Nvidia agrees to buy Hugging Face for nearly $13 billion, Shopify acquires Tailwind after AI gutted its commercial business, and DuckDB joins AWS. OpenAI cuts off Cursor and its agents attacked RubyGems without disclosure, Altman and Musk back Dario Amodei's call to pace the frontier as a safety researcher resigns, GLM-5.3-Flash undercuts the premium labs, and a skill tries to stop coding agents from burying the answer. [...]
A free desktop whiteboard with no accounts and no servers, where your AI agent does the drawing. Why that beats slides and Excalidraw for explaining code. [...]
Cache tokens are the parts of your prompt that AI providers bill at a big discount because the model has already processed them. Here's what cache reads and writes are, how Claude, OpenAI, Gemini, and DeepSeek charge for them, and how to structure your prompts so you actually pay the cheap rate. [...]
Content negotiation is making a comeback as a way to serve clean Markdown to AI agents. Here's how it works, why developers are split on it, and whether it's worth implementing. [...]
Stripe acquires OpenRouter for over $7 billion, Cursor launches Origin as a GitHub alternative, an AI-generated Copilot autofix breaches Snowflake's Jira, a grey market resells AI credits and startup perks as cheap inference, an on-device model autocompletes piano, and a look at how Bluesky watermarks its logo onto screenshots. [...]
GitHub introduces stacked pull requests, Gemma 4 uses roughly 2 GB of memory for weights and KV cache on Apple Silicon, and Kimi K3 offers an unusual open-weight model architecture. Portable AI sessions, PostgreSQL 3D visualization, and why faster coding does not always produce better software. [...]
In this post, I explain what effort levels actually are, why more thinking costs more money, why some models are great on low effort while others fall apart, and what really happens when you pick an effort level in Claude Code, opencode, or Codex. [...]
TypeScript 7's native compiler lands, a firm charges $10,000 a week to delete AI generated slop, Clawk provides disposable Linux VMs for coding agents, and Claude Code sends 33k tokens before it even reads your prompt. [...]
AI models are becoming a commodity while the "harness" wrapped around them turns into the real moat. I break down what inference and a harness actually are, how they differ, and why every AI lab is now shipping its own, using a self-driving car analogy. [...]
Fable 5 quietly returns with a usage cap that undercuts its own doomsday marketing, HackerRank's open-sourced ATS can't score the same resume twice, and ZCode wraps GLM-5.2 in a Claude Code-style harness. Plus DeepSeek's DSpark paper, Box3D's open 3D physics, and Claude Code steganographically watermarking your prompts. [...]
OpenAI's GPT-5.6 Sol ships behind US government vetting, Anthropic's Mythos and Fable get rationed by Washington, open-weight models rise as the cheap alternative, smart model routing, Node's new Nub toolkit, Google firing a CLI maintainer, and Microsoft's quantum claim undone by Python bugs. [...]
GLM-5.2 takes the open-weights crown at a fraction of the price, DeepSeek finally gains vision, Lore emerges as a version control system built to challenge Git at scale, MCP ships enterprise-managed OAuth, and Charity Majors argues AI demands more engineering discipline, not less. [...]
The US government orders Anthropic to suspend Fable 5 and Mythos, SpaceX acquires Cursor for $60 billion, Homebrew 6.0 ships with macOS 27 support, an AI agent racks up a massive bill unsupervised, Apple enters the container space, going HTML first doubles a startup's users, and a LinkedIn job offer hides a backdoor. [...]
Anthropic ships Opus 4.8 with dynamic workflows, OpenRouter raises $113M to become the model marketplace, SQLite draws a hard line against AI generated code, and a developer's open source project gets weaponized for phishing. [...]
The Weekly Diff is a curated roundup of the week's most interesting developer stories. This week: a trojanized VSCode extension breaches 3,800 GitHub repos, Google silently replaces Antigravity with a chatbot, Microsoft kills Claude Code licenses, Files.md offers a no-lock-in Obsidian alternative, and Node.js 26 finally ships the Temporal API. [...]
