5 minBusiness
Companies should build intelligent organizations, not digital employees, argues innovation expert
An innovation professor argues that the industry's push to treat AI agents as digital employees is a flawed architecture and that companies should instead build persistent systems that preserve organizational knowledge across models and agents.
The corporate technology sector is increasingly describing AI agents as digital employees, and an innovation expert is urging companies to reconsider that framing. In a commentary published by Fast Company, the author argues that the industry is rushing to build an organizational chart in software, reproducing human hierarchies instead of asking what a truly machine-native company should look like.
The author points to several recent product moves as evidence of the trend. Microsoft, according to the piece, gives agents identities, sponsors, access rights, and employment-like lifecycle management, governing them in a manner similar to human identities. Salesforce describes agents as «digital workers» and promotes a digital workforce that mimics human capabilities. ServiceNow has introduced individual AI passports that specify what each agent may do at runtime. The author notes that companies and managers have begun to boast about the number of agents they supervise, treating it as a badge of honor.
The central objection is that human organizational structures exist largely because people have limited attention, memory, and communication capacity. Software does not share those constraints, the author argues, and therefore does not need to inherit the same roles and hierarchies. Identity and permissions remain necessary, the piece acknowledges, but excessive anthropomorphism risks becoming technical debt. A growing collection of increasingly autonomous agents could re-create organizational silos, develop conflicting incentives, and generate coordination overhead, now operating at machine speed.
In the author's view, agents should be treated as temporary execution processes rather than permanent members of the organization. They would be instantiated for a specific purpose, constrained, observed, and then discarded. The piece argues that knowledge, identity, or operational history should never become trapped inside a particular agent, and that agents, like models, should be replaceable components. The author describes an agent as a software process equipped with an objective, a set of tools, access to context and memory, and some capacity to decide what to do next. Such systems should, ideally, generate feedback and learning loops. The employee metaphor, the author concludes, may be useful for adoption and access management but is inadequate as the foundational architecture of an intelligent company.
If agents are replaceable, the piece contends, the company's learning must reside elsewhere. The formal representation of the business, including its customers, products, employees, suppliers, contracts, and processes, along with the relationships among them, should persist independently. The author refers to these as «the nouns» and «the verbs,» a concept the piece notes is called «the ontology» by the data platform Palantir. The operational state of the company, meaning what is happening, what has changed, and what remains unresolved, should also be preserved, as should the objectives and trade-offs set by leadership, the permissions governing each action, and the rules determining when decisions must be escalated.
The article also calls for an institutional memory in the form of a journal in which every relevant decision, action, and outcome is recorded and that no agent or model can silently rewrite. Such a record would support future accountability and, the author notes, is increasingly relevant to cybersecurity. That record should feed back into future behavior so the company can learn from successes, failures, and unintended consequences. Models may be upgraded or replaced, agents created or retired, and workflows redesigned, but accumulated organizational experience must remain. Without that continuity, the author writes, a company is repeatedly hiring «intelligent amnesiacs.»
The piece concludes that the most important unit of analysis in corporate AI should be the organizational loop rather than a synthetic worker. Better models may improve individual cognitive performance, and additional agents may increase activity, but neither automatically makes a company learn coherently. The author predicts that the next enterprise platform will not resemble a workforce or management system for bots. Instead, it will be a persistent substrate in which models and agents can be exchanged while the company's accumulated intelligence stays secure and under control. The author says the product category that preserves state, authority, evidence, and learning across models and agents is still missing, and suggests it will emerge.
7
