Don't optimize tokens — max-effort everywhere as policy

Deliberate counter-stance to A-025 ("medium reasoning by default") and B-073 (tiered model strategy): every agent session runs on the highest-effort tier of the strongest available model. Cap subscription cost only at the team level (if any one IC outgrows the base plan, the founder gets paged); do not cap per-task reasoning effort. Thesis: token cost today is small money relative to value and to the cost of a wrong answer; vendors will tighten limits later; spend the abundance window now.
Valid when all three hold:

• (a) founder-scale capital (LLM bill is a rounding line, not a budget line),
• (b) team size < 10 (every seat is a high-leverage operator, not a cost-center IC),
• (c) workload is non-customer-facing (no per-request unit-economics pressure from a paying customer).
  Tripwire — downgrade to a tiered strategy (B-073) when any of these fire:
• monthly LLM spend per seat exceeds ~1 day of that seat's loaded cost,
• a customer-facing workload starts dominating volume (unit economics now matter),
• vendor tightens limits and max-effort sessions start getting throttled in production.