential drop of the day. Anthropic released ten ready-to-run agent templates covering pitchbook generation, KYC screening, audit checks, credit memos, and month-end close. Each ships as a plugin in Claude Cowork and Claude Code, and as a cookbook for Claude Managed Agents. The templates are reusable patterns for any structured back-office work, so even if you do not work in finance the scaffolding is worth a look.

The launch landed at an invite-only briefing in New York where Dario Amodei shared a stage with JPMorgan chief executive Jamie Dimon, who publicly endorsed AI capital expenditure at trillion-dollar scale. Moody's full ratings and risk data, covering more than 600 million companies, is now available inside Claude as a native app. That removes the friction of separate seats and copy-pasting between systems. If you do due diligence, supplier risk reviews, or background research on private companies, this is the kind of data partnership that quietly changes how a chat assistant earns its place in the workflow.

Claude Add-ins for Excel, PowerPoint, and Word Go Generally Available; Outlook Enters Beta

If you live in Microsoft 365, this is the practical headline. The Claude add-ins for Excel, PowerPoint, and Word are now generally available, while Claude for Outlook is in beta. Context carries automatically between the four applications, so a workflow that starts as analysis in Excel can finish as a deck in PowerPoint without re-pasting source material or rewriting the brief.

In Excel, Claude builds models from filings and feeds, audits formulas across linked workbooks, and runs sensitivity analyses. In Word, it edits documents against your own templates. For anyone who has been forced to choose between Copilot and Claude, you can now run them side by side inside Office.

Anthropic Commits $200 Billion to Google Cloud over Five Years

Reported by The Information and confirmed in wider coverage today, Anthropic has committed to spend $200 billion on Google Cloud and TPU capacity across the next five years. This is infrastructure news with a direct product implication: it locks in the compute headroom needed to keep Claude available at low latency as demand from the new Microsoft 365 add-ins and finance agents scales. It also signals that, despite the AWS partnership, Anthropic's primary training and serving stack will continue to lean heavily on Google silicon.

GPT-5.5 Instant Becomes ChatGPT's New Default, Cutting Hallucinations by More Than Half

If you use ChatGPT without manually picking a model, you are now talking to GPT-5.5 Instant. OpenAI says the new model produces 52.5% fewer hallucinated claims than GPT-5.3 Instant on high-stakes prompts in medicine, law, and finance, and 37.3% fewer inaccurate claims on conversations users had previously flagged for factual errors. Responses are about 30% shorter with roughly 29% fewer lines, and the model deliberately uses fewer emojis and less filler formatting.

For developers, the model is exposed via the API as chat-latest, with GPT-5.3 still available for paid users for three months. The personalisation upgrade is also notable: ChatGPT can now reference past conversations, files, and Gmail to ground answers in your own context.

OpenAI Opens Its Self-Serve ChatGPT Ads Platform to Any US Business

OpenAI has flipped its ChatGPT ads programme from invite-only to a self-serve platform open to any business that wants in. This matters even if you never plan to buy ads: it confirms that ad-supported answers are now a core part of the free-tier experience. The platform adds third-party measurement and CPA bidding, putting ChatGPT's pricing discipline closer to Google Ads norms. Plus, Pro, Business, Enterprise, and Education subscribers are unaffected.

Gemini for Home Gets a 3.1 Upgrade, Enabling Compound Voice Commands

If you have early access to the Gemini for Home assistant, your speakers and displays have just received the Gemini 3.1 brain transplant. The practical improvement is compound voice commands: you can add new shopping list items and tweak existing ones in a single utterance, or chain a smart-home action, an alarm, and a calendar event together without three separate prompts. Camera UI improvements ride alongside the model update for Nest users.

A Gemini Agent for macOS Surfaces, Aimed at File Organisation and Computer Control

Code references uncovered in the Gemini macOS app point to a desktop agent that can drive your Mac and organise files, positioning Google directly against Claude Cowork and Perplexity's Personal Computer. This is still pre-release plumbing rather than a public launch, but the direction is unambiguous: every major chatbot is converging on the desktop as the next surface, with file system access as the first capability. Expect a clearer comparison set within weeks.

Microsoft Reframes Copilot as the Operating Layer for an Agent-First Workplace

Microsoft used the 2026 Work Trend Index publication to make a strategic pivot worth paying attention to even if you are not on Microsoft 365: agents, not chat, are now positioned as the primary interface to enterprise software. The company says 78% of knowledge workers now use AI agents at least weekly, up from 12% in 2024, and 67% of business leaders say agent-powered automation has reshaped at least one core process.

Copilot Chat is gaining a marketplace for agent plugins, with most agent logic authored in natural language. For builders, the practical takeaway is that Microsoft now expects line-of-business staff to package their own workflows as reusable agents.

Think 2026 Ships watsonx Orchestrate, IBM Bob, Confluent Integration, and IBM Sovereign Core

At Think 2026 in Boston, IBM unveiled the next generation of watsonx Orchestrate as the agentic control plane for enterprises that have ended up with agents from many vendors. The pitch is consistent policy enforcement and accountability across agents from any source, whether Claude, ChatGPT, or home-grown. It is in private preview today, with general availability expected later in the year.

IBM Bob is now generally available as an end-to-end software development agent covering generation through testing, security, and deployment. The differentiator versus Claude Code or Codex is explicit security and cost controls baked in from the start, which matters in regulated environments where unrestricted shell access is a non-starter. The Confluent and Tableflow integration with watsonx.data also went generally available, connecting real-time streaming with batch workloads, and IBM unveiled Sovereign Core for organisations that need operational independence from US-based hyperscalers.

ElevenLabs Names BlackRock, Nvidia, and Salesforce as Series D Backers; ARR Tops $500 Million

ElevenLabs has filled in the investor list behind its $500 million Series D, naming BlackRock, Wellington, D.E. Shaw, and Schroders alongside strategic backers Nvidia, Salesforce Ventures, Santander, KPN, and Deutsche Telekom. The company also disclosed it has crossed $500 million in ARR, up from roughly $350 million at the end of last year, and ran a second $100 million tender within six months.

For users and builders, the practical implication is that ElevenLabs has the runway and strategic relationships to remain the default voice and audio layer for the next phase of agent products. A roster of individual backers including Jamie Foxx, Eva Longoria, and Squid Game creator Hwang Dong-hyuk adds cultural reach to the strategic footprint.