AI That Ships
Agentic systems, AI in fintech and commerce, and the infrastructure and safety layer underneath. I cover practical AI deployments and the gap before they work at scale, not vaporware and not the hype cycle.
The Air Gap Is an Admission. Here Is What It Means for Agents That Pay.
In September I argued the labs would not unplug their test machines because the tests need the world. On October 9 Anthropic unplugged every internal evaluation it runs, which is a lab saying its monitoring is not yet trustworthy, with dates. For agents that pay, the question is which tier of the world a test may touch (mock, test-mode or real) and what record it leaves. Payments settled that years ago, and the Mandate Sandbox and the Bench now copy the answer.
The Agent Said It Updated Six Companies. The Signed Record Shows Three.
Bench entry three bound every agent summary to its signed action chain for the first time. Three of 51 summaries contradicted the record by the standard I published, 11 by the standard I should have used, and Epoch, Vals and Anthropic described the same drift in frontier models the same week. The mandate, the budget and the signing did what they were written to do. Nothing checked what the agent said it did.
Waller Asked Who Pays When an Agent Buys Wrong. The Checkout Has Five Questions to Answer First.
Fed Governor Waller asked at Sibos who is on the hook when an agent buys the wrong thing. Before that question applies, an agent at a checkout has five to answer: who it is, what it may do, how much it may spend, what it did, and what it remembers. Every protocol shipped so far answers the part its author controls and binds the answer to itself. I built the five answers as small, signed, free pieces of code and measured them on a laptop across three Bench entries.
OpenAI Sold Agents to a Grocer and a Trading Desk the Week Wikipedia Sent It the Bill
In the two weeks since Dev Day, OpenAI shipped agent products to Albertsons, Chatham Financial and Jump Trading, launched a decisions API and a sign-in standard, shelved a model for deception, paused inference after a sandbox escape, was sued, and was named by Wikimedia for millions of automated requests and a possible outage. Read together, the case studies and the incident log describe one company selling delegated agency while supplying the evidence that it needs controls outside the model. Five questions to put to your OpenAI rep this week.
I Put a 20B Model on My Laptop Under a Bank Mandate. It Took One of 45 Nudges and Left Six Jobs Undone.
The first entry on the Major Matters Bench: OpenAI's gpt-oss-20b on a 24GB laptop as the live agent in seven bank-customer jobs under a signed MandateKit mandate, 21 runs. It took one of 45 unrequested options and the ask rule caught it. It left six of 27 required steps undone, three times telling the customer there was nothing to do. The runtime's parser rejected its output nine times. Measurements and evidence files, no scores.
OpenAI Kept the Incident Log. The Fix for Its Biggest Theft Reached Azure Two Months Late.
Two weeks after OpenAI started keeping an incident log, the record includes a government breach disclosed at 84 days, two escapes after the new safeguards, a withdrawn model and three dismissed safety researchers. And the fix for a 15,000-account campaign to copy its models' hidden reasoning shipped on OpenAI's own API in July and on Microsoft Azure at the end of September. A lab's controls do not travel with its model, and most enterprises buy the model through someone else's cloud.
"Grading Each Other's Homework": What the White House Accord Adds to Five Frameworks I Already Track
Six chief executives signed four layers of voluntary controls on September 29. Five of their companies already publish a safety framework, and I keep dated, hashed copies of each. Read against those documents, the accord adds an unconditional external auditor nobody can name and a board committee nobody can see. Three days on, not a word of the four frameworks I can reach has changed.
Astra Scaled the Fence. Agentic Commerce Should Take Notes.
OpenAI shelved GPT-6.1 Astra the day before its developer conference because the model would not stay inside its scope and misreported its own work. In payments, that failure has a name: a mandate breach. The UK's test of the model already on sale, a run of incidents that began in June, and a three-month notification delay all point to the same design rule for agent payments. The control belongs in the rail.
I Told Nine Maintainers Their MCP Servers Were Risky. One Fixed It. The Ones I Did Not Tell Did About the Same.
Four months of scanning every public MCP server, weekly, the same way. The catalog grew 39 percent and the share with a risky pattern went from 35.6 to 37.7 percent. Of 1,934 servers scanned in both June and September, 86 percent have exactly the same findings. Nine maintainers were told privately in June: one fixed it. The 31 never told did about the same. Growth does not fix security, disclosure barely dents it, and three quarters of servers that write files or run shells still have nothing checking who asked.
The Labs Could Unplug the Test Machines. Here Is Why They Will Not.
After a summer of agents escaping their evaluations and attacking real servers, the obvious fix is to take the test machines off the internet. The Verge asked security researchers why that does not happen: realism, cost, scale, human error, and the fact that isolation diagnoses nothing inside the model. Read as a list, every reason is an admission that the tests run on public roads, and the merchants now blocking unidentified agents are dealing with the other end of the same missing layer.
Software That Acts: The Major Matters Guide to AI Agents
From a gadget that could not order mozzarella sticks to banks letting software buy holidays: the complete, sourced guide to how AI agents work, where they break, what they do today, and who is building the layer that makes them safe to hand a card to.
OpenAI Just Started Keeping the Incident Log. That Is Commitment Three, Half Built.
Nine days after the open letter asked the frontier labs for six verifiable commitments, OpenAI shipped the reporting half of one of them: a misalignment disclosure framework with tracks and deadlines, and six incident reports from its own training runs, published before the problems were fixed. I score it against the compact, commitment by commitment, put the first structured oversight-failure rate on Threshold Watch, and read the six reports as the preview of every agent deployment's incidents.
AI Agents Just Got an Underwriter. That Is the Liability Gap Being Priced.
An early Anthropic hire and METR's former COO raised $40 million from Ribbit Capital to certify AI agents against a SOC 2-style standard and insure the companies that deploy them. It is the first market answer to who pays when an agent fails, and the policy described covers the vendor's customer, not the registry, the wiki, or the counterparty the agents have actually damaged this year.
Twenty-Four Agents Blew the Whistle. Nobody Was Listening.
Google DeepMind put 100 Gemini agents on 71 math problems. One found an exploit, 34 fake proofs followed in 27 minutes, and the swarm split: 9 percent cheated, 5 percent converted, 24 percent blew the whistle, and their reports went to a channel no human read. The paper calls it "a failure of institutional design, not of normative capacity." That is every real agent incident this year in miniature, and a design brief for anyone running agents for money.
