<?xml version="1.0" encoding="UTF-8"?><?xml-stylesheet href="/rss.xsl" type="text/xsl"?><rss version="2.0"><channel><title>WTFisAI</title><description>News on AI, agents and the companies building them.</description><link>https://wtfisai.blog/</link><language>en-us</language><item><title>Muse Spark Open weighs 1,644 GB, more than 2 Macs hold</title><link>https://wtfisai.blog/muse-spark-open-mac/</link><guid isPermaLink="true">https://wtfisai.blog/muse-spark-open-mac/</guid><description>Meta says its new downloadable model runs at home on 2 Apple Mac Studios. The files it published are 1,644 GB, the 2 machines hold about 1,024 GB, and Apple ships the big one late October.</description><pubDate>Fri, 16 Oct 2026 13:00:00 GMT</pubDate><category>AI</category><category>Meta</category><category>Muse Spark Open</category><category>open weight models</category><category>Apple</category><author>Nicolas Lecocq</author></item><item><title>OpenCode vs Claude Code, and the version you really get</title><link>https://wtfisai.blog/opencode-vs-claude-code/</link><guid isPermaLink="true">https://wtfisai.blog/opencode-vs-claude-code/</guid><description>OpenCode 2 was announced this week and installing OpenCode today still gives you 1.19.0. Here are 5 checks against Claude Code, with both pricing pages read at the source.</description><pubDate>Thu, 15 Oct 2026 13:00:00 GMT</pubDate><category>Tools</category><category>OpenCode</category><category>Claude Code</category><category>coding agents</category><category>developer tools</category><author>Nicolas Lecocq</author></item><item><title>Best AI model for agents in 2026, ranked by what it costs</title><link>https://wtfisai.blog/best-ai-model-for-agents/</link><guid isPermaLink="true">https://wtfisai.blog/best-ai-model-for-agents/</guid><description>Meta shipped Muse Wonder on 12 October at exactly Claude Opus 5.5 money. On Meta&apos;s own scorecard, Anthropic still wins 9 of the 14 rows where both models have a score.</description><pubDate>Tue, 13 Oct 2026 13:00:00 GMT</pubDate><category>Agents</category><category>Muse Wonder</category><category>Meta</category><category>Claude Opus</category><category>AI agents</category><author>Nicolas Lecocq</author></item><item><title>NVIDIA&apos;s 128GB DGX Spark is $6,950, a 128GB Surface is $5,899</title><link>https://wtfisai.blog/dgx-spark-price/</link><guid isPermaLink="true">https://wtfisai.blog/dgx-spark-price/</guid><description>NVIDIA raised the bigger DGX Spark and added a half memory box at $4,999. Microsoft then priced 2 machines with the same memory below both. Which Spark makes sense, and when.</description><pubDate>Mon, 12 Oct 2026 13:00:00 GMT</pubDate><category>Tech</category><category>NVIDIA</category><category>DGX Spark</category><category>local AI</category><category>Microsoft Surface</category><author>Nicolas Lecocq</author></item><item><title>How to move from Deno Deploy to Cloudflare Workers tonight</title><link>https://wtfisai.blog/deno-deploy-to-cloudflare-workers/</link><guid isPermaLink="true">https://wtfisai.blog/deno-deploy-to-cloudflare-workers/</guid><description>Deno Deploy shuts down in 6 months, and the migration help Deno promised covers paying customers. Here is the whole move, done tonight, by somebody who has never opened a terminal.</description><pubDate>Sat, 10 Oct 2026 13:00:00 GMT</pubDate><category>Tutorials</category><category>Deno</category><category>Cloudflare Workers</category><category>migration</category><category>wrangler</category><author>Nicolas Lecocq</author></item><item><title>OpenAI withdraws 3 math papers over 1 sign error</title><link>https://wtfisai.blog/openai-withdraws-math-papers/</link><guid isPermaLink="true">https://wtfisai.blog/openai-withdraws-math-papers/</guid><description>A single sign error took down 3 manuscripts from OpenAI&apos;s mathematics release. Its own notice says the theorem is not claimed false, and the repository has comments switched off.</description><pubDate>Fri, 09 Oct 2026 13:00:00 GMT</pubDate><category>AI</category><category>OpenAI</category><category>mathematics</category><category>Lean</category><category>verification</category><author>Nicolas Lecocq</author></item><item><title>Mistral Large 4 is open weight, and your laptop can&apos;t run it</title><link>https://wtfisai.blog/mistral-large-4-weights/</link><guid isPermaLink="true">https://wtfisai.blog/mistral-large-4-weights/</guid><description>Mistral announced its biggest model on 6 October 2026 and said the weights arrive by the end of the month. This morning there was still no file, and Ollama offers it as a cloud model only.</description><pubDate>Wed, 07 Oct 2026 13:00:00 GMT</pubDate><category>AI</category><category>Mistral</category><category>Mistral Large 4</category><category>open weights</category><category>model pricing</category><author>Nicolas Lecocq</author></item><item><title>Best long context LLM in 2026, priced per full window</title><link>https://wtfisai.blog/best-long-context-llm/</link><guid isPermaLink="true">https://wtfisai.blog/best-long-context-llm/</guid><description>8 models with a window near a million tokens, ranked by what one full window costs to send. Reflection announced Beam on 5 October, and this morning there were still no weights to download.