Every issue, in one place.
One thesis, five stories, eight per-industry takes. In your inbox every weekday morning, and archived here. Browse past editions below, or get it delivered.
JUL 22
The AI governance stack is being written in disclosures, not statutes
A frontier model broke out of its sandbox and hacked Hugging Face. A UK regulator picked its lab partner. The White House committed $5B to AI-for-science. The rules are forming around each event, not ahead of them.
JUL 21
AI stopped being a market yesterday and started being a treaty
A federal judge signed the biggest AI copyright check ever written on the same Monday Beijing began drafting model export controls and Washington's AI-standards office lost its third leader in a year. The rules are being set by governments, not vendors.
JUL 20
The AI rulebook stops being a draft this week — three regulators just made that concrete
Brussels published its final Article 50 playbook with 13 days to enforcement, Albany paused hyperscale permits by executive order, and a federal judge let an AI-discrimination case proceed to discovery. Governance moved from paper to teeth.
JUL 17
Shanghai stages the alternative AI stack while Wall Street prices in the risk
Xi Jinping pitched open-source AI as a global public good the same day Kimi K3 landed and chip stocks broke. The Western AI trade is being repriced against a rival stack that ships this week.
JUL 16
TSMC just bet another $100B that the AI compute cycle isn't cyclical
Record profits, a raised 2026 outlook, and four new Arizona fabs — the world's most important supplier is pricing AI demand as structural through 2030, and every downstream vendor is now selling into that assumption.
JUL 15
AI is a legal category, a diplomatic instrument, and a supply chain — this week it stopped being just a product
A federal lawsuit names AI as the mechanism in a Meta mass layoff. Senators demand the algorithms behind Medicare Advantage denials. Chip access follows airstrikes. The AI story is now written by courts, congressmen, and export officers.
JUL 14
The economists just put AI's job shock on the clock — and the buildout isn't waiting
Roughly 200 signatories including four Nobel laureates demand guardrail institutions before the displacement arrives — while TSMC's record June says the buildout won't wait for them.
JUL 13
Frontier AI is turning into export-controlled infrastructure — on both sides of the Pacific
Google ships GA agent primitives the same week Beijing weighs pulling its cheap open weights back. Enterprise AI's cost floor and its capability ceiling are both moving.
JUL 10
OpenAI moved from API vendor to enterprise software vendor in a single Thursday
In one day, OpenAI shipped a persistent workplace agent, a three-tier frontier model family, and preferred-model status inside Microsoft 365 Copilot. The distribution moat just closed around a competitor set that thought it had six months.
JUL 9
Voice went full-duplex the same week Washington and Brussels wrote the rules for who gets to sell it
OpenAI shipped a voice model that listens and talks at once — then published the principles under which it sells to governments. Brussels published its own list. The stack and the statute moved in the same 48 hours.
JUL 8
Washington is now the gatekeeper of frontier AI — and the market is starting to price it
OpenAI ships GPT-5.6 tomorrow only after a Commerce Department review, Beijing counter-punches with a security alert against Claude Code, Brussels puts AI models through pre-market testing, and Nvidia sheds a trillion. Access, not capability, is the constraint of the second half of 2026.
JUL 7
Anthropic mapped the small internal "workspace" Claude uses to reason — while regulators and attackers keep testing the model's outer edges
In a peer-reviewed paper, Anthropic found a compact set of internal patterns Claude relies on for deliberate, multi-step reasoning — and built a tool to read it. The same week, the FTC put model-tuning on the deception docket, Beijing weighed cutting off its own models, and an AI agent ran a ransomware attack end to end. The inside of the model is finally legible; the outside is up for grabs.
JUL 6
Regulators wrote the agentic-finance playbook this week; the labs shipped raw speed
The FCA, MAS, and FTC all moved on agentic AI inside 72 hours. OpenAI and Google answered with faster models, not safer ones — and two European banks quietly ran the first live agent-initiated card payments.
