Anthropic Wins Office; Alexa Devours Rufus; Gamers Train AI

Today's AI Outlook: 🌤️

Anthropic Grabs The Corporate Card

Anthropic has overtaken OpenAI in paid business adoption, according to Ramp’s latest AI Index. Ramp tracks corporate card and invoice payments across 50,000-plus U.S. businesses, so this is a spending signal rather than a full market-share scoreboard. Still, the trend is loud: Anthropic rose to 34.4% adoption in April, while OpenAI slipped to 32.3%.

Why it matters

Claude is moving from developer darling to office workhorse. The Rundown AI points to Claude Code as a major driver, with Anthropic expanding beyond technical teams into finance, legal and research workflows. OpenAI still has the bigger consumer brand, but enterprise buyers are clearly willing to spread the budget around.

The Deets

• Overall AI use among Ramp-tracked businesses climbed to 50.6%.
• Anthropic’s usage has quadrupled over the past year.
• OpenAI’s business adoption has leveled off in Ramp’s data.
• Ramp flagged risks for Anthropic, including Claude outages and rising costs.

Key takeaway

The enterprise AI race is a paid adoption knife fight, and Claude just landed a very expensive jab.

🧩 Jargon Buster - Spend Signal: Data that shows what companies are paying for, which can reveal adoption trends without capturing the entire market.

Meta Employees Reject Being AI Trainers

Meta employees are protesting mouse-tracking software installed on work computers, according to AI Secret. Meta says the data helps train agents to use computers more like people do. Employees see a darker possibility: with layoffs looming, their clicks may be teaching the systems that eventually replace pieces of their jobs.

Why it matters

Meta’s AI ambition is not the issue. The problem is execution by surprise. Worker surveillance, weak communication and job cuts make AI transformation feel less like innovation and more like being asked to dig your own moat.

The Deets

• Employees spread flyers across U.S. offices protesting the tracking software.
• Meta says the data is for training AI agents.
• Staff concerns are rising alongside expected layoffs.
• The broader worry is trust: workers may resist AI systems they believe are being built from their own monitored behavior.

Key takeaway

AI labor automation needs trust to scale inside companies. Meta appears to be testing the opposite approach with a flamethrower and a clipboard.

🧩 Jargon Buster - Agent Training: The process of teaching AI systems how to complete tasks, often by learning from human behavior, examples or feedback.

Cisco Cuts Jobs While AI Infra Thrives

Cisco is cutting nearly 4,000 jobs, less than 5% of its workforce, while raising its 2026 revenue forecast. The reason is not collapsing demand. Quite the opposite: AI infrastructure orders are surging, and Cisco is shifting capital toward the parts of the business tied to data centers, networking and AI compute demand.

Why it matters

This is the awkward corporate math of the AI boom. Companies can grow richer from AI infrastructure while cutting workers tied to older business lines. The winners are chips, switches, optics, power and data centers. The losers may be legacy org charts.

The Deets

• Cisco has $5.3B in AI infrastructure orders this fiscal year.
• It now expects that number to reach $9B.
• Networking orders rose more than 50%.
• Data-center switching rose more than 40%.
• Cisco shares jumped 16% after hours, according to AI Secret.

Key takeaway

The AI buildout is creating growth, but it is also giving companies cover to prune anything that does not fit the new infrastructure stack.

🧩 Jargon Buster - AI Infrastructure: The hardware and networking systems needed to run AI, including chips, servers, switches, storage, power and data centers.

🛠️ Tools & Products

Alexa Eats Rufus

Amazon is folding its shopping chatbot Rufus into Alexa for Shopping, turning Amazon’s search bar into a more personalized shopping agent. The new Alexa can answer product questions, compare items, track prices, remember preferences and even buy items automatically when a price hits a target.

Why it matters

Amazon has the one thing every shopping agent wants: a mountain of purchase history, product data, reviews and delivery signals. That gives Alexa a serious edge in agentic commerce, provided customers trust it to do more than nudge them toward extra batteries and a suspiciously discounted waffle maker.

The Deets

• Rufus drew 300M-plus users in 2025 while still in beta, according to The Rundown AI.
• Alexa for Shopping uses catalog data, reviews, delivery timing, past purchases and Alexa conversations.
• It can run side-by-side comparisons and price tracking.
• Auto-Buy can purchase items when prices hit a target.
• Buy for Me can handle checkout on non-Amazon stores.
• Scheduled Actions can automatically restock products.