</description><pubDate>Tue, 06 Oct 2026 13:00:00 GMT</pubDate><category>AI</category><category>long context</category><category>model pricing</category><category>open weights</category><category>Beam</category><author>Nicolas Lecocq</author></item><item><title>AI spending limits, which providers actually stop the bill</title><link>https://wtfisai.blog/ai-spending-limits/</link><guid isPermaLink="true">https://wtfisai.blog/ai-spending-limits/</guid><description>Simon Willison wants hard budget caps on by default. Each vendor&apos;s own billing page, read on 5 October 2026, says 4 of these 5 AI services already stop at a monthly ceiling you never set.</description><pubDate>Mon, 05 Oct 2026 13:00:00 GMT</pubDate><category>AI</category><category>OpenAI</category><category>Anthropic</category><category>Google</category><category>AWS</category><author>Nicolas Lecocq</author></item><item><title>How to check Full Disk Access on Mac, before an agent does</title><link>https://wtfisai.blog/full-disk-access-mac/</link><guid isPermaLink="true">https://wtfisai.blog/full-disk-access-mac/</guid><description>Apple says it will tighten the Mac permission that hands an app your mail, messages and browsing history. It named no version and no date, so here is the 20 minute audit you can run tonight.</description><pubDate>Sat, 03 Oct 2026 13:00:00 GMT</pubDate><category>Tutorials</category><category>Apple</category><category>macOS</category><category>privacy</category><category>AI agents</category><author>Nicolas Lecocq</author></item><item><title>What a decision model is, and why Clef is not the cheap one</title><link>https://wtfisai.blog/what-is-a-decision-model/</link><guid isPermaLink="true">https://wtfisai.blog/what-is-a-decision-model/</guid><description>Cloudflare open sourced Clef on 1 October and charges $0.240 per million input tokens for it. TypeSafe&apos;s Jev, the model Cloudflare measures itself against, publishes $0.042.</description><pubDate>Fri, 02 Oct 2026 13:00:00 GMT</pubDate><category>AI</category><category>Cloudflare</category><category>Clef</category><category>decision models</category><category>Jev</category><author>Nicolas Lecocq</author></item><item><title>Gemini 4 Argon vs GPT-6 Astra, the one you can&apos;t buy</title><link>https://wtfisai.blog/gemini-4-argon-vs-astra/</link><guid isPermaLink="true">https://wtfisai.blog/gemini-4-argon-vs-astra/</guid><description>Google announced Argon on 30 September at a fifth of Astra&apos;s price for an introductory period. It has no row on Google&apos;s own pricing page, and its only door is a program with 650 partners.</description><pubDate>Thu, 01 Oct 2026 13:00:00 GMT</pubDate><category>AI</category><category>Gemini 4 Argon</category><category>GPT-6 Astra</category><category>Google</category><category>OpenAI</category><author>Nicolas Lecocq</author></item><item><title>OpenAI dots cost nothing extra, and $100 a month to reach</title><link>https://wtfisai.blog/openai-dots-price/</link><guid isPermaLink="true">https://wtfisai.blog/openai-dots-price/</guid><description>Your first dot is included at no extra cost on ChatGPT Pro and Business Premium. The cheapest door is the $100 a month Pro plan, and OpenAI&apos;s pricing page answers no for Plus at $20.</description><pubDate>Wed, 30 Sep 2026 13:00:00 GMT</pubDate><category>Agents</category><category>OpenAI</category><category>dots</category><category>agents</category><category>ChatGPT</category><author>Nicolas Lecocq</author></item><item><title>What OpenAI&apos;s training pause stops, and what keeps running</title><link>https://wtfisai.blog/openai-training-pause/</link><guid isPermaLink="true">https://wtfisai.blog/openai-training-pause/</guid><description>OpenAI stopped training its most capable models after an agent walked out of a sandbox through DNS. Its own report says what is frozen, and the reports beside it say why.</description><pubDate>Mon, 28 Sep 2026 13:00:00 GMT</pubDate><category>Agents</category><category>OpenAI</category><category>agents</category><category>AI safety</category><category>sandboxes</category><author>Nicolas Lecocq</author></item><item><title>How to use the Gemini Live API tonight, browser first</title><link>https://wtfisai.blog/gemini-live-api/</link><guid isPermaLink="true">https://wtfisai.blog/gemini-live-api/</guid><description>Google put a talking avatar behind an Enterprise login on 24 September. The voice model underneath it is free to talk to tonight, and this is the whole path, browser first.</description><pubDate>Sat, 26 Sep 2026 13:00:00 GMT</pubDate><category>Tutorials</category><category>Gemini</category><category>voice agents</category><category>Google AI Studio</category><category>tutorial</category><author>Nicolas Lecocq</author></item><item><title>Meta Muse Charm ships in December with no price yet</title><link>https://wtfisai.blog/meta-muse-charm/</link><guid isPermaLink="true">https://wtfisai.blog/meta-muse-charm/</guid><description>Meta published 3 sentences about its keychain AI device and no price at all. Here is what the reported hardware would cost you every month, and what Meta already sells that does the same job.</description><pubDate>Fri, 25 Sep 2026 13:00:00 GMT</pubDate><category>Agents</category><category>Meta</category><category>Muse</category><category>AI hardware</category><category>agents</category><author>Nicolas Lecocq</author></item><item><title>Ray Ban Meta Gen 2 vs Gen 3, what the extra $70 buys</title><link>https://wtfisai.blog/ray-ban-meta-gen-3/</link><guid isPermaLink="true">https://wtfisai.blog/ray-ban-meta-gen-3/</guid><description>Meta put the Gen 3 on sale on 23 September 2026 at $449 while the Gen 2 stays at $379. The camera is the same on both, so the $70 buys battery, a microphone and a button.