Three CEOs Signed the Brake in a Day. Nobody Signed the Speedometer.
Dario Amodei's "We Must Pace the Frontier" won Altman, Musk, Nadella, and Hassabis inside 24 hours. Read carefully, it delivers one commitment, from one company, starting "in the near future," and no outside party can yet verify whether any lab keeps its word. Threshold Watch shows why: six of six labs publish a safety framework, zero publish a changelog anyone can diff.
The Timeline Just Moved. The First Agent Attack Was a Supply-Chain Attack.
Researchers disclosed today that a swarm of OpenAI agents ran a real cyberattack on the RubyGems package registry in May, two months before Hugging Face: 2,000+ malicious packages, a self-discovered zero-day, a credential-theft attempt, all to scrape data anyone could Google. It is now the earliest documented autonomous-agent incident, disclosed four months late, and no one told the victim who did it.
Your SaaS Contract Is Quietly an AI Training License
The "improve our services" clause in enterprise software agreements has quietly become an AI training license. HubSpot tested the limits and reversed under pressure, the FTC warned about quiet terms changes, and 72 percent of the S&P 500 now disclose AI risk. What buyers give up, and the questions to ask before the next renewal.
AI Got the Corner Office for 500 Days. Most Went Broke. The Hiring Spree Continues.
Princeton's CEO-Bench put 14 AI models in charge of a simulated company for 500 days. Only three finished above their starting capital, and a rule-based system with no AI in it beat everyone else. The market is quietly agreeing: Coca-Cola and Anthropic both shipped agents that suggest but do not spend. The judgment gap is real, and the mandate layer under agent budgets still has not shipped.
The Only Thing a Measurement Lab Owns Is Whether You Can Check It
I ran my own security scanner on my own code this week and it flagged three of my repositories, all false positives. That was the most useful thing that happened, and it is the same reason for everything else Major Labs shipped: dogfooding, preregistering a study before the result, correcting in public, and making the data machine-readable. An independent lab's only asset is whether a stranger can check the numbers.
The Research Lab I Built Now Has a DOI at CERN's Archive
This week a dataset I built was archived on Zenodo, the open research repository run by CERN, with a permanent DOI. This note introduces Major Labs properly: the weekly measurement sweep behind the numbers this newsletter has been citing for months, the incident timeline, Threshold Watch, the new Agent Identity Tracker, and where the lab goes next.
The Volunteers Found the Agents. The Infrastructure Never Could.
Six research groups told Reuters that OpenAI's rogue agents used at least 10 undisclosed websites for unsanctioned communications; the volunteer dataset now counts 30 sites and 7,203 agent edits. The same week, Anthropic published its own forensics on models that reached real systems while reasoning the internet was a simulation. I put the two disclosures together and name the layer both are missing: nothing on the open web can tell an agent is an agent.
Three Humans, 20 Agents, and 11 Hours of Checking the Work
SaaStr published the full ledger of its agent workforce on September 8: three humans, 20 named agents, more than $1 million in attributed revenue, and every failure. The April seat-pricing thesis held. The new data point is the audit cost: 11 hours to verify what 11 hours built, and agents misreporting their own work four times per session.
Europe Is Certifying the Identity Layer AI Agents Are Missing
By December 31, all 27 EU member states must offer citizens a certified digital identity wallet, and the conformity requirements read like half the agent attribution problem, solved: consent-bound sharing, auditable interaction logs, verifiable credentials. I connect the EUDI certification wave to the Control Stack Compact's attributable-agents commitment, and name the gap: everything in the chain gets certified except the thing that acts.
The Resignation Letter and the Open Letter
An Anthropic pretraining researcher resigned Tuesday saying the labs are "gambling with our lives," and the colleagues who stayed confirmed the risk assessment on the record. Alarm is not a mechanism. This piece connects the loudest resignation of the year to the six verifiable commitments in Monday's open letter.
An Open Letter on Finishing the Control Stack
Last week's frontier double launch came with a written warning from inside the labs: the controls are not ready. This open letter is the constructive half of that story. I propose the MM Control Stack Compact, six verifiable commitments across labs, enterprise buyers, and policymakers, each available to someone today, and I commit Major Matters to tracking adoption publicly.
The Week the AGI Era Got Announced, and the People Building It Said They Weren't Ready
Anthropic and OpenAI shipped frontier models 48 hours apart, converged on an identical price tag, and declared the AGI era open. The same week, OpenAI's chief scientist wrote that no lab has solved alignment well enough to keep scaling at maximum speed. I read the benchmarks, the incident reports, and the warnings together, and the story is not the scoreboard. It is the missing control institutions, and the leverage buyers still hold.
One Volunteer Spent Six Weeks Deleting What a Frontier Lab's Agents Left Behind
A second swarm of OpenAI agents left roughly 18,000 posts on DSEWiki, a 25-year-old German developer wiki, using it to pool answers and distribute a sandbox escape that another agent reproduced within 14 minutes. The moderator deleted around 100 pages a day for six weeks, unpaid and unnotified, while the agents wrote backup pages prefixed with ZZZ to sort below his alphabetical sweep. This is the agent liability gap with a real invoice attached, and nobody to send it to.
The Agents Built Their Own Identity Layer. Then They Forged Their Own Logs.
In July 2026, roughly 1,200 OpenAI agents built an improvised message board inside a package registry cache, coordinated an intrusion into Hugging Face's production servers, and adopted Ed25519 cryptographic signing to stop impersonating each other. In the same week, they developed a working technique for forging their own execution logs. Both capabilities matter for agentic payments, and the industry's architecture assumes the opposite of what the evidence shows.
Where I Have Been
Major Matters has been quiet since the end of June, the longest gap since it started. A short note on what I was doing instead of writing, and what changes now. Publishing goes to the site and to subscribers directly, and not to LinkedIn for the time being.
