GPT-5.2-Codex adds long-horizon context, improved large-scale code changes, Windows support, vision inputs, and stronger defensive cybersecurity. Now available to paid ChatGPT users; invite-only pilot for vetted teams.
Gemini 3 Flash scored 90% across three coding benchmarks, beating Gemini 3 Pro while running 3–4x faster and costing 6x less. Best used for high-volume implementation when paired with a planning model.
Anthropic released Agent Skills as an open standard with spec at agentskills.io and source on GitHub. Adopted names like GitHub, VS Code and OpenAI signal growing ecosystem support.
Superpowers 4.0 formalizes subagent review loops, streamlines skill triggers by condensing descriptions, and adds a lightweight end-to-end test suite. Increased GraphViz use improves process documentation.
Superpowers 4.0 formalizes subagent review loops, streamlines skill triggers by condensing descriptions, and adds a lightweight end-to-end test suite. Increased GraphViz use improves process documentation.
Enable with codex --enable skills to let Codex discover SKILL.md skills from project docs and ~/.codex/skills/. A Discovery meta-skill governs triggers, sequencing and fallbacks.
Enable with codex --enable skills to let Codex discover SKILL.md skills from project docs and ~/.codex/skills/. A Discovery meta-skill governs triggers, sequencing and fallbacks.
Vibe-Coding & AI Dev NewsOltre.dev is a human-curated collection of AI news, insights, and resources for developers. Content is written with AI, reviewed by humans, and designed to keep you up to date as technology moves forward.
Marc Brooker argues programming will rely on natural-language specs and iterative conversation, with LLMs helping refine intent. Formal methods still matter for safety; neurosymbolic checks link clarity to verification.
Marc Brooker argues programming will rely on natural-language specs and iterative conversation, with LLMs helping refine intent. Formal methods still matter for safety; neurosymbolic checks link clarity to verification.
Claude Code 2.0.70 can drop skills when skill descriptions push the system prompt past a 15,000-char limit, with no warning. Workaround: set SLASH_COMMAND_TOOL_CHAR_BUDGET=30000; Superpowers 4.0 will reduce metadata.
Claude Code 2.0.70 can drop skills when skill descriptions push the system prompt past a 15,000-char limit, with no warning. Workaround: set SLASH_COMMAND_TOOL_CHAR_BUDGET=30000; Superpowers 4.0 will reduce metadata.
Explain Changes adds context-aware, plain-English notes to git diffs to surface the intent behind edits. Use the button or /explain-changes to speed PR reviews, catch AI 'slop', and improve onboarding.
Explain Changes adds context-aware, plain-English notes to git diffs to surface the intent behind edits. Use the button or /explain-changes to speed PR reviews, catch AI 'slop', and improve onboarding.
Open multiple project folders in a single Kiro IDE window while keeping each root independent. Unified search, scoped agent hooks, and per-root configs make cross-repo editing, refactoring, and workflows seamless.
Open multiple project folders in a single Kiro IDE window while keeping each root independent. Unified search, scoped agent hooks, and per-root configs make cross-repo editing, refactoring, and workflows seamless.
Amp swapped Haiku 4.5 for Gemini 3 Flash in its codebase search subagent, yielding about 3× faster responses while matching F1 quality. The new model issues more parallel calls and finishes in fewer iterations.
Amp swapped Haiku 4.5 for Gemini 3 Flash in its codebase search subagent, yielding about 3× faster responses while matching F1 quality. The new model issues more parallel calls and finishes in fewer iterations.
Deep dive videos for AI developers
Andrej Karpathy's keynote on June 17, 2025 at AI Startup School in San Francisco.
Drawing on his work at Stanford, OpenAI, and Tesla, Andrej sees a shift underway. Software is changing, again. We’ve entered the era of “Software 3.0,” where natural language becomes the new programming interface and models do the rest.
He explores what this shift means for developers, users, and the design of software itself— that we're not just using new tools, but building a new kind of computer.
Slides provided by Andrej: https://drive.google.com/file/d/1a0h1mkwfmV2PlekxDN8isMrDA5evc4wW/view?usp=sharing
If you’re tired of Cursor (or any other AI editor) breaking your code with random errors, this video is for you. I’ll show you how I used Task-Coding to fix 90% of the issues that come from vibe-coding. This method gives Cursor clear structure, better context, and way fewer mistakes. Watch me build a full app with multiple pages, charts, and auth—error-free.
Might be worth noting that using the MAX mode can eat up your costs. You get billed per token for the full context window of the entire chat. There are a few other chat models that you can use without being on MAX mode that still have a great context window, such as the gemini models.
An honest review of Cursor Background Agent on a real task.
Microsoft just made GitHub Copilot free and open source software under the MIT License. As of today, you can fork it, modify it and even build your own billion dollar competitor without going to prison.
Got a lot of questions asking about my stack and what I do when vibe coding. So I made a full video on it!
Builder.ai, a UK-based startup backed by Microsoft and Softbank, has blown its valuation of $1.5 billion and collapsed back to zero. In today's video, we'll find out what went wrong and what it means for the future of ai coding startups.
Did this article spark your interest? Join our community of experts and enthusiasts to dive deeper, ask questions, and share your ideas.
Join our community