Key takeaway

Amazon is turning shopping search into an AI agent, and its biggest advantage is knowing what you bought before you remembered you needed it.

🧩 Jargon Buster - Agentic Shopping: Shopping handled by an AI assistant that can search, compare, track prices and complete purchases with limited human input.

Microsoft Finally Turns Edge Into A Browsing Co-Pilot

Microsoft added new Copilot features to Edge that can reason across open tabs, browsing history, mobile screen sharing, quizzes and AI-generated podcasts, according to AI Secret.

Why it matters

Browsers are becoming operating systems for AI work. The more context Copilot can see, the more useful it gets, but also the more sensitive the data layer becomes.

The Deets

• Copilot can reason across open tabs and browsing history.
• It can work with mobile screen sharing.
• New features include quizzes and AI-generated podcasts.
• The direction is clear: more context, more automation, more browser-native assistance.

Key takeaway

The browser is becoming the action layer for everyday AI, which means your tabs are no longer just tabs - they are raw material.

🧩 Jargon Buster - Action Layer: The part of an AI system that does not just generate text, but actually does things like browsing, buying, updating or organizing information.

💸 Funding & Startups

Origin Lab Wants Game Worlds To Train AI Worlds

Origin Lab raised $8M to help video game companies sell licensed game-world data to AI labs building world models, according to AI Secret.

Why it matters

Games are full of structured environments, movement, physics, objects and interactions. That makes them valuable training grounds for AI systems that need to understand how worlds behave.

The Deets

• Origin Lab raised $8M.
• The company helps game studios license game-world data.
• The buyers are AI labs building world models.
• The pitch gives gaming companies a new revenue stream from existing digital assets.

Key takeaway

The next AI training gold rush may include synthetic worlds, and game studios are sitting on a lot of map packs.

🧩 Jargon Buster - World Model: An AI system that learns how environments work so it can predict actions, movement and outcomes inside them.

🔬 Research & Models

AutoScientist Wants To Automate The Model Whisperers

Adaption, the startup from former Cohere VP of Research Sara Hooker, introduced AutoScientist, a system that customizes AI models for specific tasks by tuning both the training data and the learning process.

Why it matters

Fine-tuning powerful models is still expert work. If AutoScientist can automate more of that process, companies could build specialized AI systems without needing rare frontier-model talent on staff.

The Deets

• AutoScientist tests training data and settings, then iterates toward a user goal.
• In internal tests, it outperformed expert-tuned models by 35% on average.
• Success rates rose from 48% to 64%.
• Results held across multiple models, dataset sizes and eight industries, including finance, legal and medical domains.
• Adaption previously released Adaptive Data in February to improve dataset quality.

Key takeaway

Custom AI models may become much easier to build if the tuning process itself becomes automated.

🧩 Jargon Buster - Fine-Tuning: The process of adapting an AI model to perform better on a specific task, industry or dataset.

⚡ Quick Hits

Notion opened its developer platform, turning its workspace into a hub for custom agents, external agents, workflow code and live data sync.

Meta launched Incognito Chat for Meta AI, saying WhatsApp and Meta AI users can ask private questions without conversations being stored.

Nvidia hit a $5.5T market cap, becoming the first company to reach that level, according to The Rundown AI.

Microsoft introduced MDASH, an AI security harness that chains 100-plus specialized agents to hunt software bugs and caught 16 flaws across Windows.

The U.K.’s AI Safety Institute warned cyber capabilities are accelerating, saying AI’s ability to complete cyberattacks is doubling every few months.

🧰 Tools Of The Day

Alexa for Shopping: Amazon’s new shopping agent can answer questions, compare products, track prices and automatically buy items when they hit a target price.

Higgsfield Supercomputer: A cloud AI agent with tools, memory and creative capabilities. The Rundown AI also highlighted a workflow connecting Higgsfield CLI with Claude Code to test prompts across multiple image models.

Incognito Chat: Meta’s private chat option for Meta AI and WhatsApp users, designed for questions that are not stored..

Claude for Small Business: New Anthropic tools aimed at everyday business tasks such as payroll, invoices and campaigns.

Today’s Sources: AI Secret, The Rundown AI