Get tomorrow's brief in your inbox
Today: Zapier released AutomationBench, the first public benchmark measuring whether AI can complete real business workflows end-to-end. Amazon locked Anthropic into $100 billion of cloud spending over 10 years in exchange for up to $25 billion in new funding. Corporate America is flooding its communications with telltale AI writing patterns, and AI-generated phrases like "it's not just X, it's Y" have quadrupled since 2023.
Zapier built the benchmark enterprises actually need: Can AI get work done?
Most AI benchmarks test whether models can solve math problems or write code. Zapier's new AutomationBench measures something more important for businesses: whether an AI agent can execute a real workflow from start to finish without breaking anything. Link
The benchmark evaluates models across six business domains (sales, marketing, operations, support, finance, HR) using realistic environments with live data, messy formats, and multi-step tool chains. Scoring is deterministic: either the right CRM records were updated and the right emails were sent, or they weren't. No LLM-as-judge grading.
This matters because Zapier processes 2 billion AI tasks per month across 3.7 million companies, giving them uniquely deep visibility into what business automation actually looks like in practice. The full leaderboard and methodology are now public at zapier.com/benchmarks, and model providers can request private-set evaluation.
If you're evaluating AI tools for your business, this is the benchmark that tells you whether an AI agent can handle your workflows or just pass academic tests. It's free to access and covers the tools enterprises actually use.
Gemini expands to 7 new countries in Chrome sidebar
Google rolled out its Gemini-in-Chrome feature to Australia, Indonesia, Japan, Philippines, Singapore, South Korea, and Vietnam. The sidebar lets you ask questions across tabs, connect to Gmail and Google Photos for personalized answers, schedule meetings, check Maps, and draft emails without leaving your browser. Link
This is desktop and iOS (except Japan). The agentic feature that can control your browser to complete tasks on your behalf is still U.S.-only and requires AI Pro or AI Ultra paid plans. If you travel internationally or work with teams in Asia-Pacific, this expands where you can use Gemini without switching tabs.
NVIDIA ships 6 million synthetic Korean personas grounded in real demographics
NVIDIA's Nemotron-Personas-Korea dataset provides fully synthetic personas based on official Korean government statistics (KOSIS, Supreme Court, National Health Insurance). Zero PII, designed for Korea's Personal Information Protection Act (PIPA). Link
If you're building AI agents that serve Korean customers, this solves the identity-blind agent problem. Load a persona into your system prompt and your agent inherits region, occupation, communication norms, and domain expertise. Works with any agent framework (NemoClaw, NVIDIA NIM, or direct API calls). The tutorial shows how to filter personas by occupation and deploy a Korean public health agent in about 20 minutes.
Free to use. Part of the broader Nemotron-Personas Collection covering USA, Japan, India, Singapore, Brazil, and France.
Epic Games lets Fortnite creators build AI characters (no romance allowed)
Fortnite's new "conversations" tool lets developers create AI-powered NPCs that talk and interact with players using unscripted dialogue. You define the character with simple prompts (who they are, what they know, how they behave) and pick a voice. Link
The tool is experimental and you can't publish experiences with AI characters until it reaches beta (no timeline yet). Epic updated its rules: no medical/mental health guidance, no romantic partners or intimate companions, no circumventing safety systems. Powered by Google Gemini 3.1 Flash-Lite for processing and ElevenLabs for voice.
If you're building interactive experiences or experimenting with conversational AI, this is a sandbox to test character design without building infrastructure.
Hyatt deploys ChatGPT Enterprise globally for productivity and guest experiences
Hyatt rolled out ChatGPT Enterprise across its global workforce, using GPT-5.4 and Codex to improve operations, productivity, and guest experiences. Link
This is a signal that enterprise AI adoption is moving from pilot programs to full deployment in hospitality. If you run a service business, this is the pattern: start with internal operations (staff productivity), then extend to customer-facing workflows (booking, concierge, support).
