The Elephants, Goldfish and the New Golden Age of Software Engineering is a software development approach created by Dave Rensin's team at Google. The basic idea is to prepare a comprehensive development plan within the "elephant" agent context, and then use many "goldfish" contexts to interrogate the plan. There is more depth to it than that, of course. What I find especially interesting is not just the approach, but also how much time you should expect to commit. Rensin writes that plans take about a week. This kind of explicit time commitment is rarely mentioned, but it seems right. This has been the norm for my entire career. You present an initial plan, discuss it with others, integrate feedback, present the next edition, perhaps divide one plan into multiple plans, and repeat. After a week of this, you have one or more plans that are far better than the initial one.
What Rensin does not address is how many people are involved. That is, is it just one human and one agent? I do not doubt that this pairing will create a robust plan with more rigor than we have created before. But is it a plan that should be implemented? A side effect of the previous approach is that multiple people have had a week to ponder the outcome. This leads to a better appreciation of whether or not the outcome is actually warranted. Those making the go/no-go decision are not required to make that decision quickly -- and they shouldn't. So, how will these plans be assessed in the context of the enterprise when few people have seen them and they appear fully baked?