OpenCode, pi, T3 Code, and Kilo are four open source AI coding agents you can try for free. I compare them by workflow fit, provider flexibility, Claude subscription access, and what "free" actually costs. [...]
AI models now support 1M+ token context windows, but bigger isn't always better. In this post, I talk about the "lost in the middle" problem, why models struggle to retrieve information from long contexts, and the trade-offs you should know before relying on large context windows. [...]
Learn how to use your Google account with Opencode via the opencode-gemini-auth plugin, and what changed after Google deprecated Gemini Code Assist consumer account access on June 18, 2026. The OAuth flow now only works for Gemini Code Assist Standard and Enterprise subscriptions, while the Gemini API key flow is unaffected. [...]
antigravity-claude-proxy, an open source project by Badri Narayanan S, masterfully links Google’s Antigravity with Anthropic’s Claude Code, letting you run Claude models powered by your Antigravity tokens in Claude Code. [...]
Google casually dropped three products on Nov 18, 2025 — upgraded versions of two of their most popular products, Gemini and Nano Banana, and Antigravity — an IDE with agentic capabilities. This brings us back to the Gemini era in the cycle of SOTA (State-Of-The-Art) models. [...]
OpenAI shut down the Assistants API on August 26, 2026. If your integration still depends on it, here's a step-by-step guide to migrate your code to the Responses API, with the before and after of every concept, from assistants to prompts and threads to conversations. [...]
The usage of AI, especially in the software industry, has increased a lot lately, but everything has a downside — and that, is, excess. Excess of anything is bad, and that includes the use of AI. "Vibe Coding" is a term coined by Andrej Karpathy, and in this post, I explore the downsides of vibe coding and how to balance it. [...]
AI models and tools are becoming more and more capable day by day, especially at generating code. In this post, I will walk you through my experience of creating a full-stack application using some of these AI coding assistants/agents and what are the implications of AI generated code. [...]
MCP (Model Context Protocol), as the name suggests, is a protocol — a way of communication and a set of rules, just like what you have in other protocols, such as HTTP or TCP. Some also term it as the "USB-C" for AI. This post is about me exploring what MCP is, building a basic MCP server, how it works, why there was a need for it, and what changed with the 2026-07-28 spec update. [...]
In this tutorial, I will demonstrate how you can use custom API keys from multiple AI providers and use any models associated with them in GitHub Copilot Chat VSCode extension. BYOK now works on all Copilot plans, including Business and Enterprise, and even without a Copilot plan at all. [...]
Anthropic announced its latest and best Claude model yet — Claude 3.7 Sonnet on February 25, 2025. OpenAI jumped on the bandwagon and announced GPT 4.5 — its latest and largest model on February 27, 2025. To make things more interesting, Microsoft unveiled its first quantum chip — Majorana 1 — which uses topological qubits. [...]
Deep research is one of the aspects AI companies are willing to implement and integrate with their current AI systems. The use cases of deep research are massive in nature, because it saves you a lot of time skimming through countless articles and generating a comprehensive report. One such use case can be in the recruitment industry. [...]
DeepSeek's API is OpenAI-compatible, which makes it easy to plug their models into almost any tool. In this guide, I walk through using DeepSeek via their own API platform and via OpenRouter, plus how to get reliable structured JSON output out of a reasoning model. The same setup works for whatever the current DeepSeek model is, not just R1. [...]
Visual Studio Code (VSCode) is an editor which is simple on it's own but enriched when combined with extensions. Here are the visual studio code extensions that I use — cline, filesize, better comments, code spell checker, git graph, markdownlint, total typescript, pretty typescript errors, sqltools and codeium (windsurf). [...]
A list of curated apps I use and will probably keep using for the rest of 2025 - continuing my spree of documenting app defaults which I started in 2023. [...]
Code reviews have always been crucial in maintaining a standard and emphasizing on the best practices of code in a project. This is not a post about how developers should review the code, it's more about delegating a part of it to AI. Introducing Better - An AI powered Code Review Tool. [...]
At the time of this writing, Google Chrome has started shipping experimental AI features such as "built-in AI" in Chrome's Dev/Canary channels. In this tutorial, you will get to know how you can try the built-in AI model in chrome dev. [...]