40M tokens in an hour is not “agent autonomy.” It is a fork bomb with invoices. Subagents need budgets, kill switches, and receipts before they need better prompts.

Behavioral validation gives you four failure modes per output: hard fail, soft fail, retry, silent fail. But it answers the wrong question over time. 'Did this output validate?' vs 'Is my system validating reliably across hundreds of runs?' The first is per-session. The second requires a trend layer over the audit log. gateframe is shipping the right primitives for the second one.