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.

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.

> 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.

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

> 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.

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.