OpenAI just made its Codex AI coding agent generally available, rolling out a direct Slack integration, enterprise admin tools, and a TypeScript SDK. Engineering teams can now delegate code tasks by tagging @Codex in Slack threads or embed the agent directly into custom CI/CD pipelines. Codex daily usage has already surged over 10x since August.
OpenAI just launched AgentKit, a full-stack toolkit designed to help developers build, deploy, and optimize AI agents. It packs a visual canvas called Agent Builder, an embeddable ChatKit UI, and expanded evaluation tools with reinforcement fine-tuning (RFT). Why it matters: Teams can now drag and drop multi-agent workflows and test them in hours instead of months.
Google DeepMind just unveiled CodeMender, an autonomous AI agent powered by Gemini Deep Think models that diagnoses and patches software vulnerabilities. Over six months, it has already upstreamed 72 security fixes to open-source projects, though human researchers still review every patch before submission.
Google just launched Jules Tools, a lightweight CLI bringing its AI coding agent straight into your terminal, alongside early access to the Jules API. Developers can now wire Jules into Slack bugs or CI/CD pipelines. Google also rolled out persistent memory for coding preferences, a file selector, and environment variable controls.
Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, just unveiled Tinker, an automated tool designed to fine-tune frontier open-source models. Developers can use a few lines of code to run supervised fine-tuning or reinforcement learning on Meta's Llama and Alibaba's Qwen, bypassing messy GPU cluster management entirely. Access is open via application for free during beta.
Microsoft just launched the Microsoft Agent Framework in public preview, bringing together research from AutoGen and the enterprise plumbing of Semantic Kernel into a single open-source SDK. It adds native support for Anthropic's Model Context Protocol (MCP) and multi-agent workflows across Azure. Why it matters: Developers no longer have to juggle fragmented tools to build and monitor collaborating AI agents.
Wikimedia Deutschland just launched an AI-friendly vector database to make 30 million Wikidata entries easily digestible for large language models. Built using Jina AI's embedding model and hosted by DataStax, the project turns clunky structured data into rich contextual vectors. The goal is to level the playing field for smaller developers outside Big Tech so their AI agents can easily surface niche facts.
Nothing just rolled out Playground and Essential Apps for Nothing Phone 3, letting users build and share custom apps, widgets, and back-display animations entirely through prompts. The feature is launching in alpha on Nothing OS 4, and developers can export the generated code to keep tweaking. It is an early step toward Nothing's goal of an AI-focused operating system.
A comprehensive guide breaks down how Factory Droid shifts software development from single-prompt code edits to coordinated agent teams. By pairing a research-first Spec Mode with native Git worktrees and milestone-driven Missions, Factory focuses on multi-file refactors and migrations that quickly overwhelm standard coding assistants.
Anthropic just released Claude Sonnet 4.5, establishing a new high watermark for coding and agent workflows. The model hits 77.2% on SWE-bench Verified and jumps to 61.4% on the OSWorld computer-use benchmark, while keeping API pricing flat at $3/$15 per million tokens. Alongside the release, Anthropic opened up its Claude Agent SDK to let developers build custom autonomous agents.
Anthropic just launched context editing and a file-based memory tool in public beta for the Claude Developer Platform, Amazon Bedrock, and Vertex AI. Paired with Claude Sonnet 4.5, context editing automatically prunes stale tool results to slash token usage by up to 84%, while the memory tool stores key learnings across sessions. Anthropic reports a 39% performance leap on complex multi-step workflows.
Factory just raised $50 million in Series B funding while fully rolling out Droids, autonomous software development agents that topped the Terminal Bench leaderboard. Unlike tools locked to a specific IDE, Droids work across CLI, Slack, and browsers to handle refactoring, reviews, and incident response. Why it matters: It shifts AI engineering from autocomplete to full-lifecycle autonomy without breaking existing workflows.
Perplexity just launched its Search API, giving developers direct access to the web-scale infrastructure powering its AI answer engine. Built specifically for AI workloads, it indexes hundreds of billions of pages and delivers pre-ranked, fine-grained snippets instead of raw links. The rollout also includes a dedicated SDK and an open-source evaluation benchmark.
Alibaba just released its new flagship model, Qwen3-Max, shipping both a standard Instruct version and a tool-using Thinking variant. The company heavily boosted coding and agentic capabilities, rivaling elite frontier models across tough benchmarks like SWE-Bench and AIME25. It is already live on their chat platform and available through Alibaba Cloud's API.
DeepSeek just launched DeepSeek-V3.1-Terminus, an update taking aim at key user feedback. The release cleans up language consistency by curbing random characters and awkward Chinese-English mix-ups, while packing stronger performance into its Code and Search Agents.
Meta just overhauled Horizon Worlds with a new engine for sharper visuals and launched Horizon Studio for creators. While devs can already generate textures and audio, an in-editor AI assistant is arriving later this year to let them build worlds and tweak NPC personalities through simple chat.
Google DeepMind just proved its AI can beat almost all elite college coders under tournament conditions. Competing remotely in the 2025 ICPC World Finals, an advanced version of Gemini 2.5 Deep Think solved 10 of 12 problems in 5 hours, ranking second overall and cracking one problem no human team solved. Why it matters: competitive coding requires real reasoning and novel synthesis, proving AI is moving far beyond mere pattern matching.
GitHub just launched the GitHub MCP Registry, creating a central home base to discover and install MCP servers without digging through random repos. It ranks servers by GitHub stars and activity, plugs directly into VS Code with one-click installs, and works across Copilot and other agentic tools. Why it matters: Connecting AI agents to dev tools was messy and fragmented, and GitHub just made finding trusted integrations frictionless.
Sakana AI just unveiled an agentic framework and a new benchmark called robust-kbench to automate low-level CUDA kernel engineering. The system translates PyTorch code into raw CUDA kernels and uses an evolutionary meta-generation loop with LLM verifiers to iteratively boost runtime speed. Why it matters: Squeezing maximum performance out of GPUs is notoriously tedious, and handing kernel optimization over to AI agents could save developers months of manual tuning.
OpenAI just overhauled Codex with a specialized model fine-tuned for real-world software engineering. The agent dynamically adjusts its thinking time, can run autonomously for over seven hours on massive refactors, and automatically reviews GitHub PRs for bugs. Why it matters: AI coding is shifting from line-by-line autocomplete into an independent teammate that tackles complex tasks end-to-end.
