YHK - yet another hacker news clone

I had wildly different results. A lot of customer service sucked, other customer service was ok-ish. But in general it always takes away my time. Often a really well-designed manual, FAQ etc... solves most issues. Only in a few instances I had to rely on customer services.

my phone provider double-charged my credit card so i had to call in and get it reversed.

would you just accept the double charge, cancel your contract (and pay the cancellation fee), and then move to one of the other handful of providers?

Obligatory https://xkcd.com/810

Usually there is a taste about AI generated messages that are just quite obvious, it's usually pretty close to corporate speak, but not exactly the same.

At a bare minimum, this sentence was AI inspired. Likely generated. Which certainly casts the whole response into doubt (that it is human-written.) Reads more to me like a person gave a quick quippy response to an AI to clean up, then just took the output and replied.

The good thing about AI support is that absolutely flipping out in the chat actually does get escalated, unlike with a human who just digs in their heels.

I find it strange that they'd have an AI-driven followup email days after the survey. If you're automating it, why the delay? "Delay the response so it seems more human" is way too cynical.

It's worth noting that OP is talking about Ocado, too. I've been a client of theirs for over ten years, and the number one reason why I keep buying from them is that their delivery drivers never feel super-hurried. They have lower quotas than drivers for other services, and they always have time for a little bit of banter. Compare that e.g. to Amazon Fresh drivers, who'll always just leave the groceries at the door and knock. They're already gone by the time I open the door.

I find it unlikely that they'd automate this sort of second-level customer support response while not squeezing the everliving shit out of their drivers.

> It is rare to get feedback that tells us exactly what is broken and how to fix it. AI doesn't remember from one conversation to another, so probably can't say how rare something like this is at Ocado (although admittedly it could learn that it's generally rare in fine tuning). I also expect AI to be trained not to compare feedback with other people's feedback.

It's the equivalent of people a hundred years ago complaining electrical publications featured radio excessively, yes it'd be annoying but hardly off-topic or outside of their typical wheelhouse.

I’ve tried this with claude. I told it to output HTML that didn’t look like it came from claude and to avoid common claudisms. It didn’t work and still used claude boxes all over the place. I’m sure with enough time I could get it there, but at that point I may as well write the HTML myself.

I'm sure to some degree people (and AI being used to build AI) are kind of doing this thing but I don't yet think that OpenAI has a distinct flavor profile for their models that they have standardized on and are intending to lock into over time.

When I go to Starbucks I know what I'm going to get and it's been reasonably consistent over a decade plus and similarly that's true of other coffee shops which have different profiles. I'm expecting something along those lines.

/s

It's already pretty straightforward to improve this default "You're absolutely right about [your email reworded]" response but you still see it everywhere. The corpos simply don't care to fix it because doing so probably doesn't affect w/e metric they're modelling.

I think the real question to ask is not really about the tools themselves but where the organisation's incentives really lie. I've come to the conclusion that customer satisfaction just isn't up there anymore. You'll eat whatever garbage is served in your trough (from their POV)

It'll never work like that, because everybody else giving that same direction will get more or less the same result, which will itself come to be recognized as sounding like "AI`" for any popular model+harness.

What you'll always need to do is instill your AI with some idiosyncratic voice that distinguishes it from others, which you can achieve in a variety of ways (stylistic direction, audience characterization, style mimickry, personalized finetuning, etc). Presumably, tools for doing this will eventually make it into the harness products but anybody who doesn't use them is always just going to end up generating recognizably AI slop prose. It's the nature of the technology.

A secondary consequence (even if you just use "avoid sounding like AI") is that because the technology works its magic mostly through prose-streams in its own voice, these idiosyncratic voice changes will also nudge its behavior away from benchmark norms. You won't necessarily lose capability on any given task (you may even gain on some), but external reference points/measures about its capabilities will become less applicable.

No, you cannot. You might get a slightly different writing style, but it will still be the same brand of slop.

If it was that easy, don't you think the AI labs themselves would just add that to their system prompt?

Your puny local AI assistant will easily be outwitted by malicious agents with a million times the power.

I wrote a bit more about it at https://shkspr.mobi/blog/2026/08/asymmetric-agents/

You cannot pull that off if you're mid, and mid is exactly what the statistical average of humanity is.

It will try though, and it will just be an asshole. Which, funnily enough, is exactly the same failure mode as mid humans trying on that costume fall into.

There are two obvious ways to handle this.

The first is to charge for human support. If you want to pay $100 each time you make a human spend 15 minutes responding to your AI agent's query, it's your money.

And the second is that you put a chatbot out front that has memorized the FAQ and thereby successfully answers 97% of queries, but cause them to get better at identifying when they're being useless and route that small percentage of issues to human support.

I personally would rather get a literal auto reply ack e-mail generated by some python script than an LLM response that just rephrases the context. The python script will be more compute efficient too for this usecase.

Having said that, I would have used a 1-bit design to reduce ghosting and set up an option for pagination over scrolling; some e-ink devices handle scrolling fine but others don't.

It may burn off customer support engagement among customers who don't have the capacity or patience to keep pushing for help (this is a bad outcome for customers, but looks deceptively positive in vendor-side metrics), but for the remaining customers it just adds latency and work.

The chat would go like this:

- I have a question about "no vehicles in the park" policy. Does it include motorized wheelchairs?

- To guarantee safety and blablabla our policy is that no vehicles are allowed in the park.

- so can i go to the park with a wheelchair with a battery and electric engine?

- To guarantee safety and blablabla our policy is that no vehicles are allowed in the park. Please see our safety policy (link).

I am far less annoyed by AI responses where I am told up front they are AI.

Just give me instant AI end to end.