3 ms·

Thats last part is either naive or just shilling. The problems isnt the cogs, its the business owners setting up the incentives. Providing reasoned support is

by samrus 1y ago

Thats last part is either naive or just shilling. The problems isnt the cogs, its the business owners setting up the incentives. Providing reasoned support is expensive from a time and effort perspective, so they dont invest in it. Service reps are specifically trained and incentivized to just fuzzy match the SOPs And AI cant reason either, so it wont be able to provide any real reasoned support either. And the pwners dont care because they already provide shit service, people like you will just hype the AI up and then make excuses for it when its just as shit

random3 1y ago
Shilling what? AI in general? What’s the incentive? Hype is harder to argue. LLMs are defective attempts at AI, but that’s also a huge blind spot— effectively confounding AI and AI progress with a specific instance like LLMs. Business incentive— if you can make an argument on why would a corporation provide shit support to its own employees you win. Also make an argument on why support is actually good at Amazon and others like them. So I don’t buy your argument on incentives either. Although there are other implications related to incentives and AI in general, but those are off topic.
ethbr1 1y ago
The part that seems to be missing these days, that was historically present in support orgs, is escalation. You’re always going to do the bulk of support through L1, and L1 support is going to be limited in terms of training and access. But! The intended flow is that L1 recognizes when an issue exceeds their ability and escalates it up to L2+. And that’s where your talented (and empowered!) support engineers work. That escalation just doesn’t seem to happen any more. My hypothesis is that it’s a consequence of L1 being outsourced, where now they’re heavily incentivized to avoid escalating and impacting their probably-contractual KPIs. Yet another cost for adding too many parties to something that could be better if it were simpler.
- dragonwriter 1y ago
  > Yet another cost for adding too many parties to something that could be better if it were simpler. But its not a cost from the perspective of the companies doj the outsourcing: the incentives are the way they are because the result is intended. Its not a cost of adding an extra party, it is part of the reason the party is added.
  - ethbr1 1y ago
    I think you’re overestimating intentionality in bad outcomes. It’s entirely possible for a whole organization to want a good outcome, but construct an org chart where bad outcomes are guaranteed by the misaligned and incomplete incentives for sub-divisions. E.g. the IT support org that is only measured on total spend and time to ticket closure
    - lcnPylGDnU4H9OF 1y ago
      > It’s entirely possible for a whole organization to ... construct an org chart You changed actors mid-sentence. The workers in the organization may want good outcomes but they are near-invariably separate from the people making decisions for the organization.
      - ethbr1 1y ago
        No, I said it exactly as I meant it. The organization (including the bulk of employees and executive leadership) may want a good outcome, but sub-divisions’ (aka the org chart) may have have their self-interests set to inevitably produce a bad outcome.

by samrus 1y ago

random3 1y ago

Shilling what? AI in general? What’s the incentive? Hype is harder to argue. LLMs are defective attempts at AI, but that’s also a huge blind spot— effectively confounding AI and AI progress with a specific instance like LLMs. Business incentive— if you can make an argument on why would a corporation provide shit support to its own employees you win. Also make an argument on why support is actually good at Amazon and others like them. So I don’t buy your argument on incentives either. Although there are other implications related to incentives and AI in general, but those are off topic.

ethbr1 1y ago

The part that seems to be missing these days, that was historically present in support orgs, is escalation. You’re always going to do the bulk of support through L1, and L1 support is going to be limited in terms of training and access. But! The intended flow is that L1 recognizes when an issue exceeds their ability and escalates it up to L2+. And that’s where your talented (and empowered!) support engineers work. That escalation just doesn’t seem to happen any more. My hypothesis is that it’s a consequence of L1 being outsourced, where now they’re heavily incentivized to avoid escalating and impacting their probably-contractual KPIs. Yet another cost for adding too many parties to something that could be better if it were simpler.

dragonwriter 1y ago
> Yet another cost for adding too many parties to something that could be better if it were simpler. But its not a cost from the perspective of the companies doj the outsourcing: the incentives are the way they are because the result is intended. Its not a cost of adding an extra party, it is part of the reason the party is added.
- ethbr1 1y ago
  I think you’re overestimating intentionality in bad outcomes. It’s entirely possible for a whole organization to want a good outcome, but construct an org chart where bad outcomes are guaranteed by the misaligned and incomplete incentives for sub-divisions. E.g. the IT support org that is only measured on total spend and time to ticket closure
  - lcnPylGDnU4H9OF 1y ago
    > It’s entirely possible for a whole organization to ... construct an org chart You changed actors mid-sentence. The workers in the organization may want good outcomes but they are near-invariably separate from the people making decisions for the organization.
    - ethbr1 1y ago
      No, I said it exactly as I meant it. The organization (including the bulk of employees and executive leadership) may want a good outcome, but sub-divisions’ (aka the org chart) may have have their self-interests set to inevitably produce a bad outcome.