Dans: stop building digital employees and build intelligent companies

Note: the downloaded content is incomplete. It ends at "Expand to continue reading," meaning it only includes the opening of the article, not the full text. This summary is based solely on that portion.

In his Fast Company column dated October 24, 2026, Enrique Dans questions an industry trend: turning AI agents into simulated employees. He describes a race to "build the org chart in silicon."

The author cites three companies as examples. According to him, Microsoft gives its agents identities, sponsors, access rights and employment-like lifecycle management, and governs them much as it does human identities. Salesforce describes agents as "digital workers" and promotes the idea of a digital workforce that mimics human capabilities. ServiceNow has introduced individual AI "passports" that specify what each agent can do at runtime.

For Dans, this reproduces human organization in software instead of asking what a truly machine-native organization would look like. He also criticizes companies and executives for boasting about the number of agents they oversee, as if it were a badge of honor. He adds that, as he often says in his innovation courses, learning is much harder than unlearning.

His central argument is that human organizations have roles and hierarchies partly because of our limits on attention, memory and communication. Software has none of those constraints and has no reason to inherit them. He acknowledges that identity and permissions are necessary and that it can be done this way if one wants to, or doesn't know any better. But he warns that this anthropomorphism will end up as technical debt. A growing set of increasingly autonomous agents can recreate silos, develop conflicting incentives and generate coordination overhead, now at machine speed.

The proposal glimpsed in the excerpt is to treat agents as temporary execution processes. They would be instantiated for a specific purpose, scoped, observed and discarded. Dans stresses avoiding knowledge, identity or operational history becoming trapped inside specific agents. Just as models should be interchangeable components, so should agents.

The rest of the argument, including the full "intelligent enterprise" alternative promised by the title, is not in the material received and is therefore not summarized here.