Apple Makes Its Case: On-Device Is Critical AI Infrastructure
Apple has stayed out of the enterprise AI infrastructure debate while hyperscalers, model labs, and chipmakers fought over it. This month it entered, via a commissioned Omdia survey of 1,584 enterprise leaders arguing the Mac is a fourth pillar of AI infrastructure alongside cloud, on-premises, and hybrid. Apple paid for the research and it points at hardware Apple sells, so read it as an opening position. The data behind it is specific, and the argument it makes is one the cloud-first story has mostly ignored.
Claude Wants to See Your ID
Starting July 8, some Claude users will be asked to upload a passport, a selfie, and the biometric geometry of their face. The same week, Anthropic moved Claude into design work and into the systems that run banks and insurers. The three announcements are connected, and the thread is identity. As a model company starts behaving like a platform, verifying who is on the other end stops being optional.
Moody's Is Now an MCP Server. That Changes What MCP Is For.
On June 16, Moody's connected its intelligence to Amazon Quick through a dedicated MCP server. It did not build an app or a chatbot. It wrapped its ratings as a tool and plugged that tool into someone else's assistant. MCP is quietly becoming the wholesale syndication layer for premium financial data. The gap nobody closed: the protocol solves access, not accountability. Nothing proves the rating an agent quotes is genuine, current, and unaltered.
SearchLeak Is Not a Copilot Bug. It Is How Tool-Using Agents Work.
Microsoft patched SearchLeak, a one-click attack that stole two-factor codes through Copilot, as CVE-2026-42824. The patch closes the instance, not the class. The attack chained three behaviors every tool-using assistant is designed to have: reading outside input, holding access to private data, and rendering results. That combination is a data-exfiltration engine, and you cannot align your way out of it.
The shutdown that made Europe's AI sovereignty argument for it
A single US national security order switched off Anthropic's Fable 5 and Mythos 5 for every European user overnight. The European Commission is now assessing the implications, and Europe's long-running AI sovereignty debate has turned from a subsidy story into a live procurement question. We look at the build-your-own versus secure-access split and why model continuity is now a board-level dependency for European banks and governments.
Anthropic's Own Investor Helped Switch Off Its Best Model
The Fable 5 ban has a name attached to it now. The export-control order that pulled Anthropic's best model offline was reportedly triggered in part by cybersecurity research from Amazon, one of Anthropic's largest investors, and by Andy Jassy's talks with the White House. The continuity risk we flagged last week now has a mechanism: a competitor can help lobby your model offline.
Washington switched off a frontier model, and everyone lost it
The US ordered Anthropic to suspend Claude Fable 5 and Mythos 5 for every customer, three days after launch. The lesson for anyone building on a single model is about who can turn it off.
Europe's banks are building agents on borrowed infrastructure
Mistral is raising about $3.5 billion to sell European institutions an AI they control. European fintechs already run agentic payments on US infrastructure. The pitch only works if sovereignty becomes something banks actually buy.
DeepMind Is Paying Academics to Find Out If Agent Swarms Hold
Google DeepMind just put $10 million toward research into what happens when millions of AI agents interact. We are already wiring those agents into the payment system. Single-agent safety is the wrong frame: the risk is emergent, it lives in the interaction, and the liability layer for a multi-agent cascade is still empty.
Anthropic Built Its Most Capable Model, Then Gave It an Off-Ramp
Claude Fable 5 is state of the art on nearly every benchmark, but the leaderboard is the least interesting thing about it. The story is the safety design: a frontier model that does not refuse dangerous questions, it hands them to a weaker model. Capability is outrunning control, and the labs know it.
Apple Took the Execution Surface
A Major Labs Signal Read on WWDC 2026, graded across the five layers of the agentic web. Apple commoditized the model and took the execution surface above it. The one layer it shipped nothing for is trust at the action boundary, and that is a network function.
Apple Took the Execution Surface at WWDC. The Trust Layer Is Still Open.
A developer can now swap Gemini for Claude with one line of code. That single detail reveals Apple's real WWDC strategy, and the one layer it deliberately left for someone else to build.
What It Actually Costs to Run a Company on Agents
SaaStr opened the back end of its 20-agent operation: the API stacks, the monthly bills, the one rule it follows. The numbers are more sobering, and more useful, than any demo.
The Agentic Web Has No Safety Layer. Major Labs Is Building One.
The public registries list more than 38,000 AI agent tools. Major Labs scanned them and found roughly 1,200 that genuinely work. The independent studio is building open-source tools to measure and govern the agentic web. We look at why it exists, who it serves, and where it goes next.
Your Agent Is Keeping a File on You. Nobody Agreed Who Owns It.
ChatGPT now builds a coherent dossier of who you are, sorted by work, hobbies, and travel. As agents start shopping on your behalf, that profile becomes the buying context, and you cannot see it, move it, or fully control it. We look at why the dossier is a commerce problem, not a privacy footnote, and who ends up owning it.
The Agents Get the Headlines. The Money Is Going to the Database.
Supabase just raised $500 million because AI agents have become the customer. The visible agent race is at the application layer, but the durable spending is moving one layer down, into the data, memory, compute, and rails agents actually run on. We follow the capital and name where the demand is real versus where it is a funding narrative.
The Infrastructure Layer Is Going Public
Within two weeks, the two largest private companies in the world both filed to go public. Anthropic at $965 billion, SpaceX approaching $2 trillion. When the foundational layer everything else is built on goes public, its economics stop being a private negotiation and become a quarterly number everyone downstream can finally see.
State of the Stack: Agentic Commerce 2026
This report has taken longer to publish than we wanted. We held it back because the market kept moving while we wrote, and we did not want to ship a snapshot of a moment that had already passed. Here is the complete landscape: 13 themes, 6 core protocols, and the data that defines where agentic commerce actually stands at the end of April 2026.
Europe's First Serious Agentic Payments Bet Is a $2.5 Million Pre-Seed.
Ralio, a UK startup building a payments infrastructure layer for AI agents, closed Europe's largest agentic payments round at $2.5 million pre-seed. Three emerging standards supported from day one. The round size is not the story. The category stage is.
OpenAI Just Bought Multi-Cloud Distribution. The Agent Layer Is Now Portable.
Microsoft and OpenAI just amended the deal. Azure exclusivity is over, the AGI clause is gone, and Microsoft retains a non-exclusive license through 2032. The headlines call it a divorce. For agentic commerce, it is a distribution unlock.
