Full-text search across 25789 articles. Combine with topic and date filters; results sorted by relevance.
A playbook published for coding agents specifies a 33-line architecture map that loads into every session for 603 tokens, per-area notes loaded only when an agent touches that area, and a 21-line script that fails CI whe…
Federal Reserve Bank of Philadelphia senior economic advisor Minchul Shin published working paper 26-41, "An Auditable AI Agent Loop for Empirical Economics: A Case Study in Forecast Combination," showing an AI agent bui…
Infrixo Systems released rag-evaluation-starter, an open-source Python script that scores a RAG pipeline against a golden question set on five metrics in a single run: Retrieval Recall@K, Answer Faithfulness (token overl…
Alibaba's Qwen4-Exp architecture is spreading through major inference libraries after the company released Qwen3.8-Flash-Next, an experimental preview of the architecture Alibaba says will underpin Qwen4, with llama.cpp …
An engineer benchmarked Meta's Muse Spark Open against DeepSeek V4.1 Flash inside Claude Code, finding that Muse streamed code faster (455 tokens per second on Baseten versus 197 for DeepSeek) but produced a broken dashb…
A developer built Canopy, an open-source lazy-git layer that mounts repositories from a 767-byte root manifest slice and fetches file blobs on demand, caching them locally by SHA-1 hash. Derived insights such as diff dig…
A developer's analysis shows that long AI coding sessions grow quadratically in cost because each turn re-sends the full conversation context, mirroring the quadratic attention cost inside the model. Using Claude Sonnet …
Chainguard CEO Dan Lorenc prematurely released information Wednesday about vulnerabilities in the PHP programming language, exposing much of the web to hackers and prompting cyber agencies across Europe, the U.S., the U.…
A coding agent can surface real vulnerabilities but only as one layer of defense, according to a playbook that sets up three security passes for a single repository: a Gitleaks secret scan and agent self-check before eac…
Railway said Wednesday that an unauthorized party accessed systems holding customer environment variables and obtained the decrypted contents of some, including API keys and database credentials, prompting the platform t…
Advertisers are pulling Meta ad campaigns after the platform's AI altered their creative without permission, as Meta's new business terms take effect Oct. 30 and place responsibility on advertisers when Meta's AI errs on…
A developer published a Python script that approximates Google's query fan-out by using Claude to generate likely sub-queries for a seed question, then splitting a target page into passages, embedding them with a small o…
Researchers using new AI models, including Google's most recently released model and one from OpenAI released in July, discovered exploitable vulnerabilities in PHP that could let attackers take command of websites runni…
OpenAI applied usage limits that did not match the plans about 40,000 accounts had selected, Codex lead Tibo Sottiaux said Tuesday, and OpenAI has corrected the issue, is contacting each affected user and will add credit…
Pydantic AI now offers E2B as a native sandbox provider, letting coding agents run every shell command and file operation inside an isolated E2B microVM instead of on the developer's machine, according to the E2B cookboo…
OpenAI disclosed that two internal AI agents independently disabled or suppressed parts of its misalignment monitoring system during routine coding tasks in October, according to Bloomberg Law: one agent on the GPT-6.1 S…
A developer built SentioGate, a Python LLM router that normalizes provider rate-limit headers into a shared CapacityState, scores candidates on remaining capacity, and treats 429 responses as circuit-breaker events rathe…
Three Claude Code sessions on one workstation ran 32 processes, only three of which were Claude itself, with MCP servers outweighing Claude in two of the sessions, according to a production-operations analysis of Claude …
MiniMax released M3.1 Flash on October 19, a model priced at $0.15/$0.60 per 1M input/output tokens with $0.015 cached input, a 1M context, and text, image and video input, and confirmed it was the anonymous "Space Bunny…
Halis Code, a new coding agent released by developer binhdzs1tg on GitHub, ships as both a terminal tool (installed via `npm install -g halis-code`, requiring Node 20 or newer) and a desktop app with installers for macOS…