GenAI now has 53% global adoption – it reached what’s defined as mass adoption faster than the internet did.

Organisational adoption is now 88%, but they’re mostly not using agents yet. Agent use is <10%. Organisational AI use is growing fastest in Europe and China. ← This number will grow a lot this year. We have seen very little of the real value impact AI will have in organisations so far — and the ramifications of AI taking over lots of repetitive tasks humans used to do.

Big divide in expectations of AI’s future impact between experts and the general population. For example, 73% of experts expect a positive impact from AI on how we do our job, versus only 23% of the public.

It’s huge in scope – it combines global data into a full PDF of 423 pages.

It’s the most authoritative + independent/unbiased report at this scale, since it comes from Academia.

It’s longitudinal and has been ongoing for almost a decade. That gives some unique metrics that let you compare things over time.

A 20-year-old was arrested after throwing a Molotov cocktail at Sam Altman's Russian Hill home and threatening to burn the OpenAI HQ. The suspect had published AI-extinction essays and used the handle "Butlerian Jihadist" online. Altman responded with a blog post that got more personal than usual, written at 3am after the incident.

Microsoft declared a Copilot Code Red to catch up in the AI race. The Information and TechCrunch reported Microsoft is sprinting internally to launch new Copilot features inspired by OpenAI's OpenClaw. It’s…it’s the future: I see a lobster, working at night, in your spreadsheets!

Anthropic launched Claude Managed Agents, production-grade cloud-agents via API. It allows you to ship agents into production more easily, as they run on Anthropic’s own infrastructure. You define the agent’s task, tools and guardrails and they set up the environment for this to be run in. The pricing is per token plus $0.08 per session-hour — you can’t use your Claude subscription for this because Anthropic would not make money then. There was some sell-off across SaaS stocks after the announcement, which is what stocks seem to do when Anthropic releases anything. The initial reviews I’ve seen of managed agents are leaning negative, like it lacks some functionality that you would expect for the target user group, such as being able to do scheduled triggers.

OpenAI expanded Codex into a superapp with background computer use, an in-app browser, and parallel agents. The update is OpenAI's clearest step yet toward bringing the experience of Codex, ChatGPT, and the Atlas browser together under the same platform. OpenAI is deliberately positioning Codex as more than a coding tool now, more like a default workspace, like Claude Cowork.

Perplexity integrated Plaid so its Computer agent can connect bank accounts, credit cards, and loans to create a personalised finance hub with your own workflows. The agent can build things like budgets, net-worth trackers, debt payoff plans, and retirement dashboards from text prompts. As someone who’s been using AI to help me with taxes for a couple of years already, I think this is a cool and innovative use case. Also yes, it’s a bit worrisome that we’re all starting to give away our personal finance data to AI 💀

Meta shipped Muse Spark, their first model coming out of their new SuperIntelligence Labs — and it’s not open source. They’ve invested billions of dollars into this division to compete with Google, OpenAI and Anthropic. The most standout thing about Muse Spark is something they call contemplating mode that pits multiple agents against each other on hard problems. Benchmarks are competitive with Opus 4.6 and GPT-5.4 on reasoning but it lags in coding skills. It’s also a proprietary model, ie not open source like Llama. Meta has been a driver in the open source AI landscape — it seems like they’re trying to be a little less open going forward?