I have a journalist friend with 0 coding experience who has used ChatGPT to help them build tools to scrape data for their work. They run the code, report the errors, repeat, until something usable results. An agent would do an even better job. Current LLMs are pretty good at spotting their own hallucinations if they're given the ability to execute code.

The author seems to have a bias. The truth is that we _do not know_ what is going to happen. It's still too early to judge the economic impact of current technology - companies need time to understand how to use this technology. And, research is still making progress. Scaling of the current paradigms (e.g. reasoning RL) could make the technology more useful/reliable. The enormous amount of investment could yield further breakthroughs. Or.. not! Given the uncertainty, one should be both appropriately invested and diversified.