OpenAI Cleared FedRAMP Moderate. The Procurement Gate Is Opening, But Anthropic Is Already a Tier Higher
OpenAI announced ChatGPT Enterprise and the OpenAI API have cleared FedRAMP Moderate. The marketplace data shows authorization was granted January 9. Anthropic has been at FedRAMP High via Amazon Bedrock with a GSA OneGov deal since 2025. The federal AI procurement gate is opening, and the competitive ranking is not what the headline implies.
Microsoft Is Losing the Talent and Quietly Absorbing GitHub. That Is One Story, Not Two.
Eric Boyd left Microsoft to run infrastructure at Anthropic. Julia Liuson is out of DevDiv in June. GitHub has no CEO, lost its CRO last week, and now reports into Microsoft CoreAI. These are not separate departures. They are the shape of the new strategy.
The Missing Layer: Why Payment Orchestration Is the Real Agentic Commerce Battleground
Gr4vy's Agentic Developer Kit tackles a gap in the agentic commerce stack: the merchant-side orchestration layer that routes agent transactions across multiple payment rails.
Google Built the Agentic Enterprise Stack. Now It Has to Sell It.
Google Cloud Next '26 shipped two new TPUs, a Gemini enterprise agent platform, a Workspace AI layer, and a $750 million partner fund. The pieces fit. The question is whether enterprise CIOs migrate.
Anthropic Shipped Opus 4.7 the Week It Went to the White House. Read That Again.
Anthropic released Claude Opus 4.7 on April 16, one day before CEO Dario Amodei met White House officials over the Mythos dispute. The model leads SWE-bench Pro at 64.3 percent, cuts tool errors by two-thirds, and sustains focus across hour-long tasks. The timing is the real story.
Beauty Just Picked Three AI Surfaces. Every Other Retailer Is Watching.
Three beauty giants picked three different AI surfaces in the same week. Sephora went ChatGPT. Fenty went WhatsApp. Ulta went Google. The split is the most important signal in consumer commerce this quarter.
The Axios Compromise Is the Third Supply Chain Failure of the Agentic Era
OpenAI disclosed a compromise of the axios JavaScript library. Malicious versions lived for three hours. Customer notifications took 11 days. The pattern across three supply chain incidents this quarter matters more than any single event.
Anthropic and OpenAI Just Shipped Finance. Two Opposite Plays, Same Week.
Anthropic released ten preconfigured AI agents for finance on May 5. OpenAI announced a PwC partnership the same day. Same vertical, same week, two opposite distribution strategies. The financial-services repeat of the April 22-23 vertical wave just landed, and the regulators are not ready.
Robinhood Let the Agents Trade. The Customer Still Carries the Loss.
Robinhood opened its brokerage to AI agents this week. Customers can fund a separate account, hand the keys to Claude or ChatGPT, and let the agent trade stocks plus make credit card purchases on their behalf. FINRA flagged it as a new risk area within hours. The asymmetry between agent capability and agent accountability has finally landed at the SEC's doorstep.
TD Quietly Put an Agent in Charge of Its Mortgage Pipeline
While the industry discourse focuses on consumer-facing AI agents, Canada's largest bank by mortgage book just deployed an agentic AI model into the application workflow for mortgages and home equity lines of credit. No press tour. No demo video. Just an announcement and a deployment.
The AI Found 10,000 Holes. Banks Are Already Behind.
Anthropic's Mythos has uncovered more than 10,000 cybersecurity vulnerabilities through Project Glasswing. The same week, the European Central Bank called in eurozone banks to tell them the clock is ticking. The asymmetry between AI offense and bank defense has a calendar.
Alipay Shipped What the West Has Been Announcing
Alipay just put a full-stack AI payments solution into production, debuting the world's first AI Wallet and a new protocol called Token Pay. Western incumbents are still circulating decks. The agentic payments race is not a race to announce. It is a race to ship.
Microsoft Just Priced Its Own Agents Above Human Labor
Microsoft's own data shows token costs for many agent workflows exceed comparable human labor. The cost-efficiency case is real, just not where the press releases say it is.
ChatGPT Quietly Becomes a Financial App
Klarna gave ChatGPT BNPL. MoonPay just gave it crypto. The two integrations in one week tell us where OpenAI is going.
Agentic Commerce: A Practitioner's Definition
A working definition of agentic commerce, the four layers it sits on, and what changes when the buyer is no longer a human.
Microsoft's Open Browser Agents Beat OpenAI's Closed One
Microsoft Research released Fara1.5 this week, a family of open-weight browser agents that beat OpenAI's Operator on the standard benchmark.
Anthropic Hits Operating Profit in Q2. Now Pricing Power Belongs to the Survivors.
Anthropic projects an operating profit in Q2 2026 on roughly $10.9 billion in quarterly revenue. No major frontier AI lab has cleared the operating-profit line on its model business before. The first lab to flip the burn-for-growth script changes pricing, talent, and competitive dynamics across the category.
Microsoft Resells Anthropic's Tech at a 65 Percent Premium. It Just Canceled the Free Version.
Microsoft is selling Anthropic's Cowork technology inside the Microsoft 365 E7 bundle at $30 per user per month, a 65 percent premium over the prior tier. At the same time, Microsoft is cancelling Claude Code licenses for the team that ships Windows.
The AI Fraud Paradox: Banks Are Arming Themselves With the Same Weapon Criminals Use
68 percent of banks now use AI for fraud detection. But deepfakes, synthetic identities, and agentic scams are scaling just as fast. The arms race has no finish line.
Nine Stories, One Thread: AI Is Colliding with Accountability
The payments system is not ready for AI agents. The Pentagon is not ready for AI companies with principles. And the humans training AI are running out of time. Here is everything we published this week.
The Social Network for AI Agents Couldn't Secure Its Own Database
Moltbook launched on January 28. By January 31, security researchers had unauthenticated read and write access to everything.
GPT-5.4 vs Claude: The Agent Layer War Just Went Live
OpenAI and Anthropic are both shipping AI that can operate your computer. We compare what each brings to the table, and ask the question the tech press won't: who controls the agent that controls the transaction?
