Explore what's new in AI

OpenAI Robotics turns world simulation into a full robotics hiring push

OpenAI is staffing hardware, ops, systems, and ML around robotics, so the world-simulation work now looks like a real product bet.

DeepSeek V4 Flash helps a coding agent improve itself

One coder says DeepSeek V4 Flash is strong enough to improve a coding agent with a minimal harness and a lot of tokens.

TheTuringPost maps 20 advanced RAG types for 2026

This RAG taxonomy is a handy 2026 bookmark, especially if you need a map of agentic, memory, and hybrid retrieval patterns.

Terence Tao says AI lowers the barrier to math research

Terence Tao's point is simple, AI tools are lowering the barrier to real math contribution much earlier in a career.

AI CLI adds multi-image inputs for style transfer and comparisons

Multi-image inputs make AI CLI more useful for side-by-side comparisons, style transfer, and image-driven edits from the shell.

LangSmith on AWS shows how to evaluate deep agents by their trajectories

Trajectory checks, offline pytest evals, and online monitoring give deep agents a better chance of being reliable in production.

GitHub CLI's device flow is a better pattern for SSH-first login

Device flow is the right answer for SSH or headless CLI login, because polling beats a localhost callback no one can click.

Conductor treats agent transcripts as first-class exports

For agent work, the transcript is an artifact, not a log file, and Conductor's export workflow treats it that way.

Claude Code's Dynamic Workflows expose token loops and harness gaps

Long-horizon workflows expose token waste, infinite loops, and handoff bugs fast, which is why you want a custom harness in control.

Cursor's Composer 2 turns production traces into new benchmark tasks

CursorBench pulls hard tasks from real engineering sessions, which is a better proxy for production coding than tiny public tasks.

OBLIQ-Bench targets retrieval models that overfit stale benchmarks

OBLIQ-Bench is meant for retrieval systems facing fresh, implicit queries, not another saturated leaderboard with old data.

Guillermo Rauch says coding agents are pulling CEOs back into shipping code

Rauch's point is that coding agents make the stack legible across the company, so even CEOs are getting hands-on again.

Box says enterprise agents fail when context is fragmented

Enterprise agents keep breaking on context, because the real problem is fragmented systems, missing permissions, and tribal knowledge.

MIT study finds AI saves far less time than people expect

People expect AI to save far more time than it does, and that optimism itself makes repeated use more likely on trivial tasks.

ChatGPT and peers ace fresh news quizzes, then miss retrieval edges

The models can handle clean multiple-choice news questions, but the failure mode is still retrieval, wrong article, wrong language, wrong scope.

Hyperbrowser lets coding agents extract a site's design system from the terminal

The `/design` flow turns a site into brand tokens and component cues from the terminal, which speeds up frontend reverse-engineering.

duetchat adds Kanban boards for long-running agents

Kanban turns long-running agents into something operators can actually track, especially when tasks stretch across days or weeks.

Microsoft's VS Code Learn Series teaches agent tools and extensions

Microsoft's learn series is a practical on-ramp for tool use, agents, and third-party extensions in VS Code, not just high-level docs.

Microsoft's Copilot super app leak shows Coding and Cowork tabs

The leak suggests Microsoft is merging coding and coworking into one shell, with Scout and separate workspaces instead of one generic chat.

Claude Code 2.1.159 ships as an internal-infra-only release

This build only moves backend plumbing, but it still matters for reliability and rollout speed when you depend on Claude Code daily.

Hermes Agent's 100-plus skills spark a modularity versus bloat debate

Users like the ready-made setup, but the preloaded skill explosion shows how quickly agent UX turns into clutter and hard-to-audit behavior.