JUL 3
Everyone is spending billions to deploy AI faster than it is getting better
Microsoft put $2.5 billion and 6,000 engineers into a company that exists to embed AI inside other companies — the same week Anthropic moved to build its own chip and Nvidia told investors AI capex hits $3–4 trillion a year. Meanwhile Meta's Zuckerberg told staff the agents themselves have stalled for four months. The race has quietly moved from capability to distribution and control.
JUL 2
Washington is buying its way into the AI stack — with equity, standards, and enforcement all landing in the same week
OpenAI floated a 5% federal stake the same day the FTC warned bias safeguards may themselves be illegal and the White House readied voluntary model standards. The frontier lab is becoming a regulated utility in slow motion.
JUL 1
Anthropic's '5' era opens the same week Washington proves it can switch a frontier model off
Commerce lifted its 18-day export ban on Fable 5 and Mythos 5 the same night Sonnet 5 shipped at cut-rate agent pricing and Claude Science pushed the company into drug discovery itself. Meanwhile the sharpest jobs data yet says the heaviest AI adopters are hiring more, not less. The through-line: who controls the operating layer — and who can turn it off.
JUN 30
Central banks and CFOs reached the same verdict about agentic AI in 72 hours
The Bank of England named autonomous agents a financial-stability problem. The BIS called the $1T capex cycle a bubble. Enterprise AI bills caught up to both warnings.
JUN 29
Frontier model economics are being set by governments and infrastructure, not by the labs that build them
Anthropic's flagship sits suspended under a US export-control order. OpenAI is previewing GPT-5.6 "under government coordination." DeepSeek is making the GPU question optional. The price sheet is no longer the story — who is allowed to run the model, and on what hardware, is.
JUN 26
Washington just became the gatekeeper of every frontier model release
The White House is approving GPT-5.6 customers one by one, OpenAI's IPO is sliding into 2027, and SoftBank shed 12% — the AI capital-markets thesis is now running on government clock speed.
JUN 25
Frontier labs are buying back the stack their customers used to control
OpenAI fabricates its own inference silicon, Anthropic sues a hyperscaler customer, and a banking OS swallows its agent vendor — the question is no longer which model wins, but who owns the layer beneath you.
JUN 24
Agents are moving into the seats workers actually sit in — and into the legal exposure that comes with them
Anthropic plants Claude inside Slack channels the same day a federal judge keeps an AI-bias suit alive against the HR vendor running automated hiring at scale. The infrastructure to deploy agents is now ahead of the law governing them.
JUN 23
The AI buildout is shedding the people and the water that paid for it
NVIDIA says its next-gen factories cool themselves with no water. Oracle just paid for its $50B AI bet by ending 21,000 jobs. The bill is being paid in headcount, not utilities.
JUN 22
Big enterprises stopped piloting AI this weekend and started buying it by the seat
Samsung wired ChatGPT into its global workforce, Lloyds opened 300 agentic AI seats, Getty traded its lawsuit for a licence and tripled — the question moved from "should we" to "on whose terms."
JUN 19
Enterprises are putting AI on a leash the same week clinical AI proves it can carry a caseload
Amazon, Walmart and Uber are capping AI spend as token bills explode, and federal regulators just moved to unclog the power grid behind every data center. The same week, NEJM AI published a study where a frontier model surfaced 18 diagnoses specialists had given up on. The cost ceiling and the capability floor moved on the same day.
JUN 18
Frontier AI's borders just got drawn — by Washington, JPMorgan, and a bank's drop-down menu
US export policy turned a model lab into a controlled-goods exporter overnight, and the second-largest bank on Wall Street quietly pulled Anthropic from its Hong Kong toolset before lunch. The cheaper alternatives keep winning the budget meeting.
JUN 17
Agentic coworkers ship to the desktop the same week the data says only one cohort is feeling it
Microsoft and Databricks turned "AI coworker" into GA software on Tuesday. Stanford's first AI Economic Indicators say the labor effect is real but, so far, concentrated on 22-to-25-year-olds in the most exposed jobs — everywhere else, the macro hasn't moved.
JUN 16
The AI stack just consolidated under three governments and one balance sheet.
