By Seb Joseph  •  February 10, 2026  •

Ivy Liu

By the end of the year, about a quarter of revenue at its Monks arm is expected to come from what it calls subscriptions — not in the Netflix sense but as a commercial model where, instead of selling hours, the agency sells ongoing access to a bundle that combines senior talent, AI workflows, agents and institutional knowledge for a steady, recurring fee.

Those agreements typically run for at least a year, with the heavy early lift of setting it all up folded into the subscription rather than billed as a one-off project, even as the tools and systems behind the work continue to improve over time.

From there, the deal works less like a fixed checklist of tasks and more like a service that gets better over time. As the AI system improves, the agency can produce more work or do it faster without having to rewrite the contract. Those efficiency gains don’t show up as fewer hours billed, they show up as more output inside the same fee.

As Monks’ co-founder and chief AI officer Wesley ter Haar put it, something that once delivered “50 of these [outputs] a month” might reach 70 as models get better and pipelines get smarter, creating ongoing value within the same commercial contract.

“With the hunger for compute about to explode do we then have to add pass through [costs], which in an ideal world, as an agency, you don’t want to do because it complicates the ability for a client to sign off on a contract if you have every variable as a pass through cost. Procurement teams hate that. So we try and wrap that up into our subscription model,” said ter Haar.

Clearly this stops well short of the outcome-based structure agency CEOs describe as the endgame. At most, subscriptions are a waypoint. The pressure behind them is more concrete. AI now handles work that used to take junior staff hours, while also introducing new, less visible costs in the form of model usage and compute. In that environment, charging by the hour fits the work less and less, pushing agencies toward models that reflect systems rather than time.

“Just like software, the subscription gets upgrades, which is why we say it’s ongoing,” said ter Haar.

Those upgrades play out across two interlocking tiers within the bundle.

On the human side, AI is absorbing more junior-heavy production, giving clients greater access to senior operators with more flexibility in how time is deployed. Hours allocated for one quarter can roll into the next, allowing the work to follow momentum rather than the calendar.

Alongside that sits the machine layer: repeatable marketing workflows powered by LLM-driven agents handling both standardized and bespoke tasks, supported by tiered knowledge bases that combine task logic, client data and broader industry intelligence. Together, these systems compress research and production time, even as most work remains under a people-in-the model rather than fully autonomous execution.

“The billable hour does not allow for any meaningful innovation, which clients understand,” said ter Haar. “They’re open to new models.”

So far one client has brought into it since last fall, and two more are expected to follow before the quarter closes.

“My goal for this year is to have about 25% of our revenue running in this model, and that is based on hopefully having three quite sizable clients signed up at the end of Q1 and then having a decent amount of new business that starts in that space,” said ter Haar.

What happens after that depends less on vision than mechanics. AI usage must remain economically manageable as tokens and inference costs fluctuate. Agencies have to decide whether to absorb or expose those tech expenses. Clients and procurement teams must accept a model that behaves differently from traditional scopes. At the same time, large legacy contracts inside organizations still built around billable hours have to unwind slowly enough for both sides to adjust. There’s also the structural risk in tying pricing too closely to raw AI consumption. The very technology meant to make work more efficiently can introduce volatility that neither side fully controls, complicating efforts to make the model feel predictable. 

“Token costs are falling while usage is exploding — it’s an unstable foundation,” said Robert Webster, founder of AI marketing consultancy TAU Marketing Solutions. “It would be wrong for an agency to try to arbitrate that situation. Yes, AI usage is going to be really important as are the subsequent inference costs that it generates but I fear getting stuck in a world where large parts of the industry charge an arbitrage on tokens.”