Context Windows Got Huge and Nobody Knows What to Do With Them
Million-token context windows were supposed to change everything. In practice, most AI products ignore them entirely.
Read more →AI
36 posts
AI Benchmarks Are Lying to You, and the Labs Know It
The numbers AI companies publish to prove their models are smarter keep going up. The actual experience of using those models tells a different story.
Read more →
AI Agents Were Supposed to Replace Workflows. They've Become Another Layer of Them.
Autonomous AI agents promised to collapse your tool stack. Instead, they've quietly become the most expensive middleware you didn't ask for.
Read more →
AI Agents Are Failing in the Boring Ways Nobody Benchmarks
Agents can book flights and write code, but ask one to handle a multi-step refund and watch it unravel completely.
Read more →
The Open-Source AI Gap Is Closing, But Not in the Way Anyone Planned
Open-source models have nearly caught up to frontier AI - but the companies celebrating that fact are the ones who should be worried.
Read more →
AI Agents Are Shipping Before Anyone Agrees on What They're Supposed to Do
Every major AI lab is racing to release autonomous agents, but the definition of 'agent' is still being argued in real time.
Read more →
AI Memory Is Getting Better at Remembering the Wrong Things
Persistent memory in AI assistants sounds like a feature. In practice, it mostly learns your bad habits.
Read more →
AI Memory Is Getting Better at Remembering the Wrong Things
Persistent memory in AI assistants sounds like a win. In practice, it surfaces a problem nobody designed for.
Read more →
OpenAI's API Pricing History Reveals the Real Business Model
Every time OpenAI cuts API prices, the narrative is democratization. The actual story is about locking in developers before the margin improves.
Read more →
AI Benchmarks Have Become a Marketing Department
The numbers labs publish to sell their models tell you almost nothing about how those models perform on your actual work.
Read more →
Context Windows Got Huge and Nobody Knows What to Do With Them
Million-token context windows were supposed to change everything. In practice, most people are still pasting in three paragraphs.
Read more →
The Reasoning Model Hype Is Hiding a Billing Problem
Reasoning models cost significantly more to run than standard models - and most products built on them haven't figured out who absorbs that.
Read more →
Benchmark Scores Are Eating AI Development Alive
The labs are optimizing their models to ace evals instead of solving real problems, and the gap between scores and usefulness keeps widening.
Read more →
Anthropic's Safety Argument Is Starting to Sound Like a Business Strategy
Claude's maker has built its brand on AI safety. That's admirable - until you notice how conveniently it maps onto competitive positioning.
Read more →
The Context Window Is Not Memory, and Conflating the Two Is Making AI Products Worse
Expanding context windows hasn't made AI assistants more capable of knowing you. It's made it easier to confuse size with understanding.
Read more →
OpenAI's Operator Strategy Is Starting to Look Like a Platform Play, Not a Product
OpenAI keeps building tools that make more sense as infrastructure than as consumer software. That's not accidental.
Read more →
OpenAI's Operator Push Is a Bet That Most Users Will Never Make
OpenAI wants AI agents to act on your behalf. The problem is that trust, not technology, is the actual barrier.
Read more →
OpenAI's Operator Push Is Solving a Problem Users Don't Have Yet
AI agents that book restaurants and file forms sound useful - until you realise the friction they remove was never the real obstacle.
Read more →
OpenAI's Operator Model Is a Bet That Most Users Won't Read the Fine Print
Giving AI agents access to your browser and accounts sounds powerful. The real question is who benefits when something goes wrong.
Read more →
OpenAI's Operator Push Is About Dependency, Not Productivity
OpenAI wants AI agents doing your tasks. That's less about helping you and more about making you impossible to leave.
Read more →
OpenAI's Operator Ambitions Are Getting Ahead of What the Models Can Actually Do
OpenAI wants GPT-4o to book your flights and manage your inbox. The gap between that pitch and reality is still enormous.
Read more →
OpenAI's Operator Vision Is Already Failing the Basic Trust Test
AI agents that browse, book, and buy on your behalf sound useful. But the trust infrastructure to make that safe barely exists.
Read more →
A $2,000 AI-Generated Film About Iran's Protests Heads to Tribeca
Dreams of Violets, a $2,000 AI-generated film dramatizing Iran's crackdown on protesters, will premiere at the Tribeca Festival next month.
Read more →
Anthropic closes $65B Series H round at near-$1 trillion valuation
Anthropic has closed a $65B Series H round at a $965B valuation, potentially its last private raise before an IPO, with backing from Sequoia, Amazon, and others.
Read more →
Anthropic Plans Public Release of Mythos-Class Claude in Coming Weeks
Anthropic says its Mythos-class Claude model could reach general users within weeks, as the company works to finalize cybersecurity safeguards ahead of a public release.
Read more →
Anthropic's Claude Opus 4.8 Is Designed to Flag Its Own Uncertainty
Anthropic's Claude Opus 4.8 is built to better acknowledge when it's uncertain, rather than presenting unsupported conclusions with false confidence.
Read more →
Anthropic's Opus 4.8 Adds Dynamic Workflows for Large-Scale Agent Tasks
Anthropic released Opus 4.8 with a new Dynamic Workflows feature for coordinating large numbers of parallel subagents, plus improved handling of uncertain data.
Read more →
Apple's Redesigned Siri for iOS 26 May Live Inside the Dynamic Island
Bloomberg's Mark Gurman has shared an early look at Apple's Siri overhaul, expected at WWDC 2026, including a Dynamic Island home and a new chatbot-style interface.
Read more →
Asana Acquires No-Code Agent Builder Stack AI for $75 Million
Asana acquired no-code agent builder Stack AI for $75 million, adding its founders and technology to Asana's growing suite of AI workflow tools.
Read more →
ChatGPT's New 'Trusted Contact' Feature Alerts Friends to Self-Harm Risk
OpenAI's new Trusted Contact feature lets ChatGPT alert a nominated friend if a user shows signs of being at serious risk of self-harm, with human review required before any notification is sent.
Read more →
Financial Exchanges Are Building Futures Markets for AI Compute Tokens
Major financial exchanges, including Shanghai's futures market and CME Group, are developing derivative products tied to AI compute tokens and GPU rental costs.
Read more →
How Paris Became a Serious Contender in the Global AI Race
Paris is emerging as a major hub for AI research, startups, and policy - challenging Silicon Valley's long-held dominance in the global tech conversation.
Read more →
How to Run a Local AI Chatbot on Your iPhone
Running an open-weight AI chatbot locally on an iPhone can cost as little as a one-time $5 purchase, with added benefits of offline use and stronger privacy.
Read more →
Microsoft 365 Copilot redesigned with faster load times and progressive UI
Microsoft is rolling out a redesigned Microsoft 365 Copilot with a faster interface, structured responses, and a context-sensitive UI approach called progressive disclosure.
Read more →
Microsoft's Copilot Gets a Stripped-Down, Monochrome Redesign
Microsoft has redesigned Copilot for Microsoft 365 with a minimal, largely black-and-white interface aimed at consistency across Word, Excel, and PowerPoint.
Read more →
YouTube Lets Users Shape Their Feed With AI Prompts
YouTube is rolling out a prompt-based AI feature that lets US users generate a personalized Home feed, with open questions about what it means for creators.
Read more →