AI Agent with Almost No AI: How Estimation Automation Works Without Neural Networks

The author shares their experience developing an 'AI Estimator' system for auditing construction cost estimates. Despite the name, the current production version operates without neural networks, relying instead on algorithmic rules. During testing on five real-world estimates, the system identified 310 issues, including critical calculation errors that had previously gone unnoticed. The author notes that while AI was used during the development phase to write code and analyze data, it was not integrated into the final product due to lack of API access and the client's requirement for offline functionality. A similar approach was used in a system for designing glass shower enclosures, where automation was implemented without AI. The article illustrates a practical development philosophy where the reliability and predictability of hard-coded rules take precedence over trendy technologies, proving that robust software logic is often more effective than 'black box' AI models.