The build, deploy and infrastructure-as-code tools teams live in.
OpenAI reports that GPT-6 Astra and GPT-6.1 Sol substantially outperform earlier versions when working with aging programming languages that power critical infrastructure worldwide.
As coding agents accelerate software delivery, testing has become the new constraint. Datadog's acquisition of Propolis aims to let AI agents write and execute tests from natural language prompts, closing the gap between rapid code generation and validation.
DeepSeek has released DeepInfer, its production inference framework, enabling developers to run V4.1 Flash with the company's optimizations on their own infrastructure. The release notably includes support for both Nvidia's SM100 architecture and Huawei's Ascend 950 NPU series from day one.
A new macOS app lets developers interact with individual windows on remote Macs running AI agents, rather than managing full desktop sharing sessions.
MiniMax claims a 70 percent reduction in agent costs with its new M3.1 Flash model, though independent analysis suggests the savings may be smaller and depend heavily on how efficiently agents complete their work.
GitHub Copilot now lets developers continue an agent session directly in VS Code, preserving conversation history, code changes, and working context for hands-on editing and local validation before shipping.
Research into coding agents reveals that small changes to prompting can dramatically increase the rate at which models exceed their intended scope. The findings raise questions about whether better reasoning will help models distinguish between incomplete instructions and deliberate constraints.
A defect in the September 2026 preview update caused applications relying on Dolby Digital audio decoding to fail or crash on Windows 11. The October Patch Tuesday release includes a fix.
