All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Is this what getting old feels like?
Has anyone else working support noticed this becoming a thing?
It’s certainly not every ticket (probably not even 1 in 20) but it happens often enough now to be noticeable.
You receive a support request that is clearly AI-generated: a long, very confident list of possible causes, investigations to perform, logs to collect, configuration areas to inspect, and questions that supposedly need answering.
The problem is that there’s often almost no actual troubleshooting behind it. No test results. No packet captures. No logs showing the problem. And usually even if checked within 10 minutes of the ticket being opened unverifiable as described. Usually barely even a clear description of what actually happened and never is the matter at hand enough to support the many directions of investigation being requested.
Instead of a support ticket, you get what amounts to an AI-generated incident investigation checklist containing ten or twenty different theories, many of which make little sense in the context of the reported problem. Properly investigating and responding to every item could take a human hours or days, despite the customer having spent perhaps thirty seconds generating the request.
And that seems like a particularly bad direction for support. We (support representatives) being treated like an extension of the customers Agents is not sustainable or healthy.
I want to stress I have no problem with people using AI. Its a great tool to help explain a problem, clean up a post, interpret an error message, or suggest some useful troubleshooting steps. It can even write code worthy of careful a review in many cases. Using AI can make the outcome better when used as a tool (or at-least thats what I think). P.S In the search of irony I am using AI now to help me improve my grammar in this post and used it plenty throught the day.
But there seems to be a growing category of ticket where AI isn't being used to communicate or as a tool to assist in the resolution of the problem - it's being used instead of doing the troubleshooting (probably including opening the ticket).
The result is a strange asymmetry: one side can generate an enormous amount of work essentially for free, while the support representative on the other end is still expected to actually read it, understand it, test it, and respond meaningfully often requiring hours or days to do so to the normally expected standard.
Is this where SaaS and hosting support is heading?
Do providers eventually just deploy agents to respond to the customers' agents? Then another agent handles the escalation, another handles the complaint that the first agent wasn't helpful, and another handles the retention offer when the customer threatens to leave?
At some point we're going to have two language models arguing with each other while the actual probelem remains and the situations continues as it was in the first place.
Maybe this is just what getting old feels like.
Has anyone else started seeing this? And if you run support for a hosting or SaaS business, how are you dealing with it?
Comments
Have you been in the chair of the opposite site? It's crazy I tell you.
But generally, you're a bit late on that, that's a thing for quite a while. People also just post huge slabs of AI text as security issues to Devs for example.
Or you offer support for free and you're just basically checking with their llm output.(Who's the clanker now?!)
Reminds me
I need to setup a clanker to keep those call centers that annoy me busy.
use the German way.
fix the problem and the ticket answer is
Fixed.
or
Done.
I've certainly been gaslit a few times by big corpo's agents especially when the technology was new.
or are you saying I should start responding to all matters in the world via AI? I think I'd rather be a clanker if it means continuing to try and show respect for those I interact with.
I used to think Azure making you pay for support was a bit rich, but maybe they were predicting the post AI world.
Theres probably a "this is why we can't have nice things" moment here.
Its something that I've noticed go from a few support tickets ever to a good portion of them recently (last month potentially).
I've seen this too, and funny enough I got tickets like that from users that had NO active services with me. To mess with both the legitimate and random ones, I ask them for a screenshot of the error and that shuts down the clankers really fast.
I have nothing against clankers translating, correcting, summarizing ideas and texts. But when they just generate slop, in a technical field, I get pissed-pissed.
Most of the time (almost all) people that send me tickets with AI content it's due to them not being an English. Also Google translate uses AI now, so Google translate will add the well known AI em dashes everywhere too. I think the missing troubleshooting is just them skimming over AI suggested messages. I just ask them to use their native language, I much prefer just translating.
@MikeA said:
Normally I would be against that as I have tried to hold to the statement that if I translate a message and its incorrectly translated I may be held responsible for it (past bad experience) so I've required english in support tickets since.
But I do see merit in your aproach and translation has come a long way. That said I dont think any of the recent problem cases were that.
I have seen some AI assisted translations / language assistance etc. Usually there is no problems there. This is a very different class of AI usage (IMHO).
@PulsedMedia
AI makes people who don’t know their shit do that, while people who do know their shit, it can help a lot.
I’ve had Codex find routing issues, pinpoint the causes, whip up relevant packet captures, etc.