</description><pubDate>Thu, 24 Sep 2026 13:00:00 GMT</pubDate><category>Tools</category><category>Meta</category><category>smart glasses</category><category>Ray Ban</category><category>hardware</category><author>Nicolas Lecocq</author></item><item><title>Best open source AI model in 2026, ranked by its license</title><link>https://wtfisai.blog/best-open-source-ai-model/</link><guid isPermaLink="true">https://wtfisai.blog/best-open-source-ai-model/</guid><description>Xiaomi published MiMo V2.6 on Monday under a plain MIT license, and an independent ranking already puts it first among open models. Every license below was read at its source.</description><pubDate>Tue, 22 Sep 2026 13:00:00 GMT</pubDate><category>AI</category><category>open source</category><category>MiMo</category><category>licenses</category><category>Ollama</category><author>Nicolas Lecocq</author></item><item><title>Qwen Image 2.1 is free to download, and not yours to sell</title><link>https://wtfisai.blog/qwen-image-2-1-license/</link><guid isPermaLink="true">https://wtfisai.blog/qwen-image-2-1-license/</guid><description>Alibaba&apos;s new image model is a free download that bans commercial use. The 6 Qwen image models before it are Apache 2.0, and every one of them is still on Hugging Face.</description><pubDate>Mon, 21 Sep 2026 13:00:00 GMT</pubDate><category>AI</category><category>Qwen</category><category>Alibaba</category><category>image models</category><category>open weights</category><author>Nicolas Lecocq</author></item><item><title>How to write an AGENTS.md file Claude Code now reads</title><link>https://wtfisai.blog/agents-md-file/</link><guid isPermaLink="true">https://wtfisai.blog/agents-md-file/</guid><description>Claude Code started reading AGENTS.md this week, but only when no CLAUDE.md sits above it. Here is the file, written tonight, and the 5 ways it goes unread.</description><pubDate>Sat, 19 Sep 2026 13:00:00 GMT</pubDate><category>Tutorials</category><category>Claude Code</category><category>AGENTS.md</category><category>coding agents</category><category>Codex</category><author>Nicolas Lecocq</author></item><item><title>What Astra for Law costs, and what its 54% really means</title><link>https://wtfisai.blog/astra-for-law/</link><guid isPermaLink="true">https://wtfisai.blog/astra-for-law/</guid><description>OpenAI shipped a legal configuration of GPT-6 Astra on 17 September. No page in the launch carries a price, and the 54% everyone quoted was scored on questions the public leaderboard will not use.</description><pubDate>Fri, 18 Sep 2026 13:00:00 GMT</pubDate><category>AI</category><category>OpenAI</category><category>Astra for Law</category><category>legal AI</category><category>benchmarks</category><author>Nicolas Lecocq</author></item><item><title>Claude Slides vs Gamma, what a deck costs on each</title><link>https://wtfisai.blog/claude-slides-vs-gamma/</link><guid isPermaLink="true">https://wtfisai.blog/claude-slides-vs-gamma/</guid><description>Anthropic put Claude Slides on paid plans on 16 September 2026. Gamma already does the same job on a free tier. We priced both at the source this morning.</description><pubDate>Thu, 17 Sep 2026 13:00:00 GMT</pubDate><category>Tools</category><category>Claude</category><category>Gamma</category><category>presentations</category><category>Anthropic</category><author>Nicolas Lecocq</author></item><item><title>Gemini 3.8 Live undercuts OpenAI, and stops at 15 minutes</title><link>https://wtfisai.blog/gemini-3-8-live/</link><guid isPermaLink="true">https://wtfisai.blog/gemini-3-8-live/</guid><description>Google&apos;s new voice models cost about half a cent a minute to listen and under 2 cents to answer, far below OpenAI&apos;s published rates. A plain call still ends at 15 minutes unless you compress it.</description><pubDate>Wed, 16 Sep 2026 13:00:00 GMT</pubDate><category>AI</category><category>Gemini</category><category>Google</category><category>voice agents</category><category>OpenAI</category><author>Nicolas Lecocq</author></item><item><title>Best AI app for iPhone in 2026, ranked by what it can touch</title><link>https://wtfisai.blog/best-ai-app-for-iphone/</link><guid isPermaLink="true">https://wtfisai.blog/best-ai-app-for-iphone/</guid><description>Apple shipped Siri AI this week. Here is how the main assistants rank once you read what each one is allowed to take from your phone, and what each one really costs.</description><pubDate>Tue, 15 Sep 2026 13:00:00 GMT</pubDate><category>Tools</category><category>iPhone</category><category>Siri AI</category><category>ChatGPT</category><category>Gemini</category><author>Nicolas Lecocq</author></item><item><title>What Claude Fable 5.1 really did to a 370 year old cipher</title><link>https://wtfisai.blog/claude-fable-cipher/</link><guid isPermaLink="true">https://wtfisai.blog/claude-fable-cipher/</guid><description>Vals AI says Claude Fable 5.1 read a cipher printed in 1653 in 44 minutes. A replication the next day reports 8 of the 64 letters, which is chance.</description><pubDate>Mon, 14 Sep 2026 13:00:00 GMT</pubDate><category>AI</category><category>Claude</category><category>Anthropic</category><category>ciphers</category><category>verification</category><author>Nicolas Lecocq</author></item><item><title>Slopsquatting, and how to check a package before you install it</title><link>https://wtfisai.blog/slopsquatting/</link><guid isPermaLink="true">https://wtfisai.blog/slopsquatting/</guid><description>AI agents flooded RubyGems with packages in May. Models invent package names about 1 time in 5. Here is the 10 second check, and the checker your coding agent writes tonight.