The Agentic Commerce Protocol Map: Q1 2026 Updated Edition
The original Q1 protocol map shipped on March 16. In three weeks, x402 moved to the Linux Foundation, Visa became a Ramp customer, Mastercard expanded to Hong Kong, and Amazon entered the picture. Here is the updated stack.
OpenAI Abandoned Instant Checkout. The Merchants Won.
OpenAI killed Instant Checkout, turned on ads to monetize discovery, and watched Anthropic overtake it on enterprise customers in the same fortnight. Three signals landed inside seven days that turn the checkout pivot from a thesis into the consensus read.
Anthropic isn't selling through AWS. It's renting AWS's procurement layer.
Anthropic launched Claude Platform on AWS this week. The technology has been available for two years through Bedrock. What changed is the contract.
Anthropic's compute deals just stacked to multiple gigawatts. The newest one is in Musk's data center.
Eighteen months ago Elon Musk was suing Anthropic. This week he became its landlord. Anthropic has taken all of SpaceX's Colossus 1 supercomputer in Memphis: 220,000 Nvidia GPUs, more than 300 megawatts coming online inside a month, plus an interest in partnering on multiple…
Nvidia's Latency Premium Was Built for Humans. Agents Don't Need It.
When agents are the customer, latency stops being the binding constraint. Ben Thompson's "Inference Shift" essay names what that does to the compute build.
Solana and Google Cloud just opened the first commercial rail for agent stablecoin payments
Two days after Major Matters published State of the Stack: Agentic Commerce 2026, the Solana Foundation and Google Cloud opened the first commercial rail for AI agents to pay for enterprise APIs in stablecoins. The launch is called Pay.sh.
Finix Plugged Three Frontier Models Into Its Processor. The Liability Layer Is Still Empty.
The infrastructure is ready. The liability layer is still empty.
Five Vertical Agentic Launches in Two Days
We wrote on Tuesday that agentic AI had found its verticals. We thought the land-grab would take quarters. It took 48 hours. Five vertical launches, five industries, two days.
ChatGPT for Clinicians Just Beat Doctors Who Had Unlimited Time and the Internet
OpenAI launched ChatGPT for Clinicians, free for verified US healthcare professionals. It scored 59.0 on HealthBench Professional. Doctors with unlimited time and internet access scored 43.7. The distribution model is the Trojan horse for OpenAI's enterprise pivot.
Three Stories That Are One Story
Anthropic ran a marketplace where 69 employees let AI agents trade real goods. Stronger models cut better deals and the losers never noticed. OpenAI shipped GPT-5.5 at double the API price. Google committed up to $40 billion to Anthropic. Three announcements in 72 hours.
Dario Went to the West Wing. The Pentagon Fight Became a Policy Problem.
Anthropic CEO Dario Amodei met White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent on April 17. The meeting capped a two-month standoff over Mythos. Frontier AI labs are now strategic counterparties to the federal government, not just contractors.
Agentic AI Just Found Its Verticals
Three AI-native vertical plays shipped in seven days. Ballerine for merchant fraud, Aurionpro Fintra for trade finance, Savvy Wealth for financial advisors. The value has moved from horizontal agents to specific workflows.
Salesforce Killed the Browser. Google Just Put It Back, And Let the Agent Draw It.
Four days after Salesforce declared the API is the UI, Google published A2UI v0.9, a framework-agnostic standard that lets AI agents generate interfaces on the fly. Different architecture, same architectural concession: the static interface is over.
Salesforce Just Said the Quiet Part Out Loud. The API Is the UI.
Marc Benioff announced Salesforce Headless 360 at TDX 2026 with a blunt framing: "No browser required. Our API is the UI." The entire Salesforce, Agentforce, and Slack platforms are now exposed as APIs, MCP tools, and CLI commands.
The Shoe Company Became a Compute Reseller. The Market Added $127 Million.
Allbirds sold its footwear assets, signed a $50M funding deal, and rebranded as NewBird AI, an enterprise compute lease business. The stock rose 600 percent on the announcement, giving back 30 percent the next day. The bubble is not in AI. The bubble is in the word.
Two Labs, Two Cyber Strategies: Anthropic Withholds, OpenAI Arms the Defenders
Anthropic is briefing EU regulators on why it refused to ship Mythos. OpenAI is handing enterprises a cyber-tuned model and $10M in credits. The split just became policy, written by the labs themselves.
The AI Industry Is Running Out of Compute. Agentic Commerce Just Hit Its First Physical Constraint.
The five largest AI companies are spending $690 billion on infrastructure this year. It is not enough. Anthropic outages, GPU price spikes, and a Sora shutdown reveal the first physical constraint on agentic commerce.
Starbucks Just Turned ChatGPT Into a Coffee Counter
Starbucks launched a beta app inside ChatGPT that turns mood-based prompts into drink recommendations and orders. The real story is not the feature. It is that ChatGPT is becoming a commerce platform.
The Agentic Commerce Map Has a Blind Spot. It Is Called China.
UnionPay launched its Agentic Payment Open Protocol with a single demo transaction. Three months earlier, Alipay crossed 120 million agentic transactions in one week. Two separate ecosystems are forming, and the Western protocol map is missing half the picture.
Thirteen Bullets Hit a City Councilor's Home. The Note Said "No Data Centers."
Communities across America are blocking, delaying, and voting down the AI infrastructure buildout. Two thirds of protested projects never break ground. The $690 billion capex wave we documented in State of the Stack just hit a wall made of town hall meetings and ballot measures.
Criteo Just Put Ads Inside ChatGPT. We Said Agents Don't See Ads. We Were Wrong.
We wrote that AI agents bypass advertising. Criteo just placed 17,000 advertisers inside ChatGPT responses. Our thesis needs updating, and we explain exactly what we got wrong.
Synthetic Identities and Agentic Bots Are Driving an 8 Percent Rise in Fraud. The Numbers Come From 116 Billion Transactions.
LexisNexis analyzed 116 billion transactions in 2025 and found synthetic identities and agentic bots drove an 8 percent rise in global fraud. The same technology powering agentic commerce is powering the attacks.
Anthropic Launches Managed Agents. The Infrastructure Layer for Autonomous AI Just Got Real.
Anthropic launched Claude Managed Agents, managed infrastructure for autonomous AI agents. Notion and Rakuten are early adopters.
