Four distinct types of AI usage require different success metrics and skills, not a one-size-fits-all approach

The author identifies four common ways companies deploy AI: as a product interface (free-flow text input replacing structured forms), as a product implementation detail (analyzing unstructured data or generating output), as a coding tool (generating or modifying code), and as an administrative tool (research, feedback, text generation).

Each type has different trade-offs. AI product interfaces gain flexibility but lose correctness and determinism. Coding tools generate large amounts of code quickly but may lack correctness and understanding. Success metrics differ: product interfaces measure end-user usage, implementation details track service levels, coding tools measure shipped features, and administrative tools measure productivity.

Many companies introduce 'AI' broadly to increase company value without focusing on actual customer value; defining the specific type of AI usage in each conversation moves away from hype toward concrete customer impact.