Monday, June 1, 2026June 1, 2026
- Ran the Godot project and checked the map generation results
- Today’s actual project work centered on running Godot to view the map state
- The model could not reliably interpret images and screen results, making it hard to judge map improvements
- Custom web‑search MCP in a private repository
- Adjusted
web-search-mcp to match the desired search flow and output
- Tools other than Qwen Code had poorer usability with web‑search MCP than expected, feeling frustrating
- Qwen Code, Codex, OpenCode comparison test
- Qwen Code frequently repeated answers and retried plans
- Codex and OpenCode showed limits in understanding Godot context and interpreting images
- Some tools failed while planning, and sub‑agents for research queried the filesystem, breaking the flow
- Godot‑specific model experiment
- Used Hugging Face
bekoozkan/godot-gemma-4-e4b-it-GGUF model on RunPod
- Reference: https://huggingface.co/bekoozkan/godot-gemma-4-e4b-it-GGUF
- Several models tend to suggest code patterns from older Godot versions rather than the latest, prompting a specialized model trial
- However, this specialized model has few parameters and weak context understanding, so it was not as stable as hoped
- AI tool usage reflection
- Qwen Code seems unlikely to overcome repetitive issues without careful tuning, regardless of the model used
- Godot work is hard to consider complete without verification on an actual running screen
- I feel this area may not be a good fit for me
- Starting tomorrow, I will gather the necessary Godot work patterns myself and attempt fine‑tuning
- Development retrospective: docs/retrospectives/2026-06-01.md
- Meeting minutes: docs/meetings/2026-06-01.md
May 2026