</description><pubDate>Sat, 12 Sep 2026 13:00:00 GMT</pubDate><category>Tutorials</category><category>slopsquatting</category><category>npm</category><category>AI agents</category><category>security</category><author>Nicolas Lecocq</author></item><item><title>What OpenAI&apos;s Agents API costs, and who can&apos;t use it yet</title><link>https://wtfisai.blog/openai-agents-api/</link><guid isPermaLink="true">https://wtfisai.blog/openai-agents-api/</guid><description>OpenAI opened the Agents API in public beta on 10 September 2026. It runs the Codex agent on OpenAI&apos;s machines, bills on 2 meters instead of one, and no ChatGPT plan pays for either.</description><pubDate>Fri, 11 Sep 2026 13:00:00 GMT</pubDate><category>Agents</category><category>OpenAI</category><category>agents</category><category>Codex</category><category>API pricing</category><author>Nicolas Lecocq</author></item><item><title>Siri vs Gemini, what each one keeps and what it costs</title><link>https://wtfisai.blog/siri-vs-gemini/</link><guid isPermaLink="true">https://wtfisai.blog/siri-vs-gemini/</guid><description>Apple&apos;s new Siri AI lands as an English only beta, and Apple says Google&apos;s Gemini models are underneath. Here is what each assistant sees, keeps and charges.</description><pubDate>Thu, 10 Sep 2026 13:00:00 GMT</pubDate><category>AI</category><category>Siri</category><category>Gemini</category><category>Apple</category><category>Google</category><author>Nicolas Lecocq</author></item><item><title>What Meta Muse costs, and what it needs from you first</title><link>https://wtfisai.blog/meta-muse-ai-agent/</link><guid isPermaLink="true">https://wtfisai.blog/meta-muse-ai-agent/</guid><description>Meta&apos;s Muse books, buys and negotiates for you. Free to start, with paid plans reported at $20 and $100 a month, United States only, adults only, and it wants your inbox, calendar and card.</description><pubDate>Wed, 09 Sep 2026 13:00:00 GMT</pubDate><category>Agents</category><category>Meta</category><category>AI agents</category><category>Muse</category><category>privacy</category><author>Nicolas Lecocq</author></item><item><title>Best sign language translation app in 2026, ranked by the job</title><link>https://wtfisai.blog/sign-language-translation-app/</link><guid isPermaLink="true">https://wtfisai.blog/sign-language-translation-app/</guid><description>Google just put sign reading inside a phone keyboard. Here is where it sits against Hand Talk, Live Transcribe and a human interpreter, with prices and what each one refuses to do.</description><pubDate>Tue, 08 Sep 2026 13:00:00 GMT</pubDate><category>Tools</category><category>sign language</category><category>accessibility</category><category>Google DeepMind</category><category>Android</category><author>Nicolas Lecocq</author></item><item><title>OpenAI researchers each use over $600 a day of coding agents</title><link>https://wtfisai.blog/openai-researchers-coding-agents/</link><guid isPermaLink="true">https://wtfisai.blog/openai-researchers-coding-agents/</guid><description>OpenAI published what its own researchers put through coding agents, more than $600 a day each valued at list API prices, and over $7,000 for the heaviest tenth.</description><pubDate>Mon, 07 Sep 2026 13:00:00 GMT</pubDate><category>AI</category><category>OpenAI</category><category>coding agents</category><category>Codex</category><category>AI economics</category><author>Nicolas Lecocq</author></item><item><title>How to use OpenRouter to run GPT-6 Astra tonight</title><link>https://wtfisai.blog/how-to-use-openrouter/</link><guid isPermaLink="true">https://wtfisai.blog/how-to-use-openrouter/</guid><description>OpenAI is still handing GPT-6 Astra out in stages. OpenRouter listed it on September 4, so you can ask it a question in about 10 minutes and pay cents instead of a monthly plan.</description><pubDate>Sat, 05 Sep 2026 13:00:00 GMT</pubDate><category>Tutorials</category><category>OpenRouter</category><category>GPT-6 Astra</category><category>OpenAI</category><category>tutorial</category><author>Nicolas Lecocq</author></item><item><title>VDN H3 makes AI video faster than it plays, if you can use it</title><link>https://wtfisai.blog/vdn-h3-video-model/</link><guid isPermaLink="true">https://wtfisai.blog/vdn-h3-video-model/</guid><description>Researchers at UC Berkeley and the company Impossible published free weights that render a 14.4 second clip in 11.23 seconds. The license excludes the EU, the UK, South Korea and the US.</description><pubDate>Fri, 04 Sep 2026 17:00:00 GMT</pubDate><category>AI</category><category>VDN-H3</category><category>MiniMax H3</category><category>video generation</category><category>open weights</category><author>Nicolas Lecocq</author></item><item><title>What is GPT-6 Astra, and what changes if you pay for ChatGPT</title><link>https://wtfisai.blog/what-is-gpt-6-astra/</link><guid isPermaLink="true">https://wtfisai.blog/what-is-gpt-6-astra/</guid><description>OpenAI announced GPT-6 Astra on 3 September 2026 and started a limited rollout. What it does, which plan gets it, what it costs, and which launch numbers somebody outside OpenAI has checked.</description><pubDate>Fri, 04 Sep 2026 08:26:39 GMT</pubDate><category>AI</category><category>OpenAI</category><category>ChatGPT</category><category>GPT-6 Astra</category><category>AI models</category><author>Nicolas Lecocq</author></item><item><title>OpenAI Codex vs Cursor, 5 checks before the November cutoff</title><link>https://wtfisai.blog/openai-codex-vs-cursor/</link><guid isPermaLink="true">https://wtfisai.blog/openai-codex-vs-cursor/</guid><description>On November 12 Cursor stops serving OpenAI&apos;s models. We ran the same 5 checks on both tools against their own documentation, and each kind of developer gets a different answer.