Stop Managing AI Projects Like Traditional Software Projects

This presentation shows you why conventional approaches to product development and software development projects break down when building AI and what to do instead.

The presenter explores how AI development requires a shift from rigid roadmaps to experimental, evaluation-driven processes. By treating development as a journey of discovery, teams can better identify systemic failures, prioritize impactful work through data-driven insights, and effectively adapt their strategies to the unpredictable nature of intelligence-based projects.

Software development projects tells the computer how to make decisions through explicit code. In AI projects, data is the first-class citizen. AI systems learn how to make decisions from examples and data, making data quality, experimentation, and ongoing monitoring central to the project’s success. In traditional projects, software developers focus is mainly on writing business logic. In AI projects, collecting, cleaning, labeling, and managing data often consumes more effort than model development itself. Software testing will also focus on the accuracy of the model.

Video producer: https://maven.com/parlance-labs/evals