Engineering, craft, and curiosity.
I'm a HUGE BELIEVER in people being able to leverage tools better when they understand how they work. Black boxes don't breed confidence in users. So let's do that with Agentic Development Systems.
I can't remember the last time I went a day without seeing an article about '[Google has 25% of code generated by AI](https://blog.google/technology/ai/google-gemini-next-generation-model-february-2024/)' or '[GitHub Copilot is writing 46% of code](https://github.blog/2023-06-27-the-economic-impact-of-the-ai-powered-developer-lifecycle-and-lessons-from-github-copilot/)' or '[Amazon Q has saved 4,500 developer-years](https://aws.amazon.com/q/developer/)'. The knowledge isn't in the line count. It's in how they got there.
AI tools are genuinely transforming software development. But the flood of executives claiming to ship 10,000 lines of code while running companies reveals a fundamental misunderstanding of what we're building.
Reflections on a year of building with Agentic AI. The good, the bad, the lazy, and why I can't go back.
Generative AI isn't an evolution of software development; it's a revolution. To leverage it, we must stop optimizing codebases for human eyes and start building them for Agentic verification.