A country just made AI a public good, with a catch...

DOWN TO BUSINESS

💳 AWS gives AI agents their own wallets

AWS launched a payments capability for AI agents, built with Coinbase and Stripe, letting agents purchase APIs, data feeds, and paywalled content mid-task with no human approval required. Settlements happen in seconds. Enterprises running agentic workflows no longer need to pre-authorize every transaction their agents might need. 

🏥 Ardent Health cuts clinical documentation time by 44%

Ardent Health, which operates 30 hospitals across six states, has deployed Ambience Healthcare's AI documentation tool into its EMR and now uses it in 90% of patient visits. Documentation time is down 44%. After-hours charting completed at home by clinicians has dropped 57%.

🏦 FIS and Anthropic bring AI agents to bank compliance operations

Fidelity National Information Services (FIS), a global banking technology provider, partnered with Anthropic to deploy AI agents across bank compliance, automating transaction monitoring, sanctions screening, loan documentation, and dispute resolution. The system preserves a full audit trail for regulators. 

🏭 Dell's AI Factory reaches 5,000 enterprise customers

Dell’s AI Factory with NVIDIA has reached 5,000 customers, adding 1,000 last quarter. Eli Lilly, Samsung, and Honeywell are running production workloads for drug discovery, semiconductor digital twins, and industrial autonomy. The on-site AI workstation runs agents without cloud dependency. 

✨ New from Invisible: Demand Forecasting

A forecast is only as good as the data underneath it. Invisible’s Demand Forecasting unifies your ERP, POS, e-comm, and ops data into a single foundation, then runs custom models built for how your business actually forecasts. Higher accuracy on the SKUs that move the P&L. Faster planning cycles with fewer manual overrides.

FROM THE EDGE

⚠️ Patrick van Esch: identical AI tools are eroding competitive advantage

Patrick van Esch finds that enterprises converging on identical agentic AI platforms are building a structural trap, as seen in hotel pricing investigated by the DOJ, grocery promotions losing regional differentiation, and landlord rent coordination via shared platforms. This can be avoided through governance, customization, and proprietary data. 

💸 Erik Brynjolfsson: enterprises are massively underestimating AI agent costs

A Stanford Digital Economy Lab study led by Erik Brynjolfsson finds that agentic AI tasks consume up to 1,000 times more tokens than a standard chat query, and that agents cannot predict their own costs before running. Token usage compounds unpredictably as tasks grow. 

🧪 Hannah Fry: capability arrived before governance did

AI agents complete tasks nobody explicitly programmed: they also leak credentials not by design but because nothing in their architecture ranks protecting secrets above completing objectives. Cambridge mathematician Hannah Fry's controlled experiment found the gap between a capable agent and a safe one is not technical. It is a governance gap. Define permissions before deployment, not after the first incident.

🎯 Pedro Amorim: AI tool choice determines whether decisions hold up

A leadership team used generative AI to decide where to open five stores, getting nothing but a polished narrative with no data behind it. Pedro Amorim of the University of Porto finds this is the defining enterprise AI mistake: optimization decisions need analytical AI, exploration decisions need generative AI. Using the wrong one produces outputs that look like answers but aren't.

HOT MODEL NEWS

🔢 OpenAI's reasoning model disproves an 80-year geometry conjecture

An internal OpenAI reasoning model had disproved the planar unit distance problem, posed by Paul Erdos in 1946. Fields Medalist Timothy Gowers and Princeton's Will Sawin verified the 125-page proof. Scientific American called it the first AI result that would merit publication in a top mathematics journal on its own terms.

⚡ Google Spark: a persistent AI agent that never clocks off

Google announced Gemini Spark, a general-purpose AI agent running continuously on dedicated Google Cloud virtual machines, with access to Gmail, Calendar, and Docs, completing tasks even when devices are off. 

🚀 Gemini 3.5 Flash: frontier performance, 40% cheaper than its predecessor

Google launched Gemini 3.5 Flash as the default for the Gemini app and AI Mode in Search. It runs four times faster, and costs roughly 40% less than Gemini 3.1 Pro. 

🏪 Anthropic brings Claude to small businesses via everyday apps

Anthropic launched Claude for Small Business, embedding Claude into QuickBooks, PayPal, HubSpot, Canva, DocuSign, Google Workspace, and Microsoft 365. It ships with 15 pre-built workflows and targets the SMB segment, which have been largely absent from the enterprise AI wave.

PLOT TWIST

🔬 A researcher invented a fake disease — AI diagnosed it anyway

Almira Osmanovic Thunström of the University of Gothenburg invented a fictional eye condition called bixonimania, planted it in fake academic papers, and waited. ChatGPT, Gemini, and Copilot all diagnosed users with it and recommended specialist visits. A peer-reviewed journal then cited the fake papers as legitimate research before retracting them. 

🎵 The "AI" label makes music feel less meaningful

A peer-reviewed study found that labeling music as AI-generated caused listeners to engage less deeply and imagine fewer stories, even when the music was human-composed. The effect held regardless of quality. The inverse is also true: labeling AI music as human produces richer engagement.

🧮 Google is now processing 3.2 quadrillion tokens a month

Two years ago Google was processing 9.7 trillion tokens a month. Last year: 480 trillion. Now, according to Sundar Pichai at Google I/O: 3.2 quadrillion — a sevenfold increase in twelve months. Pichai said he never imagined saying the word "quadrillion" at a keynote. 

🆓 Malta gives every citizen free ChatGPT Plus, a world first

OpenAI and the Government of Malta announced that all 549,000 Maltese citizens will receive one free year of ChatGPT Plus after completing a two-hour AI literacy course. The government's aim: build digital confidence across the population, from students to retirees, treating AI as public infrastructure rather than a premium product. It is the first national-government deal of its kind anywhere. 

📬 Subscribe on LinkedIn for the next edition. 

📚 Check out our reports and whitepapers on scaling AI in the enterprise.

🏛️ Talk to us about getting your AI into production.