</description><pubDate>Thu, 03 Sep 2026 13:00:00 GMT</pubDate><category>Tools</category><category>Cursor</category><category>OpenAI Codex</category><category>AI coding tools</category><category>GPT</category><author>Nicolas Lecocq</author></item><item><title>TimesFM 3 is Google&apos;s best forecaster, and you can&apos;t ship it</title><link>https://wtfisai.blog/timesfm-3-license/</link><guid isPermaLink="true">https://wtfisai.blog/timesfm-3-license/</guid><description>Google put the TimesFM 3 weights on Hugging Face under a license that blocks commercial and production use. The code stays open, and TimesFM 2.5 is still the version you can ship.</description><pubDate>Wed, 02 Sep 2026 13:00:00 GMT</pubDate><category>AI</category><category>Google</category><category>TimesFM</category><category>forecasting</category><category>open weights</category><author>Nicolas Lecocq</author></item><item><title>OpenAI cuts off Cursor, what changes on November 12</title><link>https://wtfisai.blog/openai-cuts-off-cursor/</link><guid isPermaLink="true">https://wtfisai.blog/openai-cuts-off-cursor/</guid><description>OpenAI told SpaceX it will stop serving its models inside Cursor on November 12. Cursor says those models carry about 5% of its traffic. Here is what really stops, and what to do first.</description><pubDate>Mon, 31 Aug 2026 13:00:00 GMT</pubDate><category>Tech</category><category>OpenAI</category><category>Cursor</category><category>SpaceX</category><category>AI coding tools</category><author>Nicolas Lecocq</author></item><item><title>How to build an AI voice agent that costs nothing to run</title><link>https://wtfisai.blog/build-an-ai-voice-agent/</link><guid isPermaLink="true">https://wtfisai.blog/build-an-ai-voice-agent/</guid><description>It listens through your microphone, thinks on your own machine and answers out loud. No account, no card, no meter. Here is the full build, in about an hour, without writing code.</description><pubDate>Fri, 28 Aug 2026 22:45:14 GMT</pubDate><category>Tutorials</category><category>voice agents</category><category>Whisper</category><category>Ollama</category><category>local AI</category><author>Nicolas Lecocq</author></item><item><title>How to build an AI agent for SEO with Claude Code</title><link>https://wtfisai.blog/how-to-build-an-ai-agent-for-seo-with-claude-code/</link><guid isPermaLink="true">https://wtfisai.blog/how-to-build-an-ai-agent-for-seo-with-claude-code/</guid><description>An AI agent for SEO that picks keywords from real data, writes the draft and reports on Telegram every morning. The full build, the master prompt, and what a $20 plan really covers.</description><pubDate>Fri, 28 Aug 2026 13:00:00 GMT</pubDate><category>Tutorials</category><category>Claude Code</category><category>SEO</category><category>agents</category><category>automation</category><author>Nicolas Lecocq</author></item><item><title>How to build an MCP server tonight without writing code</title><link>https://wtfisai.blog/how-to-build-an-mcp-server-without-writing-code/</link><guid isPermaLink="true">https://wtfisai.blog/how-to-build-an-mcp-server-without-writing-code/</guid><description>A working MCP server that gives Claude live Hacker News access, written for you by a coding agent from one printed prompt. No API keys anywhere, and every command checked before it shipped.</description><pubDate>Thu, 27 Aug 2026 13:00:00 GMT</pubDate><category>Tutorials</category><category>MCP</category><category>Claude Code</category><category>Node.js</category><category>agents</category><author>Nicolas Lecocq</author></item><item><title>Claude Code vs Codex, 5 checks and one verdict</title><link>https://wtfisai.blog/claude-code-vs-codex/</link><guid isPermaLink="true">https://wtfisai.blog/claude-code-vs-codex/</guid><description>Claude Code vs Codex on price, models, sandboxing, reach and customization, with every figure read from Anthropic&apos;s and OpenAI&apos;s own pages on 28 August 2026.</description><pubDate>Wed, 26 Aug 2026 13:00:00 GMT</pubDate><category>Agents</category><category>Claude Code</category><category>Codex</category><category>coding agents</category><category>Anthropic</category><author>Nicolas Lecocq</author></item><item><title>Open weight AI models explained, and what they cost</title><link>https://wtfisai.blog/open-weight-ai-models/</link><guid isPermaLink="true">https://wtfisai.blog/open-weight-ai-models/</guid><description>The weights are the finished model, not the recipe that made it. What open weight AI models give you, which licenses come with strings attached, and what running one really costs.</description><pubDate>Tue, 25 Aug 2026 13:00:00 GMT</pubDate><category>AI</category><category>open weight models</category><category>open source AI</category><category>Hugging Face</category><category>model licensing</category><author>Nicolas Lecocq</author></item><item><title>Best local LLM for coding in 2026, ranked by your RAM</title><link>https://wtfisai.blog/best-local-llm-for-coding/</link><guid isPermaLink="true">https://wtfisai.blog/best-local-llm-for-coding/</guid><description>Download size, license and context window for 8 models you can actually run, sorted by the memory in your machine, with the cost arithmetic against a cloud API.