A new model and a harness to contain costs: the AI market takes on token economy

Writer, a developer of enterprise AI tools, presented a new model and an upgraded harness — the software rigging that controls how many tokens agents spend on tasks. The very fact that cost containment became the headline of a product launch speaks louder than any figures: the industry has admitted that AI bills have slipped out of control for many customers.

The cause lies in the nature of agents: unlike chat, where spend is bounded by the dialogue, an agent works in cycles — reads, thinks, tries, re-reads. A poorly framed task can spin up a spiral of thousands of requests, and the cost of one operation on a sloppily configured agent differs from a careful one by tens of times. Hence the demand for a harness: per-task limits, per-process budgets, automatic loop cutoffs.

For a business adopting AI, the lesson is best absorbed before scaling, not after. When choosing a platform or contractor, ask not only "what can the agent do" but "how is its spend bounded": are there budgets, limits, per-process cost reports. Inside the company, appoint an owner of the AI budget — just as someone owns the telecom and cloud budgets. Agent economics is manageable — but only for those who manage it.