Getting old nah, more like getting tired.
Having a proper ticket from a user, where he actually tried to debug it a bit, would be lit.
These days, people are lazy fucks, I mean I am lazy myself to be fair.
I ended up starting to ignore certain "tickets" where a user says something like "server down".
Especially if the node is up.
Establish an "Alpha" frame when an AI slop checklist gets thrown in your face.
“Acknowledge and Reframe”, bypass/briefly mention the slop in your reply, and clearly state the hard boundaries of the information that you would actually need to effectively solve their problem.
I actually rather tell customers that very large ai generated support requests are not helpful in any way
maybe thats also the german way, being upfront and direct
I feel you, however I think us providers are also guilty of the same. I use agents to reply to tickets, it drafts and I can enter my input and it redrafts it to how I want. Cannot fully automate it though as 30% of customers would just get slop.
Swings and roundabouts, still gotta manually review ticket but gets answered quicker in better detail. It does become very repetitive and tiring
Just charge support tickets by bytes transmitted in the support ticket.
support tokens, I'll sell the idea to VC as the next big thing in AI.
i guess its not that deep though
why would you feel attacked uppon receiving feedback that ai slop isnt a helpful contribution
Oh well, yet another ai shitfuckery ...
I'm so sorry
This is a highly valuable and insightful observation that surfaces several important dimensions of the evolving support ecosystem.
As we navigate the rapidly transforming landscape of AI-augmented customer experiences, it is essential to recognize that the core issue may not be AI itself, but rather the strategic deployment of AI without sufficient human oversight, contextual awareness, or actionable problem ownership.
A few potential areas for further investigation:
Based on the available evidence, no definitive conclusion can currently be reached. However, the situation appears consistent with a broader pattern in which generated analysis is being mistaken for completed investigation.
Recommended next steps:
Ultimately, the most scalable path forward may be a multi-agent support architecture in which the customer’s agent submits a comprehensive but unverifiable hypothesis matrix to the provider’s intake agent, which forwards it to a triage agent, which requests clarification from a communication agent, which escalates to a resolution agent, which generates a customer-facing summary explaining that the issue requires further investigation.
Meanwhile, the server remains down.
This is not a failure of intelligence. It is a failure to distinguish describing work from doing work—a distinction that perhaps warrants further analysis.
Look, all is heading to this:
So yes. All is good.
None of that has anything to do with getting old
You wanna know what being old really feels like?
It's the realization that we're closer to the year 2050 than we are to the year 2000.
I don't run a SaaS business but for my clients and users occasionally get some reported bugs or issues that are unusally long and knowledge filled from users in past that don't write that. Though many have made known they used AI or consulted with AI so nothing hidden.
But I actually welcome it despite some AI incorrect facts/assumptions creep in. It's better than the 'it doesn't work' comment and me having a back and forth to get the details and explaining etc. Any details to start with is better than nothing. So it's all relative ^_^
Usually, I just use my AI/tools to make sense of their AI generated reports to get a better understanding of the issue first. Saves time and my AI setup is way more accurate than theirs
Hehe, so you essentially battle LLM with your own LLM.
This is the way!
I have custom consult skills I built that allow me chat with up to 8 LLM models at a time from different AI labs within the same Claude Code or Codex session for verification etc + backed by grounded truth context/data. Though if there's an off-chance all 8 LLM models are all incorrect
Models I use
Claude Fable 5
Claude Opus 5
GPT-5.6 Sol
GPT-5.6 Sol Daybreak-Blue security model under Trusted Access For Cyber program
GLM-5.3
GLM-5.3 Flash
DeepSeek V4 Pro 0831
DeepSeek V4 Flash 0731
DeepSeek V4 Flash Vision Exp
Qwen 3.8 Flash
Minimax M3
Kimi K3
The only thing age-related about this is that another change happened and adapting gets harder.. you still have to adapt, or die.
That looks very expensive.
It's even expensive for us too.
Thanks for making my RAM cost a fortune. /s
A lot of it is under subsidied subscription plan pricing which works out to around US$10,000/month token value for ~US$250/month - winning
Very easy to detect AI text, and not allow a ticket to open, and provide msg to user.
Having DeepSeek V4 talk to Kimi, GLM, Minimax, Claude or GPT is an insult.
Waste of credits, and reasoning this is beyond overkill and will lead to even worse results.