Blog
From the Forge
Lessons, patterns, and A/B test results from building with AI coding agents.
Blog
Lessons, patterns, and A/B test results from building with AI coding agents.
Most pipelines end. This one doesn't. A 4-pass discovery harness on a production Next.js site feeds the Crucible, tempering catches regressions, and the bug registry auto-smelts failures back into Discovery. The loop only pauses when there's nothing left to find.
Read the case study →AI coding agents get you to 80% fast — then start breaking things trying to close the last 20%. You've spent $1,500 in tokens to watch an agent destroy its own work. The fix isn't a better model. It's a better methodology.
Read article →I started with Spec Kit and loved it. Specification-based planning and agentic coding is like peanut butter and jelly. As my ideas grew, Plan Forge was born. Here's how they work together — and an honest comparison of when to use which.
Read article →After auto-loading guardrails, 21 reviewer agents, and 9 tech stack presets, here are the seven hard-won lessons from building Plan Forge — and the mistakes that taught them.
Read article →We built a feature twice — once with a single model, once with three models in parallel consensus. Both passed all gates. But the code quality difference was measurable. Here are the A/B results.
Read article →Three open-source tools wired into a closed-loop system. Describe a feature from your phone. The system hardens the plan, builds it, captures every decision, sends progress updates, reviews independently, and ships — all while you're at dinner.
Read article →Your team uses Copilot, Claude, Cursor, and Gemini. Plan Forge generates native guardrail files for all 7 agents — same quality gates, same pipeline, same 21 reviewer agents. One setup command.
Read article →We built the same .NET app twice: once with Plan Forge guardrails, once with pure vibe coding. Same model (Claude Opus 4.6). Same time (~7 min). The quality score: 99/100 vs 44/100. 60 tests vs 13. 6 interfaces vs 0. The data speaks for itself.
Read the results →A year ago, getting enterprise-grade code from an AI agent was nearly impossible. Today we build a 99/100 app in 7 minutes. This is the story of that journey — the failures, the breakthroughs, and what I learned along the way.
Read the full story →