Full-text search across 67659 articles. Combine with topic and date filters; results sorted by relevance.
AI agent reliability depends less on the underlying large language model than on the surrounding "harness" of rules, tools, workflows, memory and guardrails, according to an analysis that cites Cole Medin's video using L…
Harness launched VegaLoad, a free, open-source load testing tool released under the Apache 2.0 license and built by the Harness Resilience Testing team, which lets AI agents write, run and report load tests. VegaLoad shi…
A developer described a lightweight standard-library Python script, claim_check.py, that uses git grep to verify every hard claim in AI-drafted documentation — environment variables, CLI flags, repo paths and version str…
A developer has declared AIPass, an open-source framework that turns single Claude Code sessions into a persistent team of 18 agents, feature-complete, shifting the project's focus to fixes and polish. Each agent is a di…
SageOx pitches a one-command setup that gives every teammate's Claude and Claude Code session access to shared team meeting context, contrasting it with a Granola-plus-Notion route the article says requires 11 steps and …
Wiz built a supply chain detection pipeline that analyzes new npm package versions at publication, using static analysis, an AI classifier and sandboxed dynamic analysis to flag malicious packages and add detection rules…
MiniMax released MiniMax M3.1 Flash, a next-generation model priced at $0.15 per million input tokens, $0.60 per million output tokens and $0.015 per million cached input tokens, with overall agentic usage costs 70% lowe…
Anthropic replaced git with a new source-control system called Scion after its daily merged code changes grew from about 2,500 in May 2026 to over 30,000 by September 2026, mostly driven by agents, according to Anthropic…
A developer building the Deployka hosting service added a pre-build scan called Deploy Doctor that checks AI-generated projects for common first-deploy blockers, including missing start scripts, hard-coded or 127.0.0.1-b…
A non-astronomer writing under the byline of a personal blog reported that Claude, Anthropic's AI assistant, helped surface an unreported 0.21% brightness dip in the star TIC 402710280, 218 light-years away in Triangulum…
Chrome DevTools MCP server 1.9, released September 8, added a --no-javascript-evaluation flag that lets AI coding agents inspect live production pages without executing scripts, alongside support for the cross-vendor Age…
A developer built GitView CodeAudit, a local security-review tool for Go backends that pairs 18 deterministic static-analysis rules with an open-weight Gemma model running on Ollama to explain findings without ever decid…
JetBrains launched the free Early Access Program for Air this week, bringing multi-agent orchestration to IntelliJ IDEA, PyCharm, WebStorm, CLion, and RustRover. Air runs Claude Agent, Codex, GitHub Copilot, Gemini, and …
NVIDIA's Nemotron-3.5 Bolt 74B-A9B, a vision-language mixture-of-experts model with 74 billion total parameters and 9 billion active per token, runs locally through llama.cpp at roughly 22 GB after 2-bit quantization-awa…
A developer built LedgerShield, an open-source, privacy-first transaction reconciliation assistant that runs entirely on a bookkeeper's local machine using an embedded SQLite database and the open-weight Google Gemma 2 m…
NVIDIA's RTX Spark combines RTX graphics, the full NVIDIA CUDA and AI software stack, and up to 128GB of unified memory shared between Grace CPU cores and a Blackwell RTX GPU to run local AI workloads on Windows laptops,…
Smart Creations released Muslim Prayer Reminder MCP v1.0.1, a production-ready astronomical prayer calculation and proactive notification engine built on Cloudflare Workers for AI models, coding agents, and developer env…
Google's Antigravity Python SDK added local model inference on September 23, with Google's own benchmark showing 97.2% of roughly 3,400 tokens ran on the developer's machine and only 95 touched the cloud. The update ship…
A Prime Intellect residency project found that GPT 6 Luna xhigh scored roughly the same on the SWE-Multimodal v2 benchmark whether it received native images, OCR text, or no images at all, passing about 175-178 of 480 ta…
Meta released its first frontier models, Muse Wonder and Muse Wonder Full, via the Meta Model API and Muse Code, pricing Wonder Full at $4 per million input tokens and $20 per million output tokens and claiming parity wi…