That's an interesting point about team composition. Companies originally split frontend from backend to scale up the number of developers they could put on projects. The idea was that each system could be a black box to the other team, and you may only need one person that understands it all end-to-end. The problem was that by splitting monoliths into multiple services they basically lost most of the advancements the industry had made to increase developer velocity. If your org was trending towards hyperscale, then you would have been transitioning to a multi-service architecture anyways, but for everyone else the move to SPAs resulted in a massive loss of productivity.
Listen, how impactful LLMs will be largely depends on the type of code teams are writing. If you're already building in a highly declarative manner, where your libraries are automatically handling most of the glue for you, then you may not have much of a need for writing code more quickly. These are the teams that are actually fast. Teams that already spend a bunch of time writing glue code will likely see significant improvements in velocity, because LLMs are good at regurgitating existing patterns. What they probably won't do is refactor your project so that your team starts operating on the level of the formerly described one.
Listen, how impactful LLMs will be largely depends on the type of code teams are writing. If you're already building in a highly declarative manner, where your libraries are automatically handling most of the glue for you, then you may not have much of a need for writing code more quickly. These are the teams that are actually fast. Teams that already spend a bunch of time writing glue code will likely see significant improvements in velocity, because LLMs are good at regurgitating existing patterns. What they probably won't do is refactor your project so that your team starts operating on the level of the formerly described one.