We’re creating a new generative AI Step-by-Step mini-series. Over the next few weeks I’ll walk through a new AI project from start to deployment, and address the major decisions and activities. For example, I’ll show you how to select the best model, prepare data, optimize prompts, and design the user experience.
Today I’m going to cover the very first step: selecting your best generative AI project. Most companies exploring generative AI opportunities soon discover they have many options and don’t know where to begin. It isn’t unusual for a CTO to call me and share a list of dozens or hundreds of potential use cases. Usually they form a steering committee to help prioritize, but after a few months they have more questions than answers - and a longer list of ideas.
I was in this position a few days ago. I needed to pick a new Generative AI project for this step-by-step mini-series and had a number of ideas. All had pros and cons, and many potential challenges were unknown. So I created the following decision criteria for selecting the best generative AI project:
Criteria 1: Fast start
Criteria 2: Accessible data
Criteria 3: Available experts
Criteria 4: Many paths to success
Criteria 5: Aligned to your experience
I evaluated each idea against these five criteria. So what project did we choose? We’re going to build a solution that uses large language models to understand and interact with our codebase. We’re calling it Ground Crew and hope to improve code maintenance, knowledge management, engineering onboarding, documentation, and identify potential code issues.
I’m going to show you how we’re building Ground Crew step-by-step over the next few weeks. In the meantime you can use these five criteria when evaluating your own options.