I’m not sure this is actually true. Technical debt can increase the amount of tokens that are needed to make changes to a codebase, and that will stay unless the models get good enough to rewrite the code completely while improving the architecture. Companies can believe what they want but profit is derived from the socially necessary human labor required to make a product. If the social necessity tends to zero because of AI, people will be able to use AI to make their own version of whatever software they need and it will not be needed to pay a company for it. But the most likely scenario in my opinion is huge codebases that induce more demand for tokens as models become cheaper to run, which means the cost savings may not actually materialize. Still, it is hard to say for sure at this stage.

However, humans can be fired, but Claude cannot. Humans are also more capable of generalized learning and following procedures meant to increase quality. Moreover, quantity of human output is roughly proportional to skill in aggregate, which can limit the damage that can result from incompetence if correct safeguards and procedures are put in place.

We are world leaders in per capita production of clowns, social murder, chronic disease, and climate collapse.