> If you ever work with LLMs you know that they quite frequently give up.

If you try to single shot something perhaps. But with multiple shots, or an agent swarm where one agent tells another to try again, it'll keep going until it has a working solution.