The token budget

Anthropic's S-1, Microsoft and Google on capacity, and gated frontier models. How to read your AI providers like strategic suppliers.

Duet, Grok Bot and Dots can act on their own, yet all stop for a human yes. What businesses must build before AI agents can operate autonomously.

Google's AI rollout, agents crossing permission boundaries and software recreation: three business stories about testing, ownership and legal limits.

Mistral's Large 4 is trained and served in Europe, but its open weights come later. What sovereignty and agent identity mean for your AI roadmap.

The biggest launches weren't new models. The biggest failures weren't in what models knew.

The AI race is accelerating. This week, Anthropic and OpenAI released new frontier models just 90 minutes apart — both faster and significantly cheaper than what came before.

Four frontier models launched in five days, Astra's top score had two values, and Apple named a hardware engineer as CEO. The old shortcuts for choosing broke.

Four frontier models launched in five days, Astra's top score had two values, and Apple named a hardware engineer as CEO. The old shortcuts for choosing broke.

Four frontier models launched in five days, Astra's top score had two values, and Apple named a hardware engineer as CEO. The old shortcuts for choosing broke.

McKinsey put a number on the quiet procurement shift while Salesforce reported its strongest quarter in four years — both are true, and the gap is the most useful thing on your desk.

Amazon let agents pay for services, Stripe bought the model switchboard, and Anthropic shipped computer use — all in the week OpenAI priced supervising a model at 20% more compute.

Open weights arrived with conditions, the cheapest inference raised its prices behind a discount headline, and the money financing the whole build-out looked circular.

Wall Street stopped rewarding AI spending on faith, and the labs started cutting prices by up to 80% within weeks of launch. The economics of AI just got squeezed from both ends.

OpenAI's model escaped a controlled test and hacked Hugging Face on its own — one day before OpenAI launched a product to help you deploy trusted AI agents.

Microsoft just bet $2.5B and 6,000 engineers that enterprise AI's bottleneck was never the model — it was deployment. What it means for your 2026 roadmap.

Microsoft just bet $2.5B and 6,000 engineers that enterprise AI's bottleneck was never the model — it was deployment. What it means for your 2026 roadmap.

Microsoft just bet $2.5B and 6,000 engineers that enterprise AI's bottleneck was never the model — it was deployment. What it means for your 2026 roadmap.

Microsoft shipped Copilot Cowork to general availability this week and announced plans to replace Anthropic’s Claude with its own Cowork One model. Here’s what it means for your AI vendor strategy.

Microsoft shipped Copilot Cowork to general availability this week and announced plans to replace Anthropic’s Claude with its own Cowork One model. Here’s what it means for your AI vendor strategy.

This week, a US export control directive took Anthropic’s Claude Fable 5 offline for every foreign national worldwide — within hours. Here’s what it means for your AI vendor risk framework.

How Microsoft Build and Anthropic’s IPO filing changed the vendor map your roadmap depends on — in three days

How Anthropic’s Mythos reversal — announced three days after the Pope called for AI to be disarmed — changed the threat model for every business on a three-to-five-year AI roadmap

How Google's price move at I/O changed the question on every AI roadmap: from "is it good enough?" to "how much will we save when we move?"

Five business days made your consultant pick your AI — and put humanoid robots on the warehouse floor.

The week the AI gap stopped being a metaphor and started being a number, across three continents and three very different strategies.

Four major model releases in five days shifted the question from which model is best to how fast you can change your mind.

PwC quantified the gap between AI leaders and the rest, and within days it hit Wall Street.

A new frontier model exposed thousands of unknown vulnerabilities, pushing cybersecurity to the top of the agenda

Operational risk persists, even among the most safety-focused AI companies

OpenAI walked away from a billion-dollar Disney deal, signalling where AI value is moving.

The fastest-growing open-source project in history didn’t build a new AI model. It built a layer on top of them.

Four launches in one week. All aimed at the same person: the knowledge worker in front of a screen. Here’s why it matters.

Users are switching assistants, enterprises are deploying agents, and trust is quickly becoming the deciding factor.

The US government labels Anthropic a supply-chain risk, OpenAI raises record funding, and Europe's dependence on US AI infrastructure comes into focus.

Accenture ties promotions to AI usage as costs fall, agents scale, and regulation accelerates.

Speed is up. So is burnout risk.

This week brought a development business leaders should pay attention to.

Earnings season showed that AI investment is judged on outcomes, not ambition.

This was the week AI accountability became unavoidable.

This week, Apple made a decision that signals where AI value is heading. It has implications for every business thinking about their AI strategy.

Regulators paused. Microsoft and Google did not.

The AI market is maturing. Costs are falling. Capabilities are converging. For buyers, that’s the best news possible.

We’re moving into an era of AI specialisation. Not just at the model level, but across the entire stack.

Something big shifted this week. Warren Buffett, the man who avoided the dot-com hype and spent two decades sceptical of tech valuations, just bought nearly $5 billion of Alphabet stock.

Robotics is moving from concept to line item, and this affects the budget planning for 2027.

Something changed this week in how big tech views the road to artificial general intelligence (AGI).

OpenAI co-founder Andrej Karpathy’s comments on artificial general intelligence provide a reality check for the industry’s expectations .

The experimentation phase is over. This is deployment at scale.

China just became the world’s largest AI market. While the US bans Chinese tools, researchers still collaborate. And Europe? Preaching sovereignty while depending on both sides. Let’s unpack what’s really going on

New AI ecosystems are threatening Google's search dominance and the open web.

You might want to sit down for this one: Nvidia and Open launch $500 billion data centre project.

AI moves fast! But don’t worry, we’ve got you covered. 

Four frontier models launched in five days, Astra's top score had two values, and Apple named a hardware engineer as CEO. The old shortcuts for choosing broke.