AI Token Costs: Ad Age Agency Pricing Analysis

The strangest budget line in creative work now looks less like a media buy and more like a parking meter. You do the thing that feels almost free, type a prompt, revise a line, generate a dozen options, ask for a cleaner version, and somewhere behind the curtain a counter advances. Nobody notices the counter during the brainstorm. They notice it later, when the work has become a bill. That is why Ad Age’s agency focused look at runaway AI token costs matters beyond the ad business. It is not really a story about whether creative teams should use generative AI. That question has already been answered in practice. The more uncomfortable question is whether agencies know how to price work when part of the labor is now metered in invisible, variable units.

The meter enters

the brainstorm Ad Age’s Asa Hiken wrote on July 30, 2026 that agencies are not immune to excessive AI token expenses. That framing is useful because agencies sit at the messy intersection of experimentation, client service, and margin discipline. They are encouraged to make AI feel magical for clients, but the magic is increasingly attached to a usage meter. Digiday describes the AI token as a new marketing currency, reporting that agencies embedding generative AI into their work are adapting pricing models to account for token costs. The publication also notes that developers such as OpenAI use tokens to meter AI compute, with both prompt text and generated output counted. That makes a familiar creative habit, trying many versions before choosing one, behave more like a variable cost center. This is the cultural turn hiding inside the accounting problem. The first phase of generative AI adoption rewarded curiosity. The next phase rewards restraint, instrumentation, and contract design. A team that treats every prompt as free will behave differently from a team that sees each prompt as a small draw on a shared operating budget.

The bill is not just usage,

it is behavior Forrester principal analyst Greg Zorella offers the most useful lens for why these bills feel slippery. Forrester says enterprises trying to understand token spend variances need to compare planned spend, actual spend, and what drove the difference. The drivers are not just tokens in the abstract, but price paid per token and quantity consumed, with multiple token types tied to different activities and different prices. That sounds dry until you map it onto agency life. A concept team may use AI for idea generation, a strategy team for research synthesis, a production team for versioning, and an account team for client recaps. Those may all look like AI usage in a dashboard, but they are not the same behavior. If the only governance rule is use less AI, the organization has learned almost nothing. The better question is where AI creates billable leverage and where it creates unpriced sprawl. A prompt that compresses hours of useful work may be cheap even when it consumes more tokens. A loop of cosmetic revisions may be expensive even when each individual request looks harmless. The unit cost is tiny, but the workflow design decides whether the system compounds value or expense.

The cloud lesson comes back wearing a creative hoodie InfoWorld’s David

Linthicum frames runaway AI token costs as cloud lessons being relearned. That comparison is doing a lot of work. Cloud computing also trained businesses to love flexible, on demand capacity before many of them learned how to govern it. The same pattern is now arriving in creative operations. The infrastructure is less visible, the interface is friendlier, and the user is not always an engineer. A copywriter asking for ten tonal options may not think of herself as provisioning compute, but economically, that is part of what is happening. This is where agencies can get ahead rather than merely clamp down. Token governance does not have to mean duller work. It can mean clearer briefs, better defaults, model choices matched to task value, and client agreements that distinguish experimentation from production. The point is not to make people afraid of the meter. The point is to make the meter legible enough that teams can design around it.

Pricing is product design now Digiday’s reporting that agencies are adapting

pricing models around tokens hints at the bigger shift. AI is not just another software subscription tucked into overhead. Once it is embedded into deliverables, revisions, research, production, and personalization, it becomes part of how an agency packages value. That means pricing becomes a product decision. Do clients buy outcomes, usage bands, AI assisted service tiers, or blended retainers that absorb reasonable experimentation? Do agencies expose token costs directly, or translate them into clearer scopes and revision rules? There is no universal answer, but hiding the cost forever is not a strategy. For readers outside agencies, the lesson travels well. Any team adopting generative AI should decide who owns the meter before the meter owns the workflow. Watch for dashboards that separate price from volume, policies that match model choice to task importance, and budgets that treat AI usage as part of operational design rather than a surprise line item. If the creative tool has become a pricing system, who in your organization is actually designing the price?