For years, GitHub was the boring part of software development. Developers wrote code, opened a pull request, waited for review, merged it, and moved on. The system was built around a fairly simple assumption: a human was creating most of the work. That workflow starts to look a little different when your coding assistant can work on multiple tasks, generate changes and open pull requests while you're doing something else. Cursor seems to be betting that this is going to change more than just how developers write code. The company just launched Origin, its own code-hosting platform, bringing repositories, pull requests and Cursor's AI agents into the same environment. Then, almost immediately, GitHub had a major outage. The timing made for an easy headline: Cursor launches a GitHub competitor as GitHub goes down. But that's not really what matters. GitHub came back up. What matters is why Cursor is moving into code hosting at all and what it sees changing as AI agents become a much bigger part of software development. Because Origin isn't really about giving developers another place to store their Git repositories.

On Sunday, Shopify CEO Tobi Lütke did something most machine learning engineers spend months trying to achieve. He improved a core model's performance by 19% while he was asleep & didn't use a massive compute cluster or a team of researchers. He used a 630-line weekend project released by Andrej Karpathy called autoresearch. By the time he woke up, the agent had run 37 experiments, tested dozens of hyperparameter combinations, and handed him a 0.8B model that outperformed the 1.6B model it was meant to replace. Karpathy's response when he heard? "Who knew early singularity could be this fun." That's the story everyone is sharing. But the more interesting story is what autoresearch actually is, how it works, and what it quietly says about where AI research is heading.

OpenAI dropped their latest model yesterday and buried inside the benchmarks is a number that deserves more attention than it's getting. On GDPval, a test that puts AI agents through real professional tasks across 44 actual occupations, GPT-5.4 matched or outperformed human professionals 83% of the time. The previous version sat at 71%. That's not a small jump. And this isn't GPT writing emails or summarizing documents anymore. This version can move a mouse, click buttons, fill out forms, and work across applications the way a person sitting at a desk would. It scored 75% on OSWorld, a benchmark that tests exactly that. The average office worker scores 72.4%. The model is already better at operating a computer than most people who use one for a living & 83% is just the beginning of what this release actually means.

Most open source models that claim agentic capability are really just instruction-tuned models with tool calling bolted on. They can call a function. They cannot think across ten steps, remember what they decided three tool calls ago, and course correct when something breaks mid-task. This is where Trinity-Large-Thinking comes into picture. Arcee AI released it this week. 398 billion total parameters, but only 13 billion active during inference. That MoE architecture means it runs closer to a 13B model in practice while carrying the knowledge of something nearly 30 times larger. And unlike most models where reasoning stops between steps, Trinity keeps its thinking tokens alive across the entire agent loop. Every decision it makes is informed by everything it reasoned through before it.

Cohere spent the past year deploying North, its enterprise AI workspace, with actual customers doing actual work. Agentic question answering over company file systems. Data analysis across spreadsheets. Multi-session memory that has to hold up in production. Command A+ is what came out of that, a model shaped by a year of watching enterprise workflows break and figuring out why. The result is a 218B mixture-of-experts model with 25B active parameters at inference time, available today on Hugging Face under Apache 2.0. It replaces five separate models in the Command A family, each of which handled one thing. This one handles all of them, and on most of the tasks those specialist models were built for, it wins.

GitHub says hackers reached thousands of internal repositories after compromising an employee device through a malicious VS Code extension. That detail matters more than the breach itself because this keeps happening now. OpenAI got hit through a poisoned developer dependency earlier this year. The European Commission got compromised through a similar supply chain route. Attackers are increasingly targeting the tools developers trust instead of trying to break company infrastructure directly. And honestly, it makes sense. A developer machine already has access to everything attackers want. This GitHub incident is another reminder that the weakest point in modern software security might not be the company. It might be the extensions, packages, and tools sitting inside a developer’s editor.

Every few years something shows up in AI that makes people stop and argue. Not argue about which model is better or whose benchmark is more honest. Argue about whether the rules just changed. SubQ is that argument right now. A Miami-based startup called Subquadratic came out of stealth last week with a single claim that's either the most important architectural shift since the 2017 transformer paper or the most sophisticated AI hype in recent memory. They say they've built the first LLM that doesn't rely on quadratic attention and that this lets them run a 12 million token context window at roughly one-fifth the cost of frontier models. The AI research community split within hours. Half are losing their minds. Half are explaining why this doesn't count. The truth is probably more interesting than either camp. Here's what we actually know.