I can recommend using 3 LLMs at once for programming.
Ask them all to do the task, get them all to refute or agree with what the others say, let them all come up with design ideas or debugging then share their solution and synthesize into one outcome.
Honestly i find it quite useful that "single developer projects" may have extra hands on code (with manual human review) to get things done faster.
my personal pet peeve is the idea that "AI is better than humans" instead of "AI is useful as a tool", LLMs have too many caveats to be the last line of decision making in pretty much anything, but under supervision they are an amazing force multiplier.
https://news.ycombinator.com/item?id=44981747
https://news.ycombinator.com/item?id=44983570
https://news.ycombinator.com/item?id=44972151