Explore what's new in AI

Trimmy 0.8 trims multi-line shell commands before you paste them

It folds wrapped shell snippets into one clean line, which cuts copy-paste mistakes without a daemon, network, or Dock icon.

TheTuringPost curates 10 agent memory resources

The list is a solid bookmark for memory, long-term context, and security papers, plus a few videos and surveys.

University of Warwick's RAVEN validates 118 exoplanets from TESS data

The model found 118 confirmed planets and 2,000 plus candidates, showing AI can chew through astronomical backlogs at scale.

OpenAI's o1 reportedly beats ER physicians at early diagnosis

A Science study says the model outperforms doctors in messy triage cases, which makes early diagnostic support look more practical.

AAFLOW speeds agentic workflows with a zero-copy Arrow data plane

The paper says most agent latency comes from serialization and coordination, and reports up to 4.64x faster pipelines.

thdxr says opencode feels like progressive rendering with voice prompting

The workflow shifts from layer-by-layer building to repeated full passes, which can make voice-led agent coding feel much smoother.

AI Engineer Europe workshop treats agentic search as context engineering

The workshop maps when to use shell tools, semantic search, or general queries, which is the core retrieval choice for agents.

trq212's Claude Agent SDK guide pushes HTML for planning and review

The guide argues HTML is a better medium for planning, reports, and reviews because agents can preserve structure while editing.

HTML to Markdown conversion can cut token counts by 92 percent

Converting HTML to Markdown can slash token counts, a huge win for crawlers, RAG pipelines, and agent context budgets.

Hyperbrowser CLI launches sandboxes and browsers from the terminal

It bundles browser automation, snapshots, and sandbox control into a single CLI, aimed at faster end-to-end agent loops.

pvncher says the coding-agent cwd can make or break a session

Starting an agent in a mixed root directory confuses it fast, so choosing the right working tree is a high-leverage setup step.

OBLIQ-Bench gives search agents a newer test set than MS MARCO

The benchmark pushes retrieval systems away from stale corpora, so search evals should get a lot more honest in 2026.

Meta's patch review template lifts verification to 93 percent

A structured template raised patch verification from 78% to 93% without new training, showing how process can beat model size.

fal adds HiDream-O1-Image for pixel-level image editing

Pixel-level conditioning makes it better for layouts, multilingual copy, and reference-sensitive edits than plain text-to-image prompting.

Hugging Face launches hf-sandbox for disposable code execution

A disposable sandbox lowers the friction of running untrusted code and gives agents a cleaner place to test terminal workflows.

Apple's Reinforced Agent reviews tool calls before they execute

Apple moves review into the execution loop, so a reviewer agent can fix bad tool calls before they ship.

BlackBar puts Blacksmith job and vCPU status in the macOS menu bar

It surfaces live job, vCPU, and API status without a dashboard, which is exactly what infra operators want during load spikes.

CodexBar 0.25 adds providers, quota warnings, and cost history

The menu bar now tracks quota, cost history, and account switching across providers, which makes multi-model spend easier to control.

Steipete teaches Codex to use social signals when reviewing PRs

The reviewer caught plugin dispatch and Windows portability bugs, showing how maintainer signals can sharpen agent reviews.

gogcli 0.16 adds Google Workspace admin commands in one binary

Workspace admins can create users, aliases, and org units from one binary, turning Google's APIs into a scriptable ops surface.