i. Discovery is solved. The work after it is not
Every platform on the market sells search: filters, creator profiles, a database. Then it hands you a list, and the list becomes a spreadsheet, and the spreadsheet becomes the job. The expensive part was never finding creators. It is everything that happens once you have.
ii. Your judgment stays human
Agencies sell taste, relationships and meticulous vetting, and they are right to. Nothing here replaces a vetter's eye. It removes the clerical work wrapped around it, so the expensive judgment gets spent on creators worth judging.
iii. Measure the work, not the worker
Every action is a timestamped event, so capacity, speed and quality are derived from the work itself. Nobody maintains the planning numbers, because the planning numbers are a query over what actually happened.
iv. The maths decides, the model advises
A deterministic layer computes what is physically possible. A language model judges the shortlist. Then the maths clamps the model back to capacity. The failure mode of an AI assignment tool is confident overcommitment, and the fix is not a better prompt.
v. Own the data, do not rent it
Your creator roster is an asset your agency builds every day. Kept properly it compounds, survives renames, and costs nothing to read again. Rented, you pay for the same answer every time you need it.
vi. Say so when the system does not know
A number that cannot be verified reads "couldn't load," never a confident zero. A tool that is confidently wrong is worse than one that is honestly unavailable, because the whole point is that people stop double-checking it.