Choosing how to construct a block is a highly opinionated/difficult thing to get right.

Even if you just want to maximize fees collected, you soon find that you need to build something akin to cluster mempools in order to graph the all the chains of CPFP transactions.

Additionally, every block, and every new tx in your mempool could potentially alter the tx set that gives you the most fees. And you need to find this optimal set in fractions of a second in some cases. Its a hard computer science problem in general to optimize.

You could just use Bitcoin Core's block template, but you can usually get more fees by using your own advanced template.

Since bitcoin core's template is suboptimal, and building a template from scratch is enormous work with a high penalty for getting it wrong, people tend to outsource it.

Outsourcing the template creates centralization pressure.

Stack on top of this, the fact that some pools are accepting tx fees out of band. Like Antpool's accelerator program, or Luxor pool for 4Meg-er inscriptions. These pools can only offer this service because they are large and create their own templates. The "rich" get "richer" in terms of the hashrate pointed at their templates.

A consequence of being a big pool, is getting regulatory pressure. The most popular block templates are likely to also censor sanctioned txns.

Luke believes that "good" miners will "do the right thing" despite what the incentives are. He hates the players not the game.