Engineering
OpenAI announced on Saturday that it had hit its "automated research intern" milestone, and hit it ahead of schedule. The number everyone quoted was 3.1: as of mid-August, OpenAI's research organisation was running 3.1 agent-workdays for every human workday. The number nobody quoted was
Changelog
Every release has a personality. This one is about trust: making sure your AI employees stay logged in, making sure your integrations are actually protected, and making sure the free tier behaves the way you expect it to. Here is what is new in v2.7.13. Your Claude employees
Engineering
The question that used to define an AI project was "can the agent do the task?" The question that defines it now is subtler and much more important: "when it can, should it do it alone?" Somewhere between a human approving every keystroke and an agent
Changelog
This release is mostly about Nova voice growing up: stronger security under the hood, calls that recover on their own, a cleaner transcript, and the groundwork for the native mobile app. There is also a Skills page improvement we are happy to ship, plus the Chrome extension chat catching up
Changelog
Some bugs announce themselves. This one hid inside the exact behavior that made everything feel broken: the longer and more valuable a conversation became, the more likely it was to fall apart. v2.7.2 is a small, focused release that fixes it, and the fix is worth understanding because
Engineering
Last week a security researcher showed that a fake bug report, slipped into a Sentry error feed, could get a coding agent to run attacker-controlled code using the developer's own credentials. Every step was authorized. Nothing in the stack had reason to flag it. The agent read a
Engineering
The benchmarks tell a confusing story. On OSWorld, the standard test for agents that operate a real computer, the best systems now score around 82% while well-known names sit down at 22 to 38%. Stanford's AI Index shows the field as a whole jumping from 12% to 66%
Future of Work
There is a quiet identity crisis happening in engineering right now, and the data backs it up. CIO reported that the engineer of 2026 spends less time writing foundational code and more time orchestrating a portfolio of agents, components, and external services. If you write software for a living, you
Engineering
There is a recurring scene that plays out in our customer onboarding calls. The user just got off another vendor's tool. They show us the bill. They tell us the price was supposed to be a flat monthly tier. The actual invoice is two or three times the
Engineering
There are two ways to put an AI worker into production. Most teams call them the same thing. They are not. The first is the deploy model. You spin up an agent, point it at an LLM, wire in some tools through MCP, give it a system prompt, and hand
Changelog
Two patches in one day. v2.2.1 went out this morning with a chat stability fix and a set of attachment upgrades. v2.2.2 followed a few hours later with one new feature worth its own paragraph and one critical bug fix that was killing phone numbers on
Product
For two years now, every conversation about AI employees has been bounded by a quiet assumption. You ask, they answer. You assign, they execute. The agent is a coworker, yes, but a coworker who only ever speaks when you speak first. v2.2.0 changes that. This release is bigger