Tokens and agents: AI shifts from being bought as a license to being metered as consumption, and SMBs must watch the cost

By Infochannel · November 10, 2026.

The article, written by Michel Braverman, founder and director of CADE Soluciones, is an opinion piece aimed at small and medium-sized businesses (SMBs) in Mexico. Its central idea: AI is changing how technology is consumed. Instead of buying licenses with fixed features, companies pay for model usage based on the amount of information they process and generate.

The unit of measure is tokens: fragments of text a model processes to read a query and generate a response. According to the author, in many services their consumption determines the cost. Braverman clarifies that they are not "the new currency" of business, but a way of measuring part of the spending.

The second focus is AI agents, systems that chain several tasks together to achieve a goal rather than answering a single question. The example in the text: receiving a customer request, checking inventory, preparing a quote, updating the sales system and scheduling a meeting. The author stresses that humans still set goals, authorize sensitive operations and oversee results.

The text lists uses in sales, customer service, administration, marketing and inventory, and warns that access to these tools does not guarantee better results. It recommends identifying which processes to automate, reviewing data quality and setting controls against errors, unnecessary spending and information exposure. Its conclusion is that the measure of success is not how many tokens are consumed or how many agents are deployed, but how much value is obtained.

Our take: the article is simple and explanatory, with no figures or measured cases, so it is best read as a framework for thinking, not as evidence. Even so, it points to a fundamental shift. With traditional software, costs were fixed and predictable; with tokens and agents, they become variable and proportional to usage. That brings freedom (starting small, without in-house infrastructure), but also a new risk: a poorly scoped agent that repeats tasks or gets stuck in loops can send the bill soaring without solving anything.

In general, that is the step that separates the pilot from governance. Many organizations are already testing agents, but few know how to budget for, audit and limit them. The risk of an agent is not just giving a bad answer, but carrying out a bad action, such as updating a system with wrong data. That is why the human oversight the author mentions is not a formality: it is part of the design.

We also share his competitive thesis, with a caveat. It is true that the cloud and on-demand models lower barriers that used to favor large companies. But that democratization benefits those with orderly processes and clean data; AI amplifies operational discipline, and disorder too.

In the short term, the transition will be uneven. The administrative and repetitive customer-service tasks the text proposes automating are precisely the most exposed in terms of jobs, and that should be said plainly. In the long term, if SMBs turn those savings into the capacity to serve more customers and focus on judgment, relationships and product, AI can be a lever for abundance and not just for cuts.

The practical conclusion is modest but useful: before asking which agent to hire, ask which specific process it improves, how much each run costs and how the result is measured. An automation that costs more than it saves is not innovation; it is expense.