In mid‑September 2026 I posted on LinkedIn about a new episode of the “RK on AI” podcast where Kashi and I debated the rapid evolution of AI models and its impact on software development.
Every week there’s a new model that makes the last one look stupid.
Demos look incredible.
Then you try to build real software with them.
Because making code faster isn’t the same as making software better.
Sometimes you just automate the mistakes too.
Maybe it’s what happens when software stops being written and starts being manufactured.
The discussion led me to a strategic pivot: I will focus research and tooling on managing a manufacturing‑style software pipeline and on preventing the propagation of existing coding mistakes.
- Open question: How can we detect and prevent the automation of existing coding mistakes?
- Open question: What tooling or processes are needed to treat software as a manufactured product?
- Open question: At what point does the cost/benefit of AI‑generated code outweigh the risk of propagating errors?
- Open question: How will developer roles evolve when the primary activity shifts from writing to overseeing manufacturing pipelines?
The software factory is presented as a change in mindset and process—not just a tool—requiring continuous adaptation, rebuilding, and compounding of proprietary workflows, systems, data, and execution.