This was one of the original (and strongest) arguments against XT, Classic, Unlimited, et al. If we simply increased the block size every time demand warranted it, services/custodians (and users) would never optimize their transactions. They would simply get used to increasing block size instead.
Segwit and Lightning aren't going to solve Coinbase's fractured UTXO set. And I suspect that's one of the reasons they've long supported hard forks. And we have to factor in onboarding costs as well (opening and closing channels). It's a bit early to say Lightning "will solve all those issues." On-chain transaction demand will only increase, so on-chain transaction optimization is going to remain extremely important going forward.