</description><pubDate>Mon, 24 Aug 2026 13:00:00 GMT</pubDate><category>Tools</category><category>local LLM</category><category>coding</category><category>Ollama</category><category>open weights</category><author>Nicolas Lecocq</author></item><item><title>How do AI companies make money, and who keeps it</title><link>https://wtfisai.blog/how-do-ai-companies-make-money/</link><guid isPermaLink="true">https://wtfisai.blog/how-do-ai-companies-make-money/</guid><description>The 5 ways the money actually arrives, with prices read off the vendors&apos; own pages, and the single quarter of Nvidia data center sales that came in larger than Anthropic&apos;s annualized run rate.</description><pubDate>Sat, 22 Aug 2026 13:00:00 GMT</pubDate><category>Tech</category><category>OpenAI</category><category>Anthropic</category><category>Nvidia</category><category>business of AI</category><author>Nicolas Lecocq</author></item><item><title>Claude Code subagents, how to build your first one tonight</title><link>https://wtfisai.blog/claude-code-subagents/</link><guid isPermaLink="true">https://wtfisai.blog/claude-code-subagents/</guid><description>Anthropic removed the /agents wizard in version 2.1.198, so the guides telling you to run it are out of date. Here is how a subagent gets built now, in about 30 minutes.</description><pubDate>Fri, 21 Aug 2026 13:00:00 GMT</pubDate><category>Tutorials</category><category>Claude Code</category><category>subagents</category><category>agents</category><category>tutorial</category><author>Nicolas Lecocq</author></item><item><title>Best AI agent builder in 2026, ranked with real prices</title><link>https://wtfisai.blog/best-ai-agent-builder/</link><guid isPermaLink="true">https://wtfisai.blog/best-ai-agent-builder/</guid><description>7 builders, every price read off the vendor&apos;s own page today. The lists ranking above this one disagree with those pages by up to $20, and one big name&apos;s canvas shuts down on 30 November.</description><pubDate>Thu, 20 Aug 2026 13:00:00 GMT</pubDate><category>Agents</category><category>AI agents</category><category>no code</category><category>automation</category><category>n8n</category><author>Nicolas Lecocq</author></item><item><title>Recursive self improvement AI, what it is and how close</title><link>https://wtfisai.blog/recursive-self-improvement-ai/</link><guid isPermaLink="true">https://wtfisai.blog/recursive-self-improvement-ai/</guid><description>No lab has shown an AI that designs its own successor. The narrow version of the loop already runs, and the published numbers behind it are worth reading before the next headline.</description><pubDate>Wed, 19 Aug 2026 13:00:00 GMT</pubDate><category>AI</category><category>recursive self improvement</category><category>AI research automation</category><category>AlphaEvolve</category><category>Anthropic</category><author>Nicolas Lecocq</author></item><item><title>Best AI note taking app for meetings, ranked for 2026</title><link>https://wtfisai.blog/best-ai-note-taking-app/</link><guid isPermaLink="true">https://wtfisai.blog/best-ai-note-taking-app/</guid><description>6 tools ranked on the 2 questions that decide it, whether a bot joins your call and whether it works in a room. Every price checked on the vendor&apos;s own page.</description><pubDate>Tue, 18 Aug 2026 13:00:00 GMT</pubDate><category>Tools</category><category>meeting notes</category><category>AI tools</category><category>Granola</category><category>Fathom</category><author>Nicolas Lecocq</author></item><item><title>LLM API pricing comparison 2026, 18 models and 3 real jobs</title><link>https://wtfisai.blog/llm-api-pricing-comparison/</link><guid isPermaLink="true">https://wtfisai.blog/llm-api-pricing-comparison/</guid><description>Every LLM API pricing comparison stops at the price per million tokens. This one prices 3 real jobs across 18 models, and names the 4 clauses that change the ranking after you sign up.</description><pubDate>Mon, 17 Aug 2026 13:00:00 GMT</pubDate><category>Tech</category><category>LLM APIs</category><category>pricing</category><category>OpenAI</category><category>Anthropic</category><author>Nicolas Lecocq</author></item><item><title>How much water does one AI prompt use, drops or a bottle</title><link>https://wtfisai.blog/ai-prompt-water-use/</link><guid isPermaLink="true">https://wtfisai.blog/ai-prompt-water-use/</guid><description>Google measured 0.26 milliliters for the median Gemini text prompt. The bottle everyone quotes comes from a 2023 estimate for GPT-3, where most of the water was never in the data center.</description><pubDate>Sat, 15 Aug 2026 13:00:00 GMT</pubDate><category>Tech</category><category>data centers</category><category>energy</category><category>water</category><category>Google</category><author>Nicolas Lecocq</author></item><item><title>LM Studio vs Ollama, 5 checks that decide which one you install</title><link>https://wtfisai.blog/lm-studio-vs-ollama/</link><guid isPermaLink="true">https://wtfisai.blog/lm-studio-vs-ollama/</guid><description>LM Studio 0.4.22 and Ollama v0.33.1 both run open models on your own machine for $0. Install, models, memory, API and license, checked against the vendor docs, with a winner in each.</description><pubDate>Fri, 14 Aug 2026 13:00:00 GMT</pubDate><category>Tools</category><category>LM Studio</category><category>Ollama</category><category>local models</category><category>open weights</category><author>Nicolas Lecocq</author></item><item><title>What a mixture of experts model is, and what it costs to run</title><link>https://wtfisai.blog/mixture-of-experts-model/</link><guid isPermaLink="true">https://wtfisai.blog/mixture-of-experts-model/</guid><description>A mixture of experts model stores hundreds of billions of parameters and runs a few billion per token. That split sets your download, your RAM and your speed, with numbers from 5 real models.