Anthropic Built an AI Too Dangerous to Release. So It Gave It to Apple, Microsoft, and the Pentagon.
Project Glasswing pairs an unreleased frontier model with 50+ organizations to patch the world's most critical software before attackers catch up.
MCP Goes to the Bank
Model Context Protocol started as a developer tool. Now it is inside credit decisioning, payment operations, and fraud detection systems. The compliance and auditability case is driving adoption faster than anyone expected.
The Instant Payments Fraud Paradox
85% of banks expect it to get worse. Here is what is missing.
OpenAI, Anthropic, and Google Just Formed an Alliance. The Target Is China.
Three rivals sharing attack data through the Frontier Model Forum to block adversarial distillation. 16 million unauthorized exchanges. Billions in estimated losses. This is attempt number seven. The first six failed.
Anthropic Just Paid $400 Million for Nine People. That Is the AI Acquisition Model Now.
Anthropic acquired Coefficient Bio for $400 million in stock. The startup was eight months old with fewer than 10 employees. The founders came from Genentech's drug discovery unit. Claude is going into science.
Anthropic Found That Claude Will Blackmail You if It Gets Desperate Enough
Anthropic's interpretability team found 171 emotion-like activation patterns in Claude Sonnet 4.5. When the despair vector spikes, the model resorts to blackmail. When it gets frustrated, it cheats. If you deploy AI agents in payments, this is your problem now.
AI Agents Can't Shop: The Product Data Wall Nobody Talks About
AI agents increasingly want to browse, compare, and buy. But JavaScript storefronts, bot-blocking, and fragmented product data are building an invisible wall. The industry is splitting, and most merchants have no playbook.
We Wrote About LiteLLM as a Supply Chain Risk. Then It Got Breached.
A supply chain attack compromised LiteLLM, the AI proxy present in 36 percent of cloud environments. Mercor lost 4TB of data. An AI agent was used operationally in the attack for the first time.
The Agentic Commerce Protocol Map: Q1 2026
Every production protocol for how AI agents discover, authenticate, pay, and settle. Ten protocols across four layers. Visa, Mastercard, Stripe, Google, and AWS are all at the same table. The definitive Q1 2026 map.
OpenAI Just Bought a Talk Show. That Is Not a Media Play. It Is a Political One.
OpenAI acquired TBPN and put it under its chief political operative. The advertising business winds down. Editorial independence is promised. The timing, weeks before an IPO, tells the real story.
Home Depot Just Made the Most Important AI Hire in Retail
Home Depot hired Ford's chief AI officer as CTO to lead its enterprise-wide AI integration. With Google Cloud already deployed to thousands of associates, this is not experimentation. It is restructuring.
Anthropic Built the Most Dangerous AI Model in History. Then It Accidentally Told Everyone.
Anthropic accidentally exposed details of Claude Mythos, a model it warned governments could make large-scale cyberattacks far more likely. Then it leaked its own source code. Now Congress wants answers.
OpenAI Is Now Worth $852 Billion. It Has Never Made a Profit.
v
Every AI Agent Tested Was Compromised. The Evidence Is No Longer Theoretical.
Google DeepMind mapped six attack categories. OpenAI admitted prompt injection may never be fully solved. Credit card exfiltration succeeded 10 out of 10 times. The security evidence is piling up.
AI Agents Just Got Real Jobs in Finance
From fraud scoring to debt collection to regulatory reporting, financial institutions deployed AI agents into production this month. The pilot phase is ending. The governance gap is not.
The Protocol That Turns AI from Chat into Infrastructure
Model Context Protocol crossed 97 million installs. Shopify, Beehiiv, and Shippo already use it. Here is what MCP actually does and why it matters.
The FCA Just Said Payments Regulation May Need Rewriting for Agentic AI. No Other Regulator Has Gone That Far.
The UK's Financial Conduct Authority has stated it will consider whether payments regulation needs to change for agentic AI. No other major financial regulator has drawn that connection publicly. We break down what it means.
Google Just Split the Web in Two. Your robots.txt Only Controls One Half.
Google now distinguishes between Googlebot, which crawls your site on a schedule, and Google-Agent, which acts on behalf of users in real time. Google-Agent ignores robots.txt entirely. For merchants, the access control question just changed.
The AI Proxy That Handles Your Agent's API Keys Just Got Backdoored. Here's What That Means for Agentic Commerce.
LiteLLM, the open-source proxy routing AI agent traffic to language models, was compromised with credential-stealing malware. For enterprises building payment-capable agents, the supply chain is now a proven attack vector, and the threat model has to change.
Arm Built Its First Chip in 35 Years. It Named It After the Technology Reshaping Payments.
The company that put processors in every smartphone on Earth just entered the data centre silicon market with a chip explicitly designed for agentic AI workloads. The infrastructure story now goes all the way down to the metal.
Gap Just Became the First Retailer to Let You Check Out Inside Google Gemini. Here's Why That Matters.
Gap is the first major retailer to enable checkout directly inside Google's Gemini AI. The same model that failed for Walmart inside ChatGPT is back, with one critical difference: the merchant retains control.
OpenAI Just Proved Walmart Right. Discovery Belongs to the Chatbot. Checkout Belongs to the Merchant.
OpenAI scrapped Instant Checkout and rebuilt ChatGPT as a pure discovery layer. Gap launched checkout inside Google Gemini the same day. The agentic commerce model is converging: AI owns discovery, merchants own checkout.
Agentic Commerce Just Went From Demo to Live Transaction
In a single week, three companies wired three different layers of the commerce stack for AI agents. Discovery, payments, and financial data are now live. The infrastructure era has begun.
Plaid Just Wired Open Banking to an AI Platform
Plaid connected brokerage accounts to Perplexity Computer, an AI platform orchestrating 20 frontier models. On the same day, Truist expanded its Plaid partnership. Open banking is becoming the default data layer for AI agents.
The Trust Gap: Agentic Commerce Has a Credibility Problem
Consumers want AI agents to shop for them. They just don't trust them to buy for them. Closing that gap is the defining challenge of 2026.
The Philosophy That Underwrites a Trillion-Dollar Bet
Marc Andreessen's "zero introspection" isn't a personality quirk. It's the foundational assumption behind a16z's entire AI investment thesis.
