My hope is that there is a sort of Cambrian explosion of small software projects built by people who have absolutely no clue what they're doing. Many such projects will go nowhere, but some percentage of them will see success and growth. My second hope is that there will always come some threshold of complexity beyond which AI cannot effectively iterate on a project without (at minimum) the prompting of an expert in the field.

The combination of these two things could lead to a situation where there is a massive, startup-dominated market for engineers who can take projects from 0.5 to 1, as well as for consulting companies or services that help founders to do the same.

Another pair of hopes is that a) the LLM systems plateau at a level where any use on complex or important projects requires expert knowledge and prompting, and b) that because of this, the hype of using them to replace engineers dies down. This would hopefully lead to a situation where they are treated like any other tool in our toolbox. Then, just like no one forces me to use emacs or vim (despite the fact that they unambiguously help me to be at least 2x more productive), no one will force me to use LLMs just for the sake of it.