Site Reliability Engineering, 2nd Edition

Chapter 19. SRE at the Tipping Point

It is now possible to instruct a machine to write code to do an arbitrary task using colloquial English. While not perfect, the output you get is generally usable, and we expect it will only get better.¹ This raises many questions.

Given that AI is redefining much of how we understand and approach software development, how is what we currently do in SRE likely to change as a result? As the underlying dynamics of development change, what is likely to happen to how software is built, maintained, and operated? How will nonfunctional requirements such as reliability and security be handled, and in particular, what will happen to production-related work? Or, to put it another way, if the cost of generating semi-useful lines of code becomes marginal, what becomes costly?

Note

Throughout the rest of this chapter, we will use the term “AI dev” to mean AI-driven software development (via CLI tool or editor plug-in, in-stream, or otherwise). This is distinct from “AI SRE,” which we use to mean production engineering activities conducted with the aid of‚ and perhaps sometimes entirely by, AI.

The Software Development Lifecycle

Software, at scale, goes through a lifecycle as it is developed and deployed. While the precise details vary between organizations and projects, most software engineers are familiar with a prototypical model that divides software development into distinct phases, roughly along the following lines (as shown in Figure 19-1): Plan ...