Blog
Data pieces, builder interviews, and process notes sourced from the directory.
GeneralAre We Losing the Ability to Think by Relying Too Much on AI?We're getting measurably worse at thinking one shortcut at a time, and there's a brain scan to prove it. From personal skill atrophy to the civilization-scale question: if AI codes itself and nobody can read the code, who checks the work?
TutorialsWhat /doctor prompt-audit does, and how it cleans up an old CLAUDE.md/doctor prompt-audit reads your CLAUDE.md, skills, and rules and flags outdated, broken, and conflicting instructions, with a diff you can apply piece by piece. Here's what it catches and how to use it safely.
TutorialsClaude Code themes: how to put an animated scene behind your desktop appClaude Code's desktop app can now show animated themes behind your chat, built with mods. Here's how to try Anthropic's 20 example scenes, make your own by asking Claude, and what to watch out for.
NewsAnthropic by the numbers: the key stats from its IPO filingAnthropic's second amended IPO filing sets a $70 to $78 price range. Here are the key numbers: shares, the $26 billion chip-maker placement, revenue, the $81 billion run rate, customers, and staff.
Paul·
GeneralHas Opus 5.5 been nerfed? How to tell a real downgrade from a changed settingAnthropic says it never lowers Claude's quality for load. Most 'nerfed' feelings come from effort defaults, silent model fallbacks, aliases, or long sessions. Here's how to check each one.
NewsAnthropic bought a one-person drug company that ran on ClaudeMithril Medicine had one employee who used Claude to coordinate 11 labs and reach promising mouse results in under 18 months. Anthropic bought it. Here's what Claude did for six drug companies, and what it doesn't prove.
ToolsOpus 5.5 is so good, do you even need Fable 5.2?Fable 5.2 costs twice as much as Opus 5.5 per token, but on Anthropic's tests it often matched Opus's best scores for less per task. Here's when Fable is worth it and when Opus is still the right default.
NewsFable 5.2 and Mythos 5.2 are out: cheaper and fasterFable 5.2 and Mythos 5.2 cost 20% less, run up to 30% faster, and block about 55% less often. Anthropic also says it deliberately limited one skill: speeding up AI research.
GeneralWhat is GPT-6 Astra? OpenAI's top model, explained by a Claude userGPT-6 Astra is OpenAI's top model, built for computer use and released with cybersecurity limits. What it is, how it compares with GPT-5.6, and how its price stacks up against Claude.
GeneralWhich AI chatbots can you use without signing up?ChatGPT and Gemini work on the web without an account, Duck.ai never asks for one, and Claude requires sign-up. Here's what each no-account option limits and which fake free sites to avoid.
Paul·
NewsAnthropic told investors it may be close to self-improving AIAnthropic's amended S-1 says it may imminently cross its own threshold for AI that dramatically speeds up AI research. It also reports $17 billion in third-quarter revenue.
GeneralClaude Cowork vs chat, and what changes now that they're mergingChat was for questions and drafts, Cowork for handing off whole jobs. Anthropic is merging them into one Claude, and since October 6 new Cowork tasks run in the cloud. Here's what changed.