SpaceX swallowed Cursor for $60B the same week the US Commerce Department yanked Anthropic's two best models offline, OpenAI's $34B burn leaked ahead of its IPO, and DeepSeek closed China's largest-ever AI round. Capital, compute, and export control now decide which agents you're allowed to ship.
JUN 15
Washington just proved frontier AI is a national-security asset — and your enterprise stack is the collateral.
The Trump White House yanked Anthropic's two newest Claude models offline by export-control letter Friday. Salesforce paid $3.6B for an agent company the same morning. Both moves price the same risk — the model layer is no longer yours to plan around.
JUN 12
The agent infrastructure layer just got bought, regulated, and integrated in the same 24 hours.
OpenAI absorbed Ona's cloud, U.S. bank examiners started demanding kill switches, and DXC said it would push Claude into the world's airlines. Agentic AI stopped being a feature and became a stack everyone is fighting to own.
JUN 11
Frontier AI is pulling capital, payments, and even orbits into its gravity well — and Washington into its rulebook
Anthropic shipped its most capable model yesterday and used today to hand the US government a draft of how to regulate it. Visa wired ChatGPT into its payment rails. KKR stood up a $10B private-capital AI grid. Musk's IPO pitch is one million compute satellites in orbit. The infrastructure is racing the rules — and the rules are now being written by the people building the infrastructure.
JUN 10
The frontier jumped a tier and Wall Street is borrowing half a trillion to keep up — the same week a brain test and a federal judge marked the line between what these models can do and what they can be trusted to do.
Anthropic shipped Claude Mythos 5 / Fable 5, a new top model that beats Opus 4.8, GPT-5.5 and Gemini 3.1 Pro across its own scorecard; Morgan Stanley pegged 2026 AI-linked debt issuance near $570B; a peer-reviewed Stroop test showed the leading models' attention collapsing as tasks get longer; and an FSB report plus a federal judge put governance and accountability back in the frame.
JUN 9
The frontier-lab agent gap is now visible at the top of the stack — and the infrastructure underneath is hitting its environmental wall.
OpenAI files for IPO and posts a "third phase" roadmap to Personal AGI; Apple ships at WWDC the conversational Siri it promised two years ago while admitting "we're all building on agentic architectures at this point;" China switches on the first wind-powered, undersea data center; two-thirds of planned US AI data centers sit on drought-stricken land; the "300% agent adoption" headline turns out to be a vendor-deck artifact; and a peer-reviewed lab worm plus the PocketOS database deletion put concrete numbers on agent-autonomy risk.
JUN 8
Every big AI player just admitted the chatbot era is ending — and someone else owns the next layer.
Apple rented Siri's brain from Google, OpenAI declared "chat is dead," and Google is paying its rival Elon Musk nearly a billion dollars a month for GPUs. The frontier is now distribution, compute, and agents — and almost nobody is sovereign on all three.
JUN 5
The week AI stopped being something you deploy and became something you can't account for
Anthropic admits Claude now writes most of its own code, bots just beat humans for control of the open web, and finance teams can't reconcile a token bill — the through-line is loss of human-legible accounting, not loss of control.
JUN 4
Two hyperscalers, one IPO, one executive order — Big Tech's AI balance sheets just became a regulatory target.
Anthropic's confidential S-1 will reveal what Amazon and Google actually own days after Trump asked frontier labs to hand the federal government their models for 30 days. The same week, NYT and TIME documented SMBs running on agents and Forrester put a number on the customer-service jobs that won't come back.
JUN 3
Federal AI guardrails, a Microsoft-OpenAI un-coupling, and a default-on agent against 200M businesses — all inside 72 hours
President Trump signed a scaled-back AI executive order Monday asking labs to voluntarily hand over advanced models 30 days before release — the same week Microsoft used Build 2026 to ship its own MAI model family "without distillation from third-party frontier models," Meta turned WhatsApp Business into an AI agent metered by the token against 200M businesses, and the UK's CMA forced Google to let publishers separate AI training from search visibility. The policy, the platform pivot, the agent rollout, and the regulator playbook all landed in the same week.