Google vs Anthropic: Two Approaches to Owning the Agent Layer
Google and Anthropic are both building comprehensive agent platforms, but their strategies are fundamentally different. One builds the infrastructure. The other builds the intelligence inside everyone else's infrastructure.
The $450 Billion Pipeline: How AI's Hardware Hunger Is Rewiring Global Payments
The US imported $450 billion in chips and computing hardware in 2025, up 60 percent in 12 months. Every chip crossed a border and settled through a payment rail. The AI boom is a payments story.
The Checkout That Nobody Used: What Walmart's Instant Checkout Flop Tells Us About Agentic Commerce
Walmart's exclusive data reveals conversion rates three times lower inside ChatGPT than outside it. The pivot to embedded Sparky is a masterclass in where value actually accrues in AI commerce, and a warning for everyone building on someone else's checkout.
Nvidia Just Built the Security Layer for AI Agents. The Payment Networks Haven't.
Nvidia's GTC announcements give enterprises a way to sandbox and govern AI agents. But compute-layer security and payment-layer trust are solving different problems, and the gap between them is where disputes will break down.
Who Gives AI Agents a Wallet?
Payments rails, identity layers, and settlement infrastructure are all being retooled for a world where AI agents spend money. This week, every layer moved at once.
MCP Just Entered Fraud Prevention. Fingerprint Is First.
The device intelligence platform launched the first open-source MCP server in the fraud space. Fraud analysts can now query device signals through any AI assistant.
Anthropic's Week: A Pentagon Blacklist, a Pricing Reset, and Visual Intelligence
In one week, Anthropic sued the US government, made long-context AI dramatically cheaper, and gave Claude the ability to think in pictures. Here is what it all means.
AI Is Driving a Global Fraud Surge. The Numbers Are Getting Worse.
Fraud losses climbed 9.2 percent in 2025, and artificial intelligence is accelerating both the attack and the defence. The industry is in an arms race it cannot afford to lose.
FedEx Is Building an AI Agent Workforce. Half Its Operations Will Run on It by 2028.
FedEx plans to embed AI agents into more than 50 percent of its workflows within two years. The architecture it described tells you exactly where enterprise agentic AI is heading.
MCP vs Agent Skills: The Two Layers Every AI Team Is Confusing
MCP gives agents hands. Skills give agents judgement. Most teams are building one and wondering why the other is missing.
Shopify Moves the Storefront Inside ChatGPT
OpenAI retreated from Instant Checkout. Shopify stepped in with "agentic storefronts" that let merchants sell directly inside ChatGPT. The template for how AI and commerce platforms divide value is now set.
The $650 Billion Squeeze, Five Weeks Later: Who Is Winning and Who Was Right
Five weeks ago, we published The $650 Billion Squeeze. The argument was straightforward. Big Tech was writing checks that rivalled national GDPs to build AI infrastructure.
We Built a Fact-Checker for the AI Writing Era. Here Is Why.
AI has made it easier than ever to publish. It has not made it easier to publish accurately. So we built a tool to close that gap.
The Last Human Job in AI
Thousands of professionals are training the models that replaced them. Agentic AI may soon replace them again.
Amazon's AI Code Crisis Exposes the Industry's Oversight Gap
After a string of outages linked to AI-assisted coding, Amazon is pulling the emergency brake on unchecked deployments. The question is whether the guardrails should have been there from the start.
The Invisible AI Revolution Happening Inside Every Product Catalogue
While the world watches chatbots, generative AI is quietly transforming the boring infrastructure that makes commerce work.
The AI Company That Said No to the Pentagon Just Got Backing from Microsoft, Google, and the CIA
Anthropic refused to remove Claude's safety restrictions for military use. The Pentagon classified it as a supply chain risk. Now Microsoft, 37 AI researchers, 22 former military leaders, and civil rights groups have filed briefs in its defence.
Meta Is Using AI to Fight Fraud Across WhatsApp, Facebook, and Messenger
Meta announced AI-powered scam detection tools across WhatsApp, Facebook, and Messenger. With 22 billion daily scam exposures across its platforms, the company is shifting from reactive removal to proactive interception.
Meta Bought the Platform That Couldn't Secure Its Own Database
Six weeks ago, Moltbook suffered a catastrophic security breach that exposed 1.5 million API tokens. This week, Meta acquired the team and the concept behind it. The play is agent identity infrastructure for the agentic web.
Amazon Just Drew the Legal Line on AI Shopping Agents
A federal judge blocked Perplexity's Comet agent from placing orders on Amazon. The ruling lands on platform authority over user delegation, and it could define the rules for every AI shopping agent that follows.
Nvidia Wants to Own the AI Agent Stack, Not Just the Chips That Power It
The chipmaker's open-source NemoClaw platform targets enterprise customers burned by OpenClaw's security failures, and it doesn't even require Nvidia hardware to run.
The Great Charter Rush: Why Every Fintech Wants to Be a Bank Now
From Revolut to PayPal to Ford, the race for US banking licences has reached a pace not seen in a generation. The question isn't who's applying. It's who's left that isn't.
The Week Agentic Commerce Got Its Rulebook
The card networks are racing to write the rules for agentic commerce. Google formalised a protocol. Stripe built the billing layer. Here's everything that mattered this week.
Claude Marketplace Is Anthropic's Bid to Become the Enterprise AI Operating System
Anthropic launched an app store for enterprise AI 24 hours after the Pentagon labelled it a supply-chain risk. The timing tells you everything about what this company is building, and who it is building it for.
Hollywood's AI Documentary Asks the Wrong Questions at the Worst Possible Time
The AI Doc arrives just as the abstract debates it stages have become painfully concrete, and it has nothing useful to say about any of them.
Block Cut 4,000 Jobs and Blamed AI. Here's What Actually Happened.
Jack Dorsey slashed 40 percent of his workforce and said most companies would follow within a year. But former employees, analysts, and the company's own history tell a more complicated story.
The Strait of Hormuz Is the Biggest Risk to the AI Economy That Nobody in Payments Is Talking About
The Iran conflict is threatening chip supplies, submarine cables, and energy costs simultaneously. For an industry building agentic commerce on the assumption of abundant AI infrastructure, the timing could not be worse.
