Another week, another avalanche of AI news that makes you question whether you're keeping up or falling behind. Spoiler: if you're reading this, you're doing better than most. Let's cut through the noise.
1. Anthropic Publishes Study Showing AI Makes You Worse at Learning
In what might be the most honest corporate move of the year, Anthropic — the company that makes Claude — published research showing that developers who used AI assistance while learning a new programming library performed significantly worse when tested without it. The control group who struggled through it the old-fashioned way retained far more.
Why this matters: This is the AI equivalent of a cigarette company funding lung cancer research. Respect to Anthropic for publishing it, but more importantly — it validates what a lot of us suspected. AI is a phenomenal productivity tool for things you already know. For things you're learning, it might be a crutch that atrophies the muscle. The Aigenda take: use AI to accelerate your strengths, not to skip the reps on your weaknesses.
2. Microsoft x Anthropic Launch "Copilot Cowork" — Your AI Office Worker
Microsoft and Anthropic have partnered on Copilot Cowork, an AI system that takes a desired outcome, breaks it into steps, pulls context from your emails, meetings, files and chats, then executes across Outlook, Teams, Excel, and PowerPoint. Essentially, an AI employee that lives inside your Microsoft 365.
Why this matters: This isn't "AI-assisted" work — it's AI-executed work. You describe what you want done, it does it. For the millions of knowledge workers whose jobs are essentially "move information between apps and format it nicely," this is the starting gun. The interesting question isn't whether this works — it's how quickly the "I describe, AI delivers" workflow becomes the baseline expectation for every office job.
3. Lightpanda: The Open-Source Browser Built for AI Agents
Someone built a headless browser from scratch — not a Chromium fork, a completely new browser written in Zig — specifically designed for AI agents, scraping, and automation. It runs 11x faster than Chrome and uses 9x less memory. It's called Lightpanda, and it's open source.
Why this matters: If you're building AI agents that need to interact with the web (and increasingly, who isn't?), the browser has been a bottleneck. Chrome is designed for humans watching YouTube, not for thousands of AI agents running parallel web tasks. Lightpanda is infrastructure for the agentic future — boring but absolutely critical plumbing. This is the kind of open-source project that quietly becomes essential.
4. OpenAI Launches Sora 2 Video API
OpenAI dropped new Video API capabilities powered by Sora 2: custom characters and objects, 16:9 and 9:16 exports, clips up to 20 seconds, video continuation, and batch generation. Video creation is now an API call away.
Why this matters: Remember when AI-generated images went from "novelty" to "how half the internet's visuals get made" in about 18 months? That's about to happen with video. The API approach is the real story here — this isn't about creatives playing with a tool, it's about developers embedding video generation into products, workflows, and automation. The content creation landscape is about to get very, very noisy.
5. Anthropic's Labour Report: The Jobs Most at Risk Aren't the Ones You Think
Anthropic released an AI labour impact report suggesting that the jobs most exposed to AI disruption over the next few years are exactly the ones people assumed were safe. The report has racked up thousands of shares and a fair amount of existential dread.
Why this matters: Every "AI won't replace you, a person using AI will" platitude just got a reality check. The specifics matter here — if the "safe" jobs are now in the crosshairs, the calculus for career planning and business strategy shifts dramatically. Whether you're hiring, managing, or doing the work yourself, ignoring this report is a choice with consequences.
6. Ex-Anthropic Scientists Raise $1B for AI Scientific Discovery Startup
A team of former Anthropic researchers is raising $175 million at a $1 billion valuation for Mirendil, a startup focused on using AI to accelerate scientific discovery. Sam Altman has also spoken about OpenAI's goal of building autonomous AI researchers by the end of the decade.
Why this matters: The AI talent exodus from big labs into focused startups is accelerating. Mirendil's bet is that the real value of frontier AI isn't chatbots — it's using these models to make scientific breakthroughs that humans alone would take decades to achieve. If this works, the ROI isn't measured in productivity gains — it's measured in new drugs, materials, and energy solutions. The ambition here is genuinely staggering.
7. Perplexity Becomes a Full-Stack Agent Platform
Perplexity announced that its API platform is now a full-stack, model-agnostic platform for building agents — replacing your model provider, search layer, and embeddings in one go. Built on the same infrastructure that powers Perplexity's search product.
Why this matters: The "picks and shovels" play in the AI gold rush just got more interesting. Perplexity is positioning itself not as a search engine but as the platform layer that agent builders use. Model-agnostic is the key word — they're betting that which LLM you use matters less than the infrastructure around it. Smart positioning in a market where model commoditisation is accelerating fast.
8. Google DeepMind Opens "Platform 37" — Major London AI Hub
Google DeepMind unveiled Platform 37, a massive new London office in King's Cross. Named to honour the area's transport heritage and AlphaGo's famous "Move 37," it signals DeepMind's continued commitment to London as a global AI research centre.
Why this matters: In the ongoing debate about whether AI's centre of gravity will remain in San Francisco, DeepMind is making a very expensive argument for London. This isn't just an office — it's a talent magnet. For the UK AI ecosystem, having DeepMind doubling down on London is enormously significant. The brain drain narrative just got a counterpoint.
That's your week in AI. Eight stories, zero fluff.
If this was useful, follow @theaigenda for the signal, not the noise. See you next week.