What Frontier Models Mean for Small Agencies (An Operator's Math)
Agencies sell hours, and models sell capability. Running the actual numbers on my ten-person shop shows where the margin went and which services got more valuable.
The panic version of the frontier model story is that agencies die when clients buy the same tools. Two years into running a ten-person shop through this transition, my actual ledger tells a stranger story: some services collapsed, some exploded, and the dividing line is verifiability, not creativity.
The collapsed line is mechanical production. Transcription, first-draft copy, standard design variants, the hours we billed for drafts that clients now generate themselves in an afternoon. That revenue fell by more than half, and no amount of prompt skill on our side puts it back; the client's subscription ate it.
The exploded line is everything where being wrong is expensive. Strategy, brand systems, campaign architecture, compliance-sensitive content, and the quiet category nobody predicted: being the client's AI reviewer. Clients can now generate ten landing pages; what they cannot generate is the judgment about which two are worth shipping, and our review retainers now outearn what the draft work used to.
The internal math changed shape too. Junior hours shifted from production to evaluation, and our onboarding now trains new hires to interrogate drafts rather than write them. Billable volume is down; billable rate is up enough that the shop earns more per head than before the transition, with fewerall-nighters.
The operator's summary: models did not kill the agency, they killed the agency's assembly line and repriced its judgment. The shops in trouble are the ones whose entire offering was hours of production, and the honest move is to have that realization from the ledger rather than from aLinkedIn post about disruption.