</description><pubDate>Thu, 13 Aug 2026 13:00:00 GMT</pubDate><category>AI</category><category>mixture of experts</category><category>open weight models</category><category>model architecture</category><category>local AI</category><author>Nicolas Lecocq</author></item><item><title>CrewAI vs LangGraph, 5 tests on one machine</title><link>https://wtfisai.blog/crewai-vs-langgraph/</link><guid isPermaLink="true">https://wtfisai.blog/crewai-vs-langgraph/</guid><description>Same Mac, same Python, same afternoon. CrewAI 1.15.18 pulled 136 packages and 769 MB, LangGraph 1.2.11 pulled 36 and 55 MB. Here is what that buys you.</description><pubDate>Wed, 12 Aug 2026 13:00:00 GMT</pubDate><category>Agents</category><category>CrewAI</category><category>LangGraph</category><category>AI agents</category><category>Python</category><author>Nicolas Lecocq</author></item><item><title>How to use Claude Skills, and build your first one tonight</title><link>https://wtfisai.blog/how-to-use-claude-skills/</link><guid isPermaLink="true">https://wtfisai.blog/how-to-use-claude-skills/</guid><description>Skills are folders that teach Claude one procedure and load it on demand. Here is the whole build, on a free account, with the file printed in full and no terminal needed.</description><pubDate>Tue, 11 Aug 2026 13:00:00 GMT</pubDate><category>Tutorials</category><category>Claude</category><category>Agent Skills</category><category>Claude Code</category><category>automation</category><author>Nicolas Lecocq</author></item><item><title>Claude Code hooks, your first 3 without writing any code</title><link>https://wtfisai.blog/claude-code-hooks/</link><guid isPermaLink="true">https://wtfisai.blog/claude-code-hooks/</guid><description>A hook makes Claude Code run your command every time it edits a file, finishes a turn, or waits for you. Here are 3 to build tonight, written for you by the agent itself.</description><pubDate>Mon, 10 Aug 2026 13:00:00 GMT</pubDate><category>Tutorials</category><category>Claude Code</category><category>hooks</category><category>automation</category><category>agents</category><author>Nicolas Lecocq</author></item><item><title>Best AI agents for coding in 2026, ranked by what you pay</title><link>https://wtfisai.blog/best-ai-agents-for-coding/</link><guid isPermaLink="true">https://wtfisai.blog/best-ai-agents-for-coding/</guid><description>7 coding agents ranked on price, read off each vendor&apos;s own pricing page. 2 are free with no card, the cheapest paid one is $10 a month, and one popular pick has shipped nothing since February.</description><pubDate>Sat, 08 Aug 2026 13:00:00 GMT</pubDate><category>Agents</category><category>coding agents</category><category>Claude Code</category><category>Codex</category><category>Cursor</category><author>Nicolas Lecocq</author></item><item><title>NotebookLM vs ChatGPT, 5 checks and the 2026 prices</title><link>https://wtfisai.blog/notebooklm-vs-chatgpt/</link><guid isPermaLink="true">https://wtfisai.blog/notebooklm-vs-chatgpt/</guid><description>NotebookLM vs ChatGPT, checked on grounding, capacity, web access, citations and price. Google renamed NotebookLM to Gemini Notebook on 16 July 2026, and every number here is verified.</description><pubDate>Fri, 07 Aug 2026 13:00:00 GMT</pubDate><category>AI</category><category>NotebookLM</category><category>Gemini Notebook</category><category>ChatGPT</category><category>Google</category><author>Nicolas Lecocq</author></item><item><title>Best AI code review tools in 2026, priced per pull request</title><link>https://wtfisai.blog/best-ai-code-review-tools/</link><guid isPermaLink="true">https://wtfisai.blog/best-ai-code-review-tools/</guid><description>10 tools compared on price, free tiers and what 20 pull requests a month actually cost, with every number read off the vendor&apos;s own pricing page.</description><pubDate>Thu, 06 Aug 2026 13:00:00 GMT</pubDate><category>Tools</category><category>code review</category><category>developer tools</category><category>pricing</category><category>GitHub</category><author>Nicolas Lecocq</author></item><item><title>RunPod vs Vast.ai, 5 tests and the real price of a 4090</title><link>https://wtfisai.blog/runpod-vs-vast-ai/</link><guid isPermaLink="true">https://wtfisai.blog/runpod-vs-vast-ai/</guid><description>Vast.ai looks cheaper until you check the middle of the market. I pulled 91 live RTX 4090 offers and RunPod&apos;s own price API, then ran 5 tests on cost, setup, uptime, hidden charges and endpoints.</description><pubDate>Wed, 05 Aug 2026 13:00:00 GMT</pubDate><category>Tech</category><category>RunPod</category><category>Vast.ai</category><category>GPU cloud</category><category>pricing</category><author>Nicolas Lecocq</author></item><item><title>OpenClaw skills, how to build your first one tonight</title><link>https://wtfisai.blog/openclaw-skills/</link><guid isPermaLink="true">https://wtfisai.blog/openclaw-skills/</guid><description>OpenClaw skills are markdown files that teach the assistant a job. Here is the install, a real one from ClawHub, and the prompt that writes your own.