Wagtail's one-month AI coding challenge failed—and that's the point
One month coding with GLM 5.3 Flash
Wagtail set out to spend a month coding exclusively with GLM 5.3 Flash. They succeeded for two weeks, then burned 1B tokens on other models. A vibe-coded MCP prototype cost $150 overnight, and infrastructure limits forced switches to DeepSeek V4.1 Flash and Qwen 3.8 Flash. They share benchmarks and lessons for budgeting experimentation and measuring energy, not just tokens.
We could have achieved similar results for most likely 5x less cost with not that much more effort.