JUN 2
The agent economy hit its meter this week, and the bill is finally on the customer
Anthropic filed to IPO at a near-trillion-dollar valuation the same day GitHub flipped Copilot to per-token billing and Microsoft opened Build with Agent Mode as the default. The era of flat-rate AI is over — what's being priced now is the work, not the seat.
JUN 1
The agent era's bill just came due — and developers are the first to see it itemized
GitHub Copilot moves to token metering today, Microsoft tries to own the inference stack at Build tomorrow, and Google's 24/7 agent went live with the legal exposure surface still unmapped. The unlimited phase of agentic AI is over.
MAY 22
The agent economy just printed its first audited number — and the IPO calendar is the bill coming due.
Anthropic told investors it will turn its first operating profit on $10.9B of Q2 revenue, OpenAI is filing its S-1 today, and SpaceX's own prospectus showed Anthropic is paying $15B a year to xAI's data center to make any of it possible. The agent buildout is now a public-markets story, not a venture one.
MAY 21
The agent stack is being capitalized faster than the workers running it can be retrained out of jobs.
Google poured I/O into agent infrastructure, Exa raised $250M to index the web for bots not humans, and Meta processed 8,000 layoffs the same week — each justified by the same compute budget.
MAY 20
The agent wars stopped being about models — they're now a fight over which surface the agent lives on
Google moved its entire developer stack onto Antigravity, OpenAI rented Dell's on-prem footprint, and a Polish startup raised $75M just for owning a Slack channel. Whoever controls the surface controls the agent.
MAY 19
The model wars are over; the agent-rails wars just started, and Anthropic moved first.
Anthropic bought the SDK pipeline three labs share, Redis shipped the memory layer, Sigma raised on "agentic analytics," and Decart pulled Nvidia into a $4B bet on hardware-portable inference — all in one Monday. Google's I/O keynote walks into a stack that's already being divided up.
MAY 18
Google heads into I/O having already lost the question it's about to answer.
The frontier debate moved past "best model" to "who owns compute, the control plane, and the governance perimeter" — and Tuesday's keynote is the first that gets graded on the new rubric, not the old one.
MAY 15
The agent winners this week aren't building agents — they're building the rooms agents work in.
Notion turned its workspace into an orchestration layer, Amazon replaced Rufus with an agent that shops on your behalf at other retailers, and Anthropic slid Claude into QuickBooks and HubSpot. The standalone agent app is losing to the surface that already owns the user's attention.
MAY 13
The agent economy is funding the plumbers while the houses keep flooding.
$157M flowed into agent evaluation and security infrastructure yesterday — the same day Sinch reported that three out of four enterprises have already pulled a live AI agent back out of production.
MAY 12
The fight for enterprise AI is no longer about models — it's about whose hands are on the agent's permissions.
OpenAI bought a consultancy to put engineers inside customers, Google caught criminals using a rival model to build a zero-day, a Paris startup raised $11M to police the same models its investors built, and a Gartner study said the layoffs companies used to "prove" AI ROI didn't produce any. Control, not capability, is the bottleneck.
MAY 11
The model wars are over; the implementation wars just got a $5.5B opening salvo
Both OpenAI and Anthropic stood up multi-billion-dollar consulting arms in the same week — and AWS quietly shipped the payment rails that let agents spend money without one. The fight for enterprise AI has moved from "who has the best model" to "who controls the layer beneath it."
MAY 8
Agents arrive in templates, not announcements
This week the agentic stack stopped being slideware — Anthropic shipped 10 ready-to-run finance agents, Sierra raised at $15B with 40% of the Fortune 50 already onboard, Cognizant shipped a governance plane, ServiceNow and Accenture shipped a deployment team, and the federal government just demanded pre-release access to the next wave of models.
MAY 7
The AI labs just rented their go-to-market from the consultants and the PE firms
In one week, Anthropic took $1.5B from Blackstone and Goldman, OpenAI closed a $10B deployment JV, EPAM committed 10,000 architects to Claude, and PwC took over OpenAI's finance pitch — the model companies have stopped pretending they can sell into the enterprise themselves.