- Author initially skeptical of AI agents for coding but found Opus 4.5 significantly improved performance.
- Created AGENTS.md files to control agent behavior, which were crucial for good results.
- Successfully built Python projects (YouTube scraper, FastAPI app) and Rust projects (icon-to-image, word cloud, miditui, ballin) using agents.
- Agents optimized Rust code to be 2-10x faster than existing libraries like Python's umap and scikit-learn.
- Developing rustlearn, a Rust crate for ML algorithms, aiming to outperform scikit-learn.
- Encourages giving modern agents another try with proper AGENTS.md and domain expertise.
- Despite toxic discourse, author remains optimistic and continues open-sourcing projects.