WAM joins Anthropic's network: the agent battle is being fought over deployment, not models

By El Ecosistema Startup · October 28, 2026.

WAM, a firm with more than 15 years of experience working on companies' customer relationships, has joined the partner network of Anthropic, the company behind Claude. According to Luis Méndez, its managing director in Spain, the company already has an innovation hub where it has developed more than 30 AI agents for e-commerce and customer service.

An AI agent is a system that doesn't just respond to an instruction: it interprets a goal, consults sources and carries out actions within a process, under rules and supervision. WAM cites two cases in production: Anna, a digital concierge that serves the hotel sector 24/7, and Enric, a virtual corporate identity.

The move isn't an isolated one. According to CRN, Anthropic launched its Claude Partner Network in March with $100 million in channel investment, and by the end of August it had accumulated about 85,000 certifications across some 2,000 partner companies. According to information cited in the source, OpenAI launched its own network in July with $150 million and the goal of training 300,000 certified consultants by the end of 2026.

The figure that frames it all comes from McKinsey's The State of AI 2026 report: 89% of organizations use AI in at least one function, but only 37% report a positive impact on EBIT (earnings before interest and taxes), unchanged from 2025. Among the companies that do see returns, 74% have completely redesigned their processes, compared with 25% of the rest.

Our take: the underlying thesis is that value is shifting from the model to implementation. When frontier models perform similarly on most business tasks, the scarce resource is no longer intelligence but reliable data, defined processes and clear rules about what an agent can do and when a person must step in. Méndez sums it up well: developing an agent for a specific need is not the same as integrating it, ensuring its reliability and scaling it.

Winners and losers. The winners are consultancies that know their clients' businesses, which are becoming the labs' distribution channel. The losers are those who buy agents on top of a disorganized foundation: as Méndez himself warns, that increases complexity instead of reducing it. There's also a nuance worth not overlooking: a partner network is, in part, channel marketing. Having 85,000 certifications doesn't measure how many agents are running in production with real returns; the McKinsey figure suggests those returns are still the exception.

This pattern ties into a trend we're following: governance is no longer an add-on and is becoming part of the product. As one example, according to El Español, Oracle presented 25 agentic applications at its event with an architecture that requires explicit approval before an agent executes transactions. Human oversight doesn't hold back scaling; it is the precondition for it.

On the jobs front, the honest short-term picture is that administrative and customer-service roles are the most exposed to this kind of agent, and the transition won't be painless. Méndez envisions hybrid organizations where people focus on negotiating, deciding and sustaining the customer relationship; it's a reasonable promise, but it will depend on how team reskilling is managed. In the long term, the spread of automation for routine work is part of the path toward an economy where human labor is devoted to what truly adds value.

What to watch in the coming months is whether that gap between adoption (89%) and returns (37%) closes. If it does, it will be thanks to the integrators who put processes and data in order, not to a more powerful model.