AI Workflow
7 articles
Karpathy Proved It — AI Agents Without a Validation Harness Will Fail Every Time
Karpathy's March of Nines math is brutal: 90% accuracy sounds great until you chain 10 steps and get 35% success. Here's how we built a 32-check Validation Harness to fix it.
Vision Eval — AI That Checks AI (Using Gemini Vision to QA AI-Generated Images)
We generate 20-30 AI images daily but never QA them — covers miss safe zones, images too dark, text gets blocked. We built vision-eval.py with Gemini Vision: 8 criteria, scored /80, 3 presets, compare mode.
Claude Channels + Telegram — Control Claude Code From Your Phone, No SSH Required
Anthropic released Claude Channels — an MCP plugin that connects Telegram to your Claude Code session. Deploy, fix code, transcribe voice messages, view images — all from your phone. No Tailscale, no tmux.
I Built an Entire AI Marketing Team — 11 Agents, 6 Real Clients, 6 Weeks
Agencies charge $5,000-$10,000/month for Website Audit + Copy Analysis + Competitor Research. I built an AI Marketing Team of 11 agents that works with real clients daily — using Claude Code.
11 AI Working Simultaneously — Agent Teams That Turn 3-Day Jobs into 2 Hours
Using Claude Code Agent Teams to run 11 AI instances in parallel — Strategy, Content, Media, Data, Design — they communicate, divide work, and we just approve.
AI Caught Fake Pricing — When 'Cheaper' Doesn't Mean 'Available'
We built an AI system to monitor competitor prices for an E-commerce client — scraped 600+ products, matched them automatically, compared prices, checked stock, and discovered competitors were faking prices on out-of-stock items.
Building a Blog with 7 AI Agents Working in Parallel — Behind INK by DopeLab
We used Claude Code Agent Teams to build INK by DopeLab from scratch — 7 AI agents working simultaneously to create an About page, CLI scripts, and a comments system in a single session.