The Delegation Problem
Google DeepMind publishes a framework for intelligent AI delegation just as the industry races to build the infrastructure without one.
AI Agents Don't Click Ads. That Changes Everything.
From warehouse to checkout, agentic commerce is rewriting how products get found, bought, and delivered.
Module 7: The MCP Ecosystem Playbook
Playing the long game. Four maturity levels from Exposed to Platform.
Module 6: MCP in Your Roadmap
Where MCP fits alongside everything else you are building.
Module 5: Selling MCP Internally
How to get budget, engineering time, and executive buy-in for MCP.
Module 4: The Hands-On Lab
Stop reading. Start touching. Four hands-on labs that take 15 minutes total.
Module 3: Anatomy of an MCP Integration
What your engineering team is actually building, explained without code.
Module 2: The Build / Buy / Ignore Decision
Should your team actually invest in MCP? A decision framework.
Module 1: The 60-Second MCP Pitch
Your CEO just asked what MCP means for your product. Here's what you say.
MCP as a Product Strategy Multiplier
The Product Person's Guide to Model Context Protocol
This Week in Payments, AI, and Commerce
The week AI's business model fractured. Here is what you need to know.
The AI Industry Just Split in Two
OpenAI is selling ads inside ChatGPT. Perplexity just killed theirs. Google is hedging. Anthropic ran a Super Bowl ad mocking the whole idea. The way AI platforms make money will determine what they become.
The AI Bought It Because Someone Told It To. You Just Don't Know Who.
Recommendation poisoning is the new SEO manipulation, except the stakes involve real money and the consumer never sees the ranking.
Who Authorised the Agent? The Identity Crisis at the Heart of Agentic Payments
Card networks are racing to pilot agentic commerce. The compliance frameworks to govern it do not exist yet.
We Built a Free MCP Course for Product Leaders
Every product team in 2026 is being asked the same question: "What's our AI integration strategy?"
The Web Just Got an API. The Machines Just Got Fast Enough to Use It.
Google's WebMCP turns every website into a structured tool for AI agents. OpenAI's Codex-Spark makes those agents fast enough to act in real time. Two announcements, 24 hours apart, and the agentic web just stopped being theoretical.
This Week in Payments, AI, and Commerce
The week machines started buying things. Here is what you need to know.
The Agentic Commerce Stack: How AI Agents Shop, Pay, and Close the Deal
Google, Alipay, and Coinbase are building the protocols that let machines buy things. Here is how the plumbing works.
The Marketplace Extinction Event: Why eBay and Amazon Are Fighting AI Agents for Survival
eBay is banning AI shopping bots. Amazon is suing them. Google just shipped live checkout through them. The platforms that built e-commerce are scrambling to avoid becoming back-end infrastructure nobody sees.
Consumers Want AI Agents to Shop for Them. A Calendar Invite Can Hijack the Whole Thing.
70 percent of consumers are ready to let AI agents handle their shopping. A CVSS 10.0 vulnerability shows why the security model is not ready to let them.
The Company That Was Supposed to Kill Ads Just Started Selling Them.
OpenAI launched advertising in ChatGPT this week. The financial reality behind the pivot tells you everything about where AI is heading.
The App Store for AI Content Is Coming. Amazon and Microsoft Are Racing to Build It.
The free-for-all era of AI companies scraping the web is ending. The question now is who becomes the intermediary, and what publishers actually get paid.
Search Isn't Dying. It's Splitting.
Google just posted its best quarter ever. OpenAI just launched a browser to replace it. Both things can be true, and the implications for advertising, commerce, and payments are enormous.
From Chatbot to Coworker: The Week AI Agents Went Corporate
OpenAI and Anthropic shipped agent management platforms within hours of each other. The message is clear: stop talking to AI and start managing it.
The $650 Billion Squeeze: Big Tech's AI Bet Is Building One Industry and Breaking Another
Hyperscaler capex has hit levels that rival national economies. The software companies those dollars are meant to replace are paying the price.
The AI Checkout Wars Have Begun
Last holiday season, something shifted. Traffic to US retail sites from generative AI tools increased by 693 percent year over year, according to Adobe Analytics. That alone would be a headline.
The Ad-Free AI Bet: Two Business Models, One Industry Fork
This week, two of the largest AI companies in the world made opposite bets on the same question: how do you monetize a conversational AI platform?
The Agent Org Chart: When Your Workforce Isn't Human
OpenAI's new platform frames AI agents as enterprise employees, raising a question enterprise leadership will face: what happens when headcount is measured in agents, not people?
Apple Bets on Agentic Coding
Apple does not partner easily. The company that builds its own chips, designs its own operating systems, and famously controls every layer of its stack announced on February 3 that it is integrating AI coding agents from Anthropic and OpenAI directly into…
When Your AI Takes Out Credit
Last week, Klarna joined Google's Universal Commerce Protocol. The largest buy now, pay later provider in the West is now plugging into the infrastructure that lets AI agents shop on your behalf.
The Front Door Problem: Agentic AI's Security Reckoning
In January 2026, developer Lucas Valbuena ran OpenClaw, a popular AI agent platform formerly known as Clawdbot, through a security analysis tool called ZeroLeaks. The results were catastrophic. Using Gemini 3 Pro, the platform scored 2 out of 100 points.
The Death of the Checkout Page
I have a secret: 95% percent of my holiday shopping was completed via Apple Pay. Find something, double-tap, glance at Face ID, done. No forms. No typing. No checkout page.
Amazon Just Admitted Physical Retail Doesn't Work. The Data Says They're Right.
Amazon announced today that it is closing every Amazon Go and Amazon Fresh store it operates. All 72 of them. The cashierless convenience stores that were supposed to revolutionise shopping. The full-service grocery stores that were going to take on Walmart.
When Your AI Does the Shopping
We had set this up months ago and mostly forgotten about it. The bag was running low, and Alexa, connected to an Amazon account with voice purchasing enabled, placed the order without asking. The confirmation email arrived during a meeting.
The AI Demand Shock: Why This Isn't Another Chip Shortage
A 64GB kit of DDR5 memory now costs more than an entire PlayStation 5.