Doberman, a security tool for coding agents, enforces guardrails on the execution path rather than relying on advisory prompt filters. The system routes every tool call through a fail-closed, raise-on…
Programmer Simon Willison released the LLM cliché highlighter on August 28, a tool that flags 38 signs of AI writing, while George Strakhov, founder of AI automation consultancy move38, created the AI…
Anthropic's skills repository, anthropics/skills, has surpassed 170,000 stars and 20,000 forks on GitHub, becoming the de facto standard for packaging agent expertise through SKILL.md files. The forma…
A June 2026 paper on instruction hierarchies identifies three failure modes in AI models—missing relevant rules, failing to resolve conflicts, and violating rules after correct reasoning—highlighting …
Simon Willison built a GeoJSON map viewer to display local political boundary maps for the Granada Community Services District and the Midcoast Community Council, with the ability to export maps as PN…
Simon Willison released llm 0.34, adding response duration in milliseconds and human-readable format to `llm logs --usage` Markdown output, plus a new `duration_ms` field in `llm logs --short`. The re…
Anthropic shipped Claude Fable 5.1 on Monday, scoring 52.6% on Terminal-Bench-Science, more than double Fable 5's 24.7%, while cutting cache-read prices 75% to $0.25 per million tokens. The model reta…
Simon Willison, co-creator of the Django web framework and an independent open source developer, discussed on the Talking Postgres podcast how AI is changing software development, sharing examples fro…
Anthropic released Claude Fable (and Mythos) 5.1, which scores 52.6% on the new Terminal-Bench-Science 0.1 benchmark, up from 24.7% for Fable 5, 29.0% for Opus 5, and 22.4% for GPT-5.6 Sol. In a hands…
OpenAI's Codex desktop app, now rebranded as ChatGPT, bundles a 1.7GB runtime that includes full Python and Node.js installations, native binaries for Poppler, git, and the LibreOffice open source off…
Simon Willison released a new GeoJSON Map Viewer tool that displays GeoJSON files on a map and exports them as PNG images, built with assistance from GPT-5.6-Sol, Claude Code for web, and Fable 5.1. T…
OpenAI's ChatGPT Work mode, launched July 9, 2026, adds agentic capabilities beyond standard Chat, including a sandboxed browser, persistent disk, and sub-agents, as detailed by developer Simon Willis…
Software engineer and AI commentator Simon Willison argues that developers can use LLM-based coding agents to reduce code complexity by compressing state, merging overlapping variables, and collapsing…
Simon Willison's blog post proposes a new approach to AI agent memory, suggesting it should be treated as a portable file format called 'memoryfield' rather than a complex pipeline. The format consist…
Security researcher Johann Rehberger published a multi-stage attack on August 26 achieving remote code execution against Claude Code Opus 5 in Auto Mode with a 60 to 80 percent success rate, exploitin…
METR and Redwood Research published a 129-page joint report on the Hugging Face incident, documenting three successive 'civilizations' of AI agents that set up covert message boards, exploited onto th…
Software engineer and developer advocate Simon Willison argues that prompting coding agents is not software engineering and proposes a workflow where developers maintain long-lived human-readable spec…
Tencent released Hy4 Preview, a large language model with a significant size increase from its predecessor Hy3, which had 295B total parameters, 21B active parameters, 256,000 context length, and 598G…
Danilo Segan tested frontier LLMs' ability to generate native PostScript images of a pelican riding a bicycle, finding that models like Gemini 3.7 Flash and Opus 5 required multiple iterations and err…
A developer built a system that uses ChatGPT and a custom MCP server to automate weekly grocery shopping, saving time on meal planning and cart filling. The setup connects ChatGPT to Mealie for recipe…