It was progressing too quickly for user skills to keep pace. This halted about a month ago after the top tier of GPT became more able to analyze, correlate, and conclude (not just research and answer) spread sheet, image, text, and tool-based data. The potential customer base for this kind of thing is minute because the average human is unable to do this and so cannot instruct the machine to make use of this ability.

Keep in mind that what you see is a distillation of the capabilities that are being developed to facilitate recursive self improvement that may also happen to have mass market appeal and which are also inexpensive enough in terms of inference compute to deploy to people paying 10s or 100s of dollars.

We're mainly seeing computer programming tools right now. But we're also seeing multi-modal inputs, longer thinking/inference times, larger context windows, the ability to ingest irregular data formats, etc. It's trickling into accounting now.

The situation is that it's now good enough that medium companies are scrambling to use it. Large companies are spinning up groups tasked with redeveloping workflows from scratch as AI-First to try and self-disrupt before they're hit with it from outside. People are starting to figure out industry-specific evals.

There's going to be disruption this year. The thousands of software developers who were laid off. Some of them, and it doesn't take many, are going to start leveraging AI and rethinking knowledge work inputs and outputs as a series of continuous loops that humans interrupt. These are going to scale massively against the companies who fired them and they may end up being cheap enough and angry enough that they don't just get bought out and mothballed.