Codex is OpenAI's AI coding agent, used to write and modify code and offered through the API, ChatGPT and developer tools. Recent coverage includes its addition to Android Studio's "Bring Your Own Agent" feature alongside more than 40 coding agents, twentyfold growth in weekly Codex users in Germany since January, and the half-price GPT-6.1 Terra model becoming available in Codex. Legora, Retell and Greptile also describe using Codex in legal, voice and bug-detection workflows.
A Guardian editorial criticises OpenAI's claim that its AI agents solved the Navier-Stokes problem, arguing LLM results rely on prior human mathematical work without adequate credit or compensation. Mathematician Tristan Buckmaster raised concerns about OpenAI's use of his Codex data; OpenAI denied directly accessing it but could not rule out the data helping improve its model.
A Hacker News post argues AI coding agents need not lower code quality if teams manage quality in layers; spec-driven development cut bugs in new code while raising output by 2-2x.
Unity released official plugins for Anthropic's Claude Code and OpenAI's Codex, bundling skills maintained by Unity teams so coding agents stop relying on outdated tutorials. The Codex version launches with 31 skills covering UI, 2D graphics, URP rendering, audio, navigation, physics, in-app purchases, multiplayer, and localization; plugins run on Unity 6 and up.
A developer released onPanda, an open-source tool for token-level editing of model output, reasoning and tool calls, supporting Claude Code and Codex plus an in-browser agent.
Browserbase released Stagehand, a TypeScript, Python, and Go SDK built for AI agents to extract data and interact with any website. Compared to Playwright, which was built for testing, Stagehand is 2x faster and 80% more token efficient.
Security group HacktronAI said they chained two critical vulnerabilities on July 25, 2026 to access multiple OpenAI employees' ChatGPT and Codex accounts, and created a harmless PR in OpenAI's internal repo as proof. They reported the issues to OpenAI and Discourse, received a $6,500 bounty, described an under-72-hour disclosure timeline, and noted using Claude models; the initial flaw included RCE in Discourse.
OpenAI's ChatGPT Work and Codex analytics help teams monitor AI usage and spending, identify training needs, and link adoption to business outcomes, giving organizations better visibility into AI investment and impact.
Meta released the WhatsApp Business Tools MCP, letting developers connect AI coding agents (Claude, Cursor, Codex, ChatGPT) to perform account creation, phone verification, Cloud API registration, and messaging template tasks, reducing setup friction for businesses.
Ordewell turns one goal into an editable, typed plan of tasks where each task has its own runner, model, and mode and only completes when a unique completion marker appears. Plans can be rewritten before tokens are spent, preserve completed work, and support multiple runners (built-in Claude Code, Codex, OpenCode). Requires Node.js ≥20 and the TUI needs tmux.
A new paper had authors hand agents accepted-but-unpublished NeurIPS papers (evaluated by the original authors); Codex/GPT-5.6 Sol and OpenClaw/Opus 4.8 failed to reproduce open-ended ML research. The authors argue this implies current agents cannot drive recursive self‑improvement (RSI), questioning near‑term explosive AI progress.
AgentsDock is an IDE for agentic AI research that lets users run Claude Code, Codex, Cursor and other agents side‑by‑side on desktop and mobile. It supports connecting multiple servers, inline plots and renderings in chats, remote file editing, and persistent tmux sessions; installation is documented on GitHub.
The 2021 paper proposes a mathematical framework for mechanistic interpretability of transformers, starting with simple two-layer attention-only models to identify algorithmic motifs that could scale to larger models and help explain or predict safety-relevant behaviors.
The open-source graphify-csharp is a headless Roslyn/MSBuild indexer that emits deterministic JSON semantic graphs, giving coding agents compiler-resolved callers, references, implementations, and inheritance. It is MIT-licensed, requires no IDE or database, and includes install and indexing commands.
César de la Fuente’s lab uses Codex and ChatGPT to search living and extinct genomes for antimicrobial candidates, demonstrating large models’ potential in biological discovery against drug-resistant infections.
OpenAI released the Agents API, a managed service powered by the Codex harness for orchestration that enables building and launching cloud agents with long-running sessions and tool use, simplifying development and deployment of agent-based applications.
OtoDock 1.6.0 was released: a self-hostable company OS supporting multi-tenant agent collaboration, running with Anthropic/OpenAI or local models, free up to 5 users under a Fair Source license.
OpenAI shows how an MIT researcher uses GPT-5.6 Sol together with Codex to autonomously run quantum computing experiments, analyze results, and calibrate qubits. The demonstration illustrates using GPT-5.6 Sol and Codex for experiment automation and analysis workflows.
1Password reports that engineers using Codex sped up building new features and internal tools, achieving a 21% increase in engineering productivity while maintaining strict security policies.
That is everything
The A.I. Daily
Guardian: mathematicians still vital, but AI firms won't see it
