AI Tutorials & Practice

The problem: code that reads well but means something else If you vibe code — or supervise...

An apology to start off with! I was too busy having fun in Sydney last week hacking with customers,...

Do you know vibe coding? Before you answer, please open 20 PRs with AI, get burned at least once,...

This week I went through every voice agent config I have exported on disk. Nine client agents: HVAC,...

Glossary Task (Snowflake): a scheduled SQL statement or procedure call. Created...

One year ago, I opened a terminal on a small home server and gave an AI agent its first real...

In 2025 the s1ngularity payload launched Claude Code, Gemini CLI and Amazon Q with approval prompts off. How it worked, and the settings that refuse it.

A vision model said the layout was fine while the footer was missing. How a pixel-level gate replaced it: per-line margins, CDP screenshots and a control that actually fails.

Originally published on my Substack. This week I launched airtaghistory.com. It keeps the AirTag...

A person opens a trip draft and asks an assistant: Prepare my SFO to JFK trip for November 12–15,...

Exploring communication efficiency in cooperative artificial intelligence. Introduction:...

Most people pick one model and ask it to do everything. We stopped doing that months ago, mostly...

HustleRadar: A Human-in-the-Loop AI Workspace for Safer Resale Decisions What I...

I've spent 8+ years building React/TypeScript front ends. Now I'm learning applied AI in public,...

We assumed AI would give developers an unfair advantage. But the people actually winning with AI aren't always the most technical ones.

A local-first, zero-telemetry PII and secrets redaction proxy with bidirectional rehydration for open-weight models (Gemma, Llama, Mistral).

Reuters says AI agents run from exposed open-source panels breached 100+ sites at about $50 each. The one break-in NorForge verified used old PHP flaws.

Tool calling, vision, speech to text, text to speech and search, measured on 4 vCPUs with no GPU. The harnesses are open so you can rerun them.

Agentic AI is everywhere right now, and for a student just getting started it can look impossible....

Every new chat with my coding agent started the same way. Ten minutes of me typing some version of...

Google has rolled out Gemini 4 Argon, a model positioned for complex, multi-step work across the...

Microsoft AI’s MAI-Transcribe-2 achieved the lowest transcription error rate in 18 of 21 languages in...

A user opens an email assistant and asks it to summarise their unread messages. Somewhere in their...

Open the network tab while you use most AI writing tools. Watch what happens as you type an email, a...

I built Sorcerer, a free, open-source macOS app (SwiftUI, macOS 15+, MIT) that simulates a community...

Day 19 of 30 Days of Search. TL;DR A vulnerability research agent investigates an alert...

Injection Defense: Seven Notes on Settling the Security Debt Source version of CodeSmith:...

There's a piece of advice I've been seeing a lot lately...reduce friction, protect your engineers'...

Claude Code's /sandbox is built on @anthropic-ai/sandbox-runtime. Install srt, run one probe plain and sandboxed, and see which layer stops each line.

A local tool that reads Panamanian receipt photos with Gemma and produces a CSV with the ITBMS sales tax, where the model transcribes and the code calculates.

One text prompt, four original 1376\u00d7768 stills: Higgsfield charged two credits per image and delivered watermarked PNGs; Kyncept charged one credit per...

For years, I was involved in developer advocacy. I organised meetups, spoke at many of them, and was...

Persistent agents and agent teams truly caught my attention when I tried Grok Bot. Publicly, Grok...

AI is becoming capable of doing more than generating test cases. AI agents can help plan test steps,...

I build a small site where people play heads-up hold'em against AI models. At the end of every hand...

I Built LineLoom: A Local-First VS Code Autocomplete Extension — v1.0.0 Is Here Write...

The fail-closed gate: a release-gate design story Draft — for review. Part 5 of the...

Frontend and backend need different agent rules. Set up nested AGENTS.md and CLAUDE.md files, then check which ones Codex and Claude Code really loaded.

Picking the best of 21 gate thresholds on 100 held-out actions overstates EV by 0.12 in a seeded simulation. Why it happens, and three fixes.

KillSwitch is an esolang designed to be unusable by LLMs: a lying spec, anagram-only function calls, values chosen by dice roll. I built its interpreter, ran every sample program, diffed its two specs (484 lines vs 887,584 words of decoy),…