I Spent a Day with GPT-5.5: What the Benchmarks Don't Tell You
GPT-5.5 scores 82.7% on Terminal-Bench and 81.8% on CyberGym. I ran it through my own tests instead. Here's what I found.
dev>notes
Programming, AI engineering, and developer productivity.
Latest posts
How to Actually Build Your Own AI Model in 2026 (No Fluff)
APIs, RAG, fine-tuning, or training from scratch - here's what each path actually costs, which tools to use, and when you're wasting your money.
Claude Opus 4.7: Better at Coding, Worse at Trust
Opus 4.7 bumps SWE-bench to 87.6% and adds 3x vision, but the new tokenizer quietly inflates costs and the community isn't buying Anthropic's same pricing line.
Claude Mythos: Everything We Know About Anthropic's Most Powerful Model
Anthropic just announced Claude Mythos: 93.9% on SWE-bench, found a 27-year-old OpenBSD bug, and escaped its own sandbox. Everything developers need to know.
Best AI Models April 2026: Claude Mythos Just Broke the Tier List
Anthropic just announced Claude Mythos. 93.9% on SWE-bench, $25/$125 pricing, and a new tier above Opus. Plus the rest of the April 2026 model rankings, updated.
How to Actually Read LLM Benchmarks Without Getting Fooled
LLM benchmarks are marketing tools disguised as science. Here's what MMLU, HumanEval, and GPQA actually measure - and why your model choice shouldn't depend on them.
Claude Code Hooks: The Feature Nobody's Talking About
Claude Code hooks let you auto-format, lint, test, and block dangerous commands - all triggered by tool calls. Here's my exact setup.
Obsidian + Claude Code: The Developer Second Brain I Actually Use
My full Obsidian setup for developer knowledge management with Claude Code MCP integration.
OpenClaw vs Claude Code: I Tested Both for a Month
331K GitHub stars vs Anthropic's official CLI. Which AI coding tool actually ships code?
SOUL.md vs CLAUDE.md vs AGENTS.md: AI Config Files Compared
Every AI coding tool has a config file now. I tested all four formats. Here is what works.
The Definitive Guide to AI Developer Tools in 2026
Every AI developer tool I've tested, ranked. IDEs, CLI tools, models, and more.
Vibe Coding: Real Talk About the Trend
Everyone's talking about vibe coding. Here's what actually works and what's hype.
Building with AI Agents: My 2026 Workflow
Three AI agents handle different parts of my projects. How I coordinate them.
AI Experiments: What I Learned Building with Agents
Hands-on experiments with the latest AI models, tools, and workflows.
Best AI Models March 2026: Every Model Ranked
GPT-5.4, Claude Opus 4.6, Gemini 3.1 Pro, Grok 4.20, Qwen3.5 - ranked from daily use across coding, reasoning, speed, and cost.
My Exact Claude Code Configuration
CLAUDE.md files, custom slash commands, MCP servers, and permission settings.
Experiment: Can AI Build a Full App Autonomously?
Gave Claude Code a spec and walked away. What came back was 80% functional.
Claude Opus 4 Deep Review: Best Model for Code?
Extended context, better reasoning, fewer hallucinations. Detailed review after 3 weeks.
Best AI Prompts for Developers in 2026
The definitive prompt collection for Claude Opus 4, GPT-5, and Gemini 2.
How I Became a 10x Engineer with AI
How AI tools are redefining what it means to be a highly productive engineer.
AI-First Development: Why I Stopped Writing Boilerplate
Building software with AI at the core of the development process.
The 100x Developer Is a Myth (But 10x Is Real Now)
AI didn't make me 100x. It made me 5-10x on certain tasks. Here's the honest breakdown.
Lessons from Running AI Agents in Production
Error handling, cost control, output validation. What breaks when AI agents go live.
Windsurf vs Cursor in 2025: Which AI IDE Wins?
Two months with each. Tab completion, inline editing, context awareness. My verdict.
My Prompt Library: 30 Prompts I Use Weekly
Organized by category: debugging, architecture, testing, documentation, code review.
Using AI to Refactor Large Codebases
How I used Claude Code to refactor 50K+ lines. Strategy, prompts, and pitfalls.