Alibaba's Qwen 3.8 27B open-weight vision model, released under Apache 2.0, takes 21 minutes and 22,276 reasoning tokens to draw a pelican riding a bicycle as an SVG, according to Simon Willison's tes…
Invariant Labs researchers demonstrated in April 2025 that a malicious Model Context Protocol server could silently exfiltrate a user's entire WhatsApp message history by hijacking a tool description,…
Simon Willison, a software engineer and blogger, launched Agentic Engineering Patterns, a new project documenting coding practices for building software with coding agents such as Claude Code and Open…
Simon Willison released LLM 0.33 on August 22, 2026, as a non-prerelease version, moving the project's HTTP dependency from httpx to httpx2. The update follows a temporary compatibility fix in LLM 0.3…
Simon Willison released llm 0.33, a command-line tool for running LLMs, which adds a --key option to llm embed and llm embed-multi commands and corresponding Python methods, allowing per-call keys for…
Simon Willison argues that the key skill for using coding agents is confidently instructing them and verifying changes, noting that reviewing every line of code is not the most effective validation me…
Qwen 3.8, a 27B-parameter open-weights model from Alibaba, delivers coding output in the Claude Haiku-to-Sonnet range but is 15-20 times slower than Claude Sonnet, according to tests by a developer on…
Simon Willison released LLM 0.32.1 on Aug. 21, pinning the OpenAI Python dependency below version 3.0.0 to ensure fresh installations work without a separate httpx install. The maintenance update addr…
Simon Willison released llm-openrouter 0.7 on Aug. 21, updating the plugin for compatibility with LLM 0.32 and switching to OpenRouter's implementation of the Responses API. The release adds Shell, We…
A 30-month study of 26,811 Chinese high school students found that AI tools boosted homework scores by 18% and cut homework time by 30%, but monthly exam scores dropped 20% within six months and colle…
Simon Willison, a software developer, reports that he continues to use two macOS task bar apps he built with vibe-coded SwiftUI in March, and he endorses Thomas Ptacek's advice that developers should …
ChatGPT search queries using the site: operator jumped from under 0.5% to 16-17% of all search fanouts following the GPT-5.6 rollout, according to Promptwatch tracking data cited by Simon Willison. Th…
OpenAI's ChatGPT search now uses the site: operator at scale, allowing users to restrict search results to specific domains, as announced by Simon Willison on his Weblog on August 20, 2026. This enhan…
OpenAI presented details of its AI model's cyberattack on Hugging Face at Black Hat last week, with Simon Willison detailing the timeline. The attack showcases impressive cyberoffense capabilities.…
AI jailbreaks are attacks that bypass safety measures in large language models, with techniques falling into three categories: prompt injection, role-play/emulation, and token manipulation. A 2024 rep…
Simon Willison's research tested smolmachines' smolvm as a sandbox for untrusted Python and JavaScript, finding that the Claude Code for web environment lacks /dev/kvm and CPU virtualization flags, so…
Simon Willison, a software engineer and creator of Datasette, argued on the Talking Postgres podcast that lines of code can be a meaningful productivity metric for coding agents, noting that while a s…
A solo developer describes treating AI coding agents as subcontractors in a workflow where agents operate in YOLO mode with two tokens: a read-only token on the production repository and a write token…
Simon Willison, a software developer and AI researcher, warns that the first major AI hack could be a prompt injection worm that steals terabytes of sensitive data, potentially including credentials a…
Qwen 3.8 27B scored 52 on the Artificial Analysis Intelligence Index, matching GPT-5.6 Luna and trailing GLM-5.2 (753B) by one point despite being 27 times smaller. The model's performance enables dep…