FYI: gpt-oss:120b is better at coding (in benchmarks and my own anecdotal testing) than gpt5-mini. More importantly, it's so much faster too. We need more of this kind of optimization. Note that gpt5-mini is estimated to be around ~150 billion parameters.
Who is the "we" in this sentence? The ultra-rich that don't want to pay white collar workers to build software?
The advantages of LLMs are tiny for software engineers (you might be more productive, you don't get paid more) and the downsides are bad to life-altering (you get to review AI slop all day, you lose your job).
This is already a fact and it's set in stone - making AI cheaper won't change anything in that regard. However, a cheaper AI will allow the laid-off software engineers to use models independently of those firing them, and even compete on an equal footing.
My ability to create software is only really useful to me because other people pay me for it. If AI agents take that from me, it won't matter that I can now create awesome software in minutes instead of weeks. The software was never really the thing that was useful for me.
That assumption is wrong, humans will always have to guide the bots to achieve human-worthy goals, be it as supervising engineers, small business owners or a combination of the two - that helps competition and avoids lock-in.