The Society of Automotive Engineers has defined six levels of autonomous driving, ranging from Level 0—where the driver is responsible for everything—to Level 5—where a car performs all driving tasks under any conditions from Point A to Point B. The same spectrum can be used with system administration tasks to determine where and to what extent AI should be leveraged.
If we apply the SAE’s levels to system administration tasks, it would look something like this:
- Level 0: No Automation
- Level 1: Assistance Required
- Level 2: Partial Automation
- Level 3: Conditional Automation
- Level 4: High Automation
- Level 5: Full Automation
I thought about the sysadmin tasks ripest for AI-based automation and leveled them based on this spectrum. It’s important to note that all of this is a snapshot in time. As AI technology matures—and organizations’ comfort level with AI increases—what’s a Level 2 today may be a Level 3 or 4 tomorrow.
It stands to reason, but none of the tasks I focused on landed at Level 0 or Level 1. As with cars, there are few system administration tasks that involve little to no automation. You could say something like racking servers and unraveling cables, but AI will never help detangle a Clark Griswold-level cable ball.
You’ll also notice that there are no Level 5 tasks—yet. (More on that later.)
I’m open to discussion and even argument on what I’ve come up with. I would also be really interested to hear how sysadmins would categorize these and other functions, as well as how they see the sysadmin role changing as AI matures.
System shutdown: In the traditional world, this could be a server shutdown, but in a modern, cloud-native world, it could be the shutdown of a critical application, load balanced across thousands of containers which run on hundreds of worker nodes. Either way, a human needs to be involved at a high level. There are a variety of reasons for initiating a shutdown, but humans should always be the ones driving them. At most, system shutdown should be Level 2 on the autonomy scale. AI can help suss out behavior anomalies or security threats. A “driver” assistance might prompt:
- “Are you really sure you want to shut that down?”
- “I noticed a couple of containers didn’t shut down correctly and were still serving traffic”
- “A critical task is hanging, and data hasn’t been flushed to disk, so shutting down now could cause database corruption”
Sort of like lane keeping. AI has the potential to really enlighten the user about the subtasks that are happening, and what their status is, in a completely new and transparent way. But the decision to initiate a shutdown should come only after a human has verified an issue and authorized defensive actions — feet on gas and brakes, if you will.
Repairing system issues: Using AI to diagnose issues and then automatically fix those issues is a promising use case, but still a Level 2. I’ve had conversations with colleagues who used agentic AI to determine whether a set of pods in Kubernetes was healthy and recommend tools to use to fix them if they weren’t. At this point, we’re staying away from automatic fixes because the prospect is a little bit terrifying, but it’s something we may see in the future. If you basically control the inputs and the outputs—for example, saying, “Here’s a set of tools you can use, and here are the things you can do with them”—AI is really good at figuring it all out. These capabilities could eventually be used to support safe automatic repairs, but might require some modifications to existing utilities.
Powering shells: Language models are being integrated into shells and CLIs, enabling customers to enter natural-language instructions somewhat than the cryptic shell instructions which have been developed organically over the past 30 or 40 years—and which can be very difficult to remember, much less understand. On this case, the instructions are driving the working system, however sysadmins must maintain their ft poised over the gasoline and the brake to make sure that telling an OS to repeat a file or listing doesn’t end in, say, the deletion of a file or listing. With all that stated, I’m giving this a Degree 3 designation as a result of we’re beginning to experiment with asking AI to make adjustments to the shell that at the moment require looking the online to seek out bizarre strings of characters that you just copy and paste (after which pray will work). We’ve seen it work in easy use circumstances, however you continue to want a human within the loop—with the power to take management at any time.
Log evaluation: Log evaluation is tedious and tiring—like driving six hours on a freeway. Log information is actually free type, pure language. Taking people utterly out of the log evaluation course of could be irresponsible, however we are able to use generative AI to cut back the cognitive load massively, say, 80-90%. For instance, sysadmins may use generative AI to summarize 1,000,000 traces of log information to a few sentences. Or, a sysadmin would possibly analyze the log information utilizing RAG, and ask interactive questions till they get the reply they’re in search of, say, the reason for an issue they’re seeing. This could be used sooner or later to adjust to laws which require “studying the logs.” However, a human nonetheless wants to judge the information and make choices on what actions to take, which I’d say places this use case at Degree 3.
Producing config recordsdata: Producing config recordsdata is pure language processing, and pure language processing is one thing that AI innately does very well—a Degree 4 process if ever there was one, particularly whenever you constrain the inputs and outputs. In actual fact, I’d say that producing config recordsdata is similar as asking AI to translate a sentence from Spanish to English and even to generate an unique story with the theme of man vs. machine. However, whereas people would possibly need to write a poem, they most likely do not need to manually generate config recordsdata. Utilizing a language mannequin to carry out the duty is a large time saver that may probably trim a whole bunch of human work hours all the way down to only a few. With that stated, people should evaluation and validate recordsdata to make sure that they, for instance, deal with organization-specific elements or adjust to business requirements. People additionally want to ensure config recordsdata are documented to assist keep away from issues with future translation.
Updating config recordsdata: Updating config recordsdata is one other tedious job that nobody desires to do—the proper candidate for generative AI and one that may be carried out virtually 100% autonomously. Virtually. Sysadmins shouldn’t utterly depend on AI to find out what config choices have been deprecated and what new ones are in place—they have to be the ultimate arbiter of what’s OK and what’s not. Nevertheless, a machine studying mannequin can present assist alongside the best way and is about as near hands-off “driving” as sysadmins can get at the moment. Put it this manner: On an excellent day, when skies are clear and the highway is straight, sysadmins may use AI to replace (or generate) config recordsdata with out placing their palms on the wheel, gasoline, or brake. However on a nasty day, whenever you’re driving up a mountain, in a snow storm, and must swerve to keep away from a deer deciding which facet of the highway he’s going to run towards, sysadmins should be totally again within the driver’s seat.
Offering peer perspective: This one won’t do a lot for sysadmins’ social pragmatic expertise, however it may be finished with little human interplay. Wish to learn how friends have dealt with a sure problem or the factors they’ve used to judge a sure kind of know-how? The place sysadmins might have reached out to their human connections previously, they will now enter any state of affairs they need assistance with right into a generative AI device and get a great deal of recommendation, anecdotes, tales, examples, and instructions. Nevertheless, simply as when you’ve gotten a dialog with a human, it’s a must to think about the supply—and biases and potential for hallucination—whenever you “discuss” to AI. To be honest, in years previous, I have been extraordinarily pissed off with the steering my colleagues have given me, so mileage might range.
Degree 5 just isn’t attainable with automobiles at present, however there are occasions when Degree 4 is achievable—below sure situations. I wouldn’t put that a lot belief in autonomous driving throughout a blinding snowstorm or on a mountain highway with hairpin turns, however I’d on a sunny day driving alongside a protracted lonesome freeway in a desert in Arizona. The identical is true for system administration duties. The extent to which AI can assist sysadmin duties is growing shortly. However, in the long run, probably the most highly effective device in a sysadmin’s arsenal is not AI—it is the mix of AI and human experience. And that can probably all the time be the case.
