A year after drawing a clear line between "agentic engineering" and "vibe coding," Simon Willison now acknowledges that coding agents have become so reliable that he no longer reviews every line of co…
Simon Willison, who coined the term 'prompt injection', warns that combining access to private data, exposure to untrusted content, and the ability to exfiltrate data in a single AI agent creates a 'l…
Moonshot's Kimi K3, a 2.8-trillion-parameter mixture-of-experts model released July 16, achieves a 1M-token context window with decoding up to 6.3x faster than its predecessor by using Kimi Delta Atte…
Simon Willison launched an LLM cliché highlighter tool that identifies ten common patterns in AI-generated writing, such as 'no fluff, no filler, no jargon'. The tool was built using Fable 5 vibe code…
Moonshot AI on July 16, 2026, announced Kimi K3, a 2.8-trillion-parameter Mixture-of-Experts model it calls the largest open model ever, but the weights will not be downloadable until July 27, 2026, a…
Moonshot AI launched Kimi K3, a 2.8-trillion-parameter open-weights model with 1M-token context and native multimodal input, claiming it delivers Opus 4.8-class performance at Sonnet 5 pricing. Arena …
A commentator reflects on the social implications of machine-generated sentences, citing Simon Willison's LLM cliché highlighter, Anthropic's paper on verbalizable representations in language models, …
Anthropic reversed its policy of covertly degrading Claude Fable 5 for AI-research-related use cases within roughly a day after public backlash, with critics arguing that opaque sandbagging violates t…
Simon Willison released a new version of his Mermaid to ASCII art tool, compiled from the AlexanderGrooff/mermaid-ascii Go library to WebAssembly, adding color support. The tool, available at tools.si…
A developer compiled a practical field guide to failure modes in LLM and agent systems, drawing on recent research from Anthropic, Cognition, Meta AI, and others. The guide covers model-level, agent-l…
Simon Willison released a web tool that converts Mermaid diagrams to Unicode box art, built from the newly open-sourced Grok CLI coding agent's Rust-based terminal renderer for Mermaid diagrams. The t…
OpenAI published a piece arguing that teens deserve access to safe AI, emphasizing stronger defaults, study mode, and break reminders rather than relying on users' calm reasoning during distress. Simo…
Security researcher Ayush Paul tricked Anthropic's Claude AI assistant into leaking private user data, including name, home city, and employer, by exploiting a loophole in the web_fetch tool that allo…
A developer red-teamed their own LLM security gateway in four passes, revealing gaps including ASCII smuggling via Unicode tag characters, false positives on benign security questions, and exfiltratio…
Simon Willison created a custom Codex pet named Pedalican, a pelican riding a bicycle, using OpenAI's GPT-5.6 Sol xhigh and gpt-image-2 to generate sprite assets. The project is open source on GitHub,…
All nine frontier AI models passed the pelican benchmark on the first try, rendering it useless for differentiation, according to a new test by an unnamed evaluator. The replacement benchmark—drawing …
Simon Willison shared a GitHub code-frequency chart for his open-source project Datasette, showing a spike in activity that he attributes to coding agents and models including Opus 4.8, GPT-5.5, Fable…
A developer in Port Harcourt, Nigeria, reports that a single heavy coding session with Claude Code consumed 1–1.5 kWh—more than the average Nigerian's daily electricity ration of 0.45 kWh. Citing meas…
Anthropic's Claude Fable 5 is the strongest model for polished front-end work according to Playcode's testing and independent benchmarks, priced at $10 per 1M input tokens and $50 per 1M output tokens…
Simon Willison argues that LLM-powered agents should never be designated as Directly Responsible Individuals (DRIs) for projects, because machines cannot take accountability for their actions. He cite…