I've seen that mentality and gone to bat to convince a boardroom that it's the wrong approach, when people were star-struck by the possibilities. Luckily I'm in a position as CTO of a (very non-tech, brick and mortar) company that entrusts me to manage their budget for new features, and prevent erosion of our software/logistics over the long term. And I've come down decidedly on the side of not having LLMs fuck with any schema or architecture changes or anything in the codebase that would touch upon business logic. When your code actually encapsulates business logic, which is often counterintuitive and full of weird exceptions, 90% of the code work is done by prior planning to map out all possible branches and the algorithms to assist employee decision making. The 10% that's actually writing code needs to be done by someone who understands the entire stack and business model perfectly. Some nice HTML/CSS fluff here and there is great to hand off to an LLM, and you don't need frontier models for that.

I shot down similar arguments in favor of outsourcing overseas for years. Outsourcing any critical logic to an LLM is even worse.

You’re in finance and you should know better… it’s not about recouping costs.

It’s about pumping up revenues, earnings and eventually cash flows (since they can’t keep raising money constantly) to support a nonsensical valuation.

This is wrong. Token costs for the same model rapidly collapse over a year. Hardware inflation is a thing but not bad enough to outweigh the massive impact of software optimizations.

Tokens billed at API prices are profitable for openAI and anthropic today and it only get more lucrative every month for them as their inference costs fall. If it weren’t for continuous massive training runs taking larger and larger capex, these companies would be massively profitable

The sad part is that you work in FINANCE of all things and this happens there.

Like: What competence do decision makers in FINANCE have, when they are this oblivious to economics?