How do you think Primer will evolve as agents get better at long-horizon tasks? The milestone and verification loop make sense now as agents fail unpredictably

How do you think Primer will evolve as agents get better at long-horizon tasks? The milestone and verification loop make sense now as agents fail unpredictably on complex tasks. Do tasks just get bigger and milestones scale with them?