Explore what's new in AI

Codex lets app shots continue a computer-use flow

App shot becomes a practical escape hatch when an app asks you to do something tedious. That turns a screenshot into an execution handoff, not just a note.

Codex adds hotkeys for switching reasoning effort

Keyboard shortcuts for reasoning effort are a small but useful speedup when you keep bouncing between light and heavy agent work. It cuts one more trip through settings.

Codex helps reverse an MP3 player's firmware from a photo and binary dump

The interesting bit is the workflow, Codex guided bootloader access, firmware extraction, binary analysis, and a patched image. That's a concrete reverse-engineering use case.

Linear Asks turns Slack mentions into auto-filed requests

The workflow trims a lot of request intake friction by collecting missing details and choosing the right template automatically. That's exactly the sort of mundane automation teams adopt fast.

Google launches Universal Cart and expands the Universal Commerce Protocol

Google is trying to standardize agentic commerce with a protocol layer and a proactive cart. If this sticks, shopping becomes another agent workflow surface.

Google Workspace expands AI Inbox and shared project workflows

Google is rolling out inbox drafting, instant file access, to-do surfaces, and shared project spaces. This is the kind of bundled enterprise workflow update that can actually move adoption.

Qwen3.7 Max ranks near Claude on Code Arena frontend tasks

The model is now showing up as a serious coding contender, not just a cheap alternative. That matters because frontend work is where agent UX gets judged hardest.

Alook lets one operator run an organized team of AI agents

Alook packages roles, reporting lines, and email coordination into a solo-founder workflow. The interesting part is the organizational model, not just the automation.

Executor previews code mode for React-style generative UI

The preview hints at a more structured way to turn APIs into interfaces, which could cut the blank-canvas tax for frontend work.

Mixedbread adds exact-match grep alongside semantic retrieval

Splitting keyword search from semantic search gives agents a cheaper first pass before they fall back to embeddings. That's a practical retrieval pattern for code and doc search.

LangChain Context Hub stores skills and AGENTS.md files in one place

Context Hub is a clean answer to context sprawl: versioned skills, AGENTS.md, and other files in a shared store. That should make agent setups easier to reuse across repos and teammates.

Gemini for Business adds collaborative projects and workflow agents

Google is pushing Gemini deeper into the enterprise workflow with shared projects and cross-app automation. That looks more actionable than the consumer-side assistant story.

AI SDK 7 canary adds step timing for better agent observability

Step timing, tool execution, TTFT, and token rate are the metrics you need when agents feel slow but you do not know why. This canary makes those bottlenecks visible without extra instrumentation.

Krea 2 lands on fal and ComfyUI with moodboards and style control

Krea 2 is positioning itself as a controllable image foundation model, not just a pretty generator. Native support in fal and ComfyUI makes it easier to slot into real creative pipelines.

Zed v1.4.2 adds reusable skills for agent instructions

Zed's new skills feature lets teams package reusable instructions instead of copy-pasting prompts. That is a clean path toward more consistent agent behavior inside the editor.

Warp sends agent conversations to the cloud when you close the laptop

Warp is making sessions durable enough to survive laptop shutdowns, which is the sort of UX change that turns agents into background workers instead of local toys.

Roboflow's RF-DETR lands in Transformers with SOTA detection

RF-DETR is now in Hugging Face Transformers, with real-time detection and segmentation claims plus demos and docs. That's a practical route for teams shipping vision agents.

OpenClaw 2026.5.26 reduces latency and hardens installs

The latest OpenClaw build focuses on responsiveness, channel reliability, and Windows or Docker install paths. That's the kind of polish that matters when agents live in production.

Claude Marketplace adds partner apps and spend commitment

Anthropic is turning Claude into a distribution channel for partner apps, and letting teams apply existing spend commitments there. Useful if you buy AI tools through one budget.

Cognition says enterprise usage is up 10x after raising over $1B

The Devin team says run-rate revenue hit $492M and enterprise usage jumped 10x, a strong signal that agentic coding has moved from demo to paid rollout.

OpenAI Foundation commits $250M to AI transition support

OpenAI's nonprofit is funding measurement, transition support, and broader prosperity work. That matters because the money is aimed at the labor shock, not just model research.

YouTube launches Ask YouTube for conversational search

Ask YouTube turns search into a structured answer interface that can follow follow-up questions. Engineers should note the retrieval stack now spans long-form video and Shorts.

OpenAI adds a Codex page and developer plugin to the API dashboard

The dashboard is becoming a launchpad, not just an admin panel, with a new Codex page and plugin for starting projects. That's a useful path to lower first-run friction.

Claude Code skills split better into commands and model-invocable skills

The proposal makes skills more composable by separating user-facing commands from model-invocable building blocks. That is a solid pattern for larger agent repos.

OpenCode's /autobrowse skill makes browser workflows 90% faster and cheaper

The team says iterative AutoResearch turned browser skills into a much faster, cheaper workflow. That's a concrete pattern for turning a slow agent loop into a reusable skill.