There is no question that the development landscape changed with the advent of git – the free, lower-case, “git” command-line source control tool that Linus Torvalds threw together over a couple of weekends, has created a multi-billion dollar enterprise uppercase “Git” cloud empire dominated by GitHub; nine years ago Microsoft paid a staggering price to take over those millions of git repos—not so much as a direct money-maker, but rather as the ultimate training tool for OpenAI, which in turn helps to make their AI Copilot assistant as powerful as it is today.
Even with all that power, AI tooling like Copilot is likely to never replace dedicated DevOps software; even if it is true that you can do very powerful “DevOps-like” development, there are two very important points to remember:
- git is Source Control, it is not Version Control
- AI is non-deterministic and largely stateless
In simple terms, source control is all about maintaining (primarily) text files, and to that extent git can do some pretty amazing things like diffs, reverts, cherry picking, etc. Version control, on the other hand, is about creating a collection of objects (typically programs, files and other related artifacts) that can be deployed to different environments. Source control is a (very) small portion of Version control, and we haven’t even discussed the more powerful features of Version control like “Change While Active” and “Version rollback” which are so critical to a successful DevOps implementation.
From an AI perspective, “non-deterministic” simply means that with exactly the same inputs, you can get very different output/results; when you use AI to do a deployment, for instance, the same object deployed in exactly the same circumstance might succeed one time, and then fail the next – you simply never know which path AI is going to choose… For creative prompts and information gathering, that might be perfectly acceptable, but for businesses who literally live and die based upon the success of their software, that’s simply not an acceptable solution.
“Stateless” simply means that what one person does in one “AI Session” (like a Chat window in Copilot), often has little or no effect on another “AI session”; and almost certainly zero affect on another person doing exactly the same thing. DevOps software on the other hand, has a persistent, centralized database, and a clearly repeatable, auditable process, always the same regardless of user of interface, that is absolutely necessary for a successful DevOps implementation.
Without a doubt, it is possible to put together a very powerful “Proof of Concept” leveraging AI and git to do many of the things that DevOps also does, but the reality is that you probably don’t have the decades of experience that DevOps companies do, nor the desire to build and maintain your own tooling. Instead, leverage AI and git as they are best suited—as a part of the DevOps process, something that enhances and speeds the process, not as a replacement; combined together with classic DevOps tooling and expertise, the resulting “DevAIOps” can truly be more powerful than anything that existed even just a few years ago.
