I have been building software and web systems since 1995. This site documents what changes when experienced engineering process is used to plan, direct, review, and ship work through AI agents.
The useful story is not that AI writes code now. The useful story is where the 8x number held, what conditions made it possible, and where the claim stops. This first article sets the boundary: production-grade, narrow-scope CRUD/business web apps, not all software.
Read article →What a transcript, screenshots, and Claude can reveal about scope, and what they still cannot see: hidden implementation work, feedback loops, and the cost of turning an estimate into delivery.
Writing topicThe conventions, review hooks, and slice structure that let AI agents produce useful work without re-learning the project every time.
Writing topicThe estimate did not fail because AI was involved. It failed in an old, familiar way: user-facing requirements were visible, but import, security, infrastructure, review work, and design changes were not.
Writing topicThe useful question is how a 1,000-hour real-world build maps against legacy effort once hidden requirements, feedback, and production risk are visible.
Writing topicI am Jochen Daum, a software engineer based in Tauranga, New Zealand. I have been building software and web systems since 1995, across custom applications, WordPress, hosting, AWS, and production support.
This site is for practical notes on software delivery with AI agents: planning, slicing, review, deployment, hidden requirements, and the difference between coding faster and engineering better.
More about me →