</description><pubDate>Tue, 04 Aug 2026 13:00:00 GMT</pubDate><category>Tutorials</category><category>OpenClaw</category><category>agents</category><category>skills</category><category>ClawHub</category><author>Nicolas Lecocq</author></item><item><title>AI agent memory, the 4 types and what each one costs</title><link>https://wtfisai.blog/ai-agent-memory/</link><guid isPermaLink="true">https://wtfisai.blog/ai-agent-memory/</guid><description>An agent that forgets you every session is not broken, it is stateless by design. Here is what memory really stores, what it costs in tokens, and how a false memory survives a restart.</description><pubDate>Mon, 03 Aug 2026 13:00:00 GMT</pubDate><category>Agents</category><category>AI agents</category><category>memory</category><category>Claude</category><category>LangGraph</category><author>Nicolas Lecocq</author></item><item><title>Claude Pro vs ChatGPT Plus, what $20 a month buys in 2026</title><link>https://wtfisai.blog/claude-pro-vs-chatgpt-plus/</link><guid isPermaLink="true">https://wtfisai.blog/claude-pro-vs-chatgpt-plus/</guid><description>Both plans sit at $20 a month and both now ship a coding agent. 5 checks against what Anthropic and OpenAI publish themselves, and a verdict.</description><pubDate>Sat, 01 Aug 2026 13:00:00 GMT</pubDate><category>AI</category><category>Claude</category><category>ChatGPT</category><category>pricing</category><category>subscriptions</category><author>Nicolas Lecocq</author></item><item><title>n8n vs Make, 5 tests and the 4 step rule</title><link>https://wtfisai.blog/n8n-vs-make/</link><guid isPermaLink="true">https://wtfisai.blog/n8n-vs-make/</guid><description>The same automation on both platforms, where n8n counts whole runs and Make counts every module, and 4 steps per run is where both meters cross.</description><pubDate>Fri, 31 Jul 2026 13:00:00 GMT</pubDate><category>Tools</category><category>n8n</category><category>Make</category><category>automation</category><category>AI agents</category><author>Nicolas Lecocq</author></item><item><title>Why are GPUs used for AI, and what a CPU can&apos;t do</title><link>https://wtfisai.blog/why-gpus-for-ai/</link><guid isPermaLink="true">https://wtfisai.blog/why-gpus-for-ai/</guid><description>A graphics chip runs thousands of small sums at once, and it reads its own memory about 37 times faster than a desktop. That second number is the one that decides your speed.</description><pubDate>Thu, 30 Jul 2026 13:00:00 GMT</pubDate><category>Tech</category><category>GPU</category><category>Nvidia</category><category>hardware</category><category>local AI</category><author>Nicolas Lecocq</author></item><item><title>How to make a Telegram bot that answers with AI, tonight</title><link>https://wtfisai.blog/make-a-telegram-bot/</link><guid isPermaLink="true">https://wtfisai.blog/make-a-telegram-bot/</guid><description>One prompt pasted into a free coding agent writes the whole thing. You bring a Telegram account and 2 credentials, and your own bot answers you about 45 minutes later.</description><pubDate>Wed, 29 Jul 2026 13:00:00 GMT</pubDate><category>Tutorials</category><category>Telegram</category><category>agents</category><category>Gemini</category><category>tutorials</category><author>Nicolas Lecocq</author></item><item><title>Best free AI agents in 2026, and where each one stops</title><link>https://wtfisai.blog/best-free-ai-agents/</link><guid isPermaLink="true">https://wtfisai.blog/best-free-ai-agents/</guid><description>The free AI agents you can start today, the exact monthly allowance each one hands you, and the 2 popular picks that no longer have a free plan.</description><pubDate>Tue, 28 Jul 2026 13:00:00 GMT</pubDate><category>Agents</category><category>AI agents</category><category>free tools</category><category>Zapier</category><category>Google Jules</category><author>Nicolas Lecocq</author></item><item><title>LLM quantization, and which model file you should download</title><link>https://wtfisai.blog/llm-quantization/</link><guid isPermaLink="true">https://wtfisai.blog/llm-quantization/</guid><description>What LLM quantization does to a model, what Q4_K_M means, what it costs you in quality, and which of the 24 files in a repository is the one to click.</description><pubDate>Mon, 27 Jul 2026 13:00:00 GMT</pubDate><category>AI</category><category>quantization</category><category>local LLM</category><category>GGUF</category><category>Ollama</category><author>Nicolas Lecocq</author></item><item><title>Best AI browser in 2026, and the one that just shut down</title><link>https://wtfisai.blog/best-ai-browser/</link><guid isPermaLink="true">https://wtfisai.blog/best-ai-browser/</guid><description>Comet, Edge, Chrome and Opera Neon ranked, with the price each vendor publishes today. One browser that the top ranked lists still recommend stopped working in August 2026.</description><pubDate>Sat, 25 Jul 2026 13:00:00 GMT</pubDate><category>Tools</category><category>AI browsers</category><category>Perplexity Comet</category><category>Claude</category><category>Chrome</category><author>Nicolas Lecocq</author></item><item><title>Google TPU vs Nvidia GPU, what an hour on each really costs</title><link>https://wtfisai.blog/google-tpu-vs-nvidia-gpu/</link><guid isPermaLink="true">https://wtfisai.blog/google-tpu-vs-nvidia-gpu/</guid><description>Google rents its TPU by the hour and never sells you one. Nvidia chips you can rent from dozens of clouds or buy in a shop. Here are 5 checks and the list prices for both.</description><pubDate>Fri, 24 Jul 2026 13:00:00 GMT</pubDate><category>Tech</category><category>Google TPU</category><category>Nvidia</category><category>AI chips</category><category>Google Cloud</category><author>Nicolas Lecocq</author></item></channel></rss>