Gemini vs. ChatGPT: Ecosystem fit matters more than model performance
ChatGPT and Gemini are now neck-and-neck on benchmarks and features. Both offer web search, deep research, canvas editing, voice mode, and one million token context windows. The choice comes down to ecosystem fit and a few specific features. Link
Gemini wins if you use Google's ecosystem (Gmail, Drive, Maps integration) and need video or music generation (Veo 3, Google Flow, Google Vids). ChatGPT has a broader third-party ecosystem and a head start on agentic AI. Both are priced similarly, but Gemini gives you more free tier usage (higher daily image generation quota).
If you're deciding which to standardize on for your team, test both for a week with your actual workflows. The technical differences are negligible; integration with your existing tools is what matters.
Amazon invests $25 billion more in Anthropic for $100 billion cloud commitment
Amazon agreed to invest up to $25 billion more in Anthropic (on top of $8 billion already invested), and Anthropic committed to spending over $100 billion on AWS over the next 10 years. Anthropic gets up to 5 GW of capacity for training and deploying Claude, including current and future Trainium chips. Link Link
The initial $5 billion investment is at Anthropic's latest valuation of $380 billion. This follows a similar deal Amazon struck with OpenAI two months ago ($50 billion investment, cloud infrastructure commitment). Anthropic says enterprise and consumer demand for Claude has created "inevitable strain" on its infrastructure, impacting reliability and performance.
For SMBs: Claude's growing enterprise adoption and infrastructure expansion means the API is getting more reliable and scalable. If you've hit rate limits or performance issues with Claude, this should improve availability in the coming months.
NSA reportedly using Anthropic's Mythos despite Pentagon feud
The National Security Agency is using Mythos Preview, Anthropic's unreleased cybersecurity model, primarily for scanning environments for vulnerabilities. This comes weeks after the Department of Defense labeled Anthropic a "supply-chain risk" because the company refused to allow Pentagon officials unrestricted access to its model's full capabilities. Link
Anthropic announced Mythos earlier this month as a frontier model designed for cybersecurity tasks, but claimed it was too capable of offensive cyberattacks to release publicly. Access is limited to around 40 organizations. The U.K.'s AI Security Institute also confirmed access.
For businesses: Mythos won't be publicly available, but this signals that AI-powered vulnerability scanning is getting good enough for intelligence agencies to rely on it. If you're building security tooling, this is the direction offensive and defensive cyber is heading.
Executives of AI automation company arrested in massive fraud scheme
Two executives from iLearningEngines, an AI automation company, were arrested in a fraud scheme. The details suggest fabricated customers and revenues. Link
This is a reminder to verify AI vendor claims before committing budget. Ask for customer references, case studies with verifiable metrics, and transparent pricing. If a vendor won't provide customer references or their case studies sound too good to be true, walk away.
Spot AI-generated corporate text with the "it's not just X, it's Y" test
Corporate communications have flooded with the sentence construction "It's not just X, it's Y," which has quadrupled from about 50 mentions in 2023 to over 200 uses in 2025, according to AlphaSense's database of corporate news releases, earnings reports, and government filings. Link
Examples from 2025: "AI won't just be a tool; it will be a collaborator" (Cisco), "The future of autonomy isn't just on the horizon; it's already unfolding" (Accenture), "DevOps teams are managing not just deployments, but also security compliance" (Workday).
If you're reviewing vendor proposals, partner communications, or even your own team's output, scan for this phrase and em-dashes. It's a tell for AI-generated text. Not a smoking gun, but a strong signal to read more critically.
The gap between academic AI benchmarks and real-world business execution is closing. Zapier's AutomationBench represents a shift from measuring what AI can theoretically do to whether it can get work done reliably. Amazon and Anthropic's $100 billion infrastructure deal signals that enterprise demand for AI is outpacing supply, which should improve API reliability and pricing competition. And the explosion of AI-generated corporate communications is a reminder to audit your own output: if your writing sounds like everyone else's, it probably came from the same place.