- AI speeds up every part of software development — and can slow the whole downEnglish | Русский AI speeds up every part of software development — and can slow the whole down Fiona Chen and James Stratton of Harvard went through five years of telemetry from 718 firms: after the switch to AI agents, 30% more lines of code and the same number of closed tasks. Review time in … Continue reading “AI speeds up every part of software development — and can slow the whole down”
- AI Day: how translating a long sentence grew into ChatGPTEnglish | Русский AI Day: how translating a long sentence grew into ChatGPT Twelve years ago a paper came out about machine translation. Step by step, ChatGPT grew out of it. September 1 — the day the school year traditionally starts, Knowledge Day in Russia and much of the post-Soviet world. It also makes a … Continue reading “AI Day: how translating a long sentence grew into ChatGPT”
- The Seven Biggest Claude Skills Collections: What’s Inside, Which to Trust, and How to Use ThemEnglish | Русский The Seven Biggest Claude Skills Collections: What’s Inside, Which to Trust, and How to Use Them TL;DR — if you only read this box: Start here: anthropics/skills for a small, production-grade, vetted set. Add glebis/claude-skills (~90 tidy personal skills) if you want more, or obra/superpowers if you want a whole workflow, not … Continue reading “The Seven Biggest Claude Skills Collections: What’s Inside, Which to Trust, and How to Use Them”
- Language, format, placement: how to write prompts an LLM understands betterEnglish | Русский Language, format, placement: how to write prompts an LLM understands better I spend my days wrangling big prompts to language models — and the same question keeps coming up: what actually works better? Is it true that if you write tersely, telegraph-style, with no filler, the model both understands more precisely and … Continue reading “Language, format, placement: how to write prompts an LLM understands better”
- How to Organize a Repository for an LLM AgentEnglish | Русский How to Organize a Repository for an LLM Agent More complex is better? I don’t think so. Five tiers of repository organization: Tier 0 — flat files. Everything in context. Prototypes, configs, small projects under 20 files. Tier 1 — text search + CLAUDE.md. How every AI coding agent works. Code projects … Continue reading “How to Organize a Repository for an LLM Agent”