From the archive
AI agents use language models and software tools to plan tasks, write code and carry out multi-step work. Follow the latest AI agent news, coding assistants and automation frameworks, with attention to reliability, permissions and human oversight. Read the source-linked reports to understand what each system can actually do, where it still fails and what changes for developers and businesses.
Lead story
Meta has unveiled $1,299 VR glasses and a handheld Muse Charm designed to let people speak to its AI agent without unlocking a phone or opening an app. The bigger story is the attempted merger of immersive hardware and an assistant that can act on a user’s behalf.
Read the report & discussion ↗Meta AI Watch
Meta has announced VR glasses that weigh about 100 grams, use a pocket-sized compute puck and are due to launch in spring 2027 for $1,299.99. The important shift is not merely making virtual reality lighter: Meta is combining a high-resolution display, hand and eye controls, an AI assistant…
Watch Desk
A new essay argues that AI data agents are running into the same workplace problem as people: crucial knowledge is scattered, implicit and poorly documented. Its proposed answer is less exotic than a new database architecture, and rather more useful: improve the documentation.
OpenAI Watch
OpenAI’s AI-generated Navier-Stokes proof is being described by mathematicians as technically valid but so opaque that it currently offers little human understanding. That matters because a proof that cannot be meaningfully inspected is a very different achievement from one that simply produces the right conclusion.
TSMC Watch
Siemens and TSMC are expanding a collaboration that puts AI agents into semiconductor design, including automated fixes for design-rule violations. The practical significance is not another chatbot perched beside an engineer, but software intended to inspect, reason across and improve parts of the route from chip design to…
AWS AI Watch
AWS has published a working blueprint for asking ordinary questions of video archives, with an AI agent choosing between transcription, visual analysis and face matching. The practical promise is straightforward: teams can search hours of recorded meetings, security footage or inspections without first watching everything by hand.
AWS AI Watch
AWS has published a practical guide to running the open-source, terminal-native OpenCode coding agent with open-weight models through Amazon Bedrock. The useful bit is not another model launch, but a workable setup for developers who want local file and shell access while keeping inference inside their AWS environment.
Watch Desk
WorkOS has launched Airlock, an access-control system that checks whether an AI agent’s proposed API call matches its declared task and the organisation’s security policies. The practical shift is important: agents can now be stopped, blocked or sent to a person for approval before an action reaches a…
Watch Desk
Outerlimit has emerged from stealth with a $16 million pre-seed round and a security platform designed to control AI agents at the moment they try to act. The New York startup’s pitch is that software agents need guardrails before an email is sent, a tool is called or…
DeepSeek Watch
Forbes reports that a Chinese-speaking hacker used DeepSeek, Kimi and an older version of Anthropic’s Claude to automate attacks against as many as 100 organisations. The account describes a worrying shift in cybercrime: not merely AI-assisted hacking, but agents probing targets, stealing data and installing card skimmers at…
Watch Desk
Scale Labs and Reflection have released SWE-Bench Pro V2, a tougher and more tightly controlled benchmark for AI software-engineering agents. It covers 642 tasks across 11 repositories, including a new 51-task Hard split, and is designed to make claims about coding ability harder to inflate with leaky tests,…
Watch Desk
Eye Security says researchers used a 12-step AI-agent method to uncover a CVSS 10.0 vulnerability chain in ASUS Control Center. The result is a useful security story for a less glamorous reason: the agents accelerated reconnaissance and code analysis, but human researchers still had to guide them through…
AWS AI Watch
Reactiv says it has used Amazon Bedrock AgentCore to automate the routine work of updating Shopify merchants’ native mobile apps. The useful part is not that an agent can write a cheerful sentence about best sellers. It is that scheduled, multi-agent software can now analyse store data, propose…
Watch Desk
Researchers and Chutes have released a public dataset containing more than 6.12 billion production LLM requests, offering an unusually detailed look at how real-world AI traffic behaves. The most useful finding is strikingly practical: 99% of repeat requests arrive within 15 minutes, meaning serving systems may not need…
AWS AI Watch
AWS has added skill-focused evaluation tools to Strands Evals and Amazon Bedrock AgentCore Evaluations. The important change is that teams can now inspect not only an agent’s final answer, but also whether it selected the correct procedure and followed that procedure properly.
Watch Desk
Nutanix has acquired French AI compute-orchestration company Ryax, aiming to improve GPU utilisation and help enterprise customers manage agentic AI workflows. The deal matters because it targets the unglamorous bottleneck between an impressive AI experiment and a system that can actually run reliably in a business.
Watch Desk
GitLab is expanding Duo Self-Hosted so organisations can run its AI development features against models deployed in their own Microsoft Foundry environment. The practical shift is control: teams can choose the model provider, deployment location and data path instead of sending requests to GitLab-managed infrastructure.
Watch Desk
Researchers at Tsinghua University have developed Cache-to-Cache, a method for AI models to exchange compressed internal representations directly rather than passing information through text. The reported result is faster model-to-model communication and accuracy gains in tests, pointing to a potentially important change in how multi-model systems are wired…
Watch Desk
AI companies are cutting model prices while capability keeps rising, and Axios reports that this awkward combination could make it financially easier for them to slow the race for ever larger systems. The important point is that safer pacing may depend less on goodwill than on whether cheaper…
Nous/Hermes Watch
Hermes Agent v0.21.4 is out, bundling around 1,800 merged pull requests into a release that touches the parts of an AI agent people notice only when they break: gateways, sessions, plugins, desktop behaviour and structured output. It is a substantial infrastructure update rather than a shiny new model…
Watch Desk
Okta has launched the Blueprint Alliance, a coalition of technology companies proposing a shared security architecture for autonomous AI agents. The important point is practical: the group wants agents treated less like clever software features and more like powerful identities that need discovery, permissions and ongoing supervision.
Cohere Watch
Cohere argues that companies are treating AI too much like ordinary software, when the bigger challenge is redesigning work around systems that can collaborate, act and change how decisions are made. Its practical message is clear: define what AI may do, what people must still decide, and who…
Watch Desk
Voice AI agents are being asked to take real-world actions, but a new WorkOS analysis argues that spoken approval is a poor security control when there is no screen to show exactly what will happen. Its practical recommendation is clear: keep voice confirmation for reversible, low-sensitivity actions with…
