The argument desk / evidence before certainty
Where the frontier
disagrees with itself.
Each dossier takes one consequential question, shows the strongest case on both sides, and gives you an instrument for testing the assumptions.
01Questions worth
Current investigations
Questions worth
getting wrong in public.
Dossier 02New · October 2026
When does an AI agent become an economic actor?
Is task completion enough—or does economic agency begin only when delegated authority creates consequences in a market?
Delegation threshold · market stress test · disagreement map→Dossier 01August 2026
Can an AI do a week of useful work?
Long-task benchmarks are rising quickly. Real workplaces introduce ambiguity, coordination, and costly mistakes.
Reliability horizon · workplace messiness test→Follow the argument
Return when the
evidence changes.
Receive new dossiers, forecast revisions, and the evidence that changed our mind.