,
8 tweets,
2 min read
Read on Twitter
The single *stupidest* VC trend of the past few years -- and yeah, I know, low bar -- has got to be AIOps.
Let's get specific.
Let's get specific.
There are some current success stories that blend some clever automation and a lot of NLP and call themselves "AIOPS" to get more money and (possibly) customers.
More power to them.
More power to them.
And I am not an AI expert. I am very far from such. I am but a humble ops engineer in the field, surveying the results.
Someday the machines *will* take over. But right now, AIOPS is stumped by change and ridden with false positives. It makes everything harder, not better.
Someday the machines *will* take over. But right now, AIOPS is stumped by change and ridden with false positives. It makes everything harder, not better.
Every time you ship code to the system, you change the system. All your fancy artificial learnings no longer apply until the learning has caught back up again.
Who is best equipped to debug that deploy? The engineer who wrote the diff that triggers the build, always.
Who is best equipped to debug that deploy? The engineer who wrote the diff that triggers the build, always.
What this suggests to me, is that AIOPS will continue to be useless until AIDEV. Until the AI has access to the original intent of the entity that generated the change. Because they wrote the diff.
Then AIOPS can know what changes to expect, and successfully validate them.
Then AIOPS can know what changes to expect, and successfully validate them.
Until then, all of this smells like the neverending pretension that greenfield development requires some special genius, while maintening, extending, scaling, and the rest of the lifetime needs of software can just be handwaved away for the robots to make do.
Which reflects our naive biases as well as what we wish were true. But reality? Not so much.
It might make more sense to have the robot spew out every possible combination of inputs, see which survive creatively, then turn our human ingenuity toward making them REALLY work.
It might make more sense to have the robot spew out every possible combination of inputs, see which survive creatively, then turn our human ingenuity toward making them REALLY work.
Anyway, my reaction to AIOPS is roughly exactly the same as my initial reaction to serverless: there's gold in those thar hills, sure.
But its branding reveals so much wishful thinking about how software gets made, I kinda don't want people to pan for it.
But its branding reveals so much wishful thinking about how software gets made, I kinda don't want people to pan for it.
