Work and agent-based AI: when algorithms enter the organisation - Il Sole 24 ORE

TEXT START: Agent-based AI is not merely automation, but a profound change in processes and leadership, bringing new challenges regarding accountability, collaboration and the promotion of professional expertise

The Dissection

The article correctly identifies agentic AI as a workflow-level transformation rather than ordinary software adoption. It recognizes that agents can execute tasks, use applications, pass outputs between processes and form a digital workforce.

But it converts a labor-displacement threat into a management redesign exercise. The central question becomes how managers should orchestrate agents, not what happens when agents make large portions of managerial and professional labor economically unnecessary. The article describes the first incision accurately while avoiding the hemorrhage.

The Core Fallacy

The article mistakes current human bottlenecks for permanent human moats.

It presents experience, relationships, creativity, contextual interpretation and exception-handling as durable human territory. Under the Discontinuity Thesis, these are temporary lag domains. As agents gain better memory, tool access, feedback loops and organizational data, many forms of judgement and coordination become instrumentable.

The manager-as-orchestrator is therefore a transition role, not a guaranteed endpoint. Allocating tasks, monitoring agents, auditing outputs, managing interdependencies and optimizing costs are themselves data-rich coordination problems. They are precisely the kinds of activities agents can progressively absorb.

The article also assumes that organizations can preserve substantial human-only domains at scale. P1 makes that unstable: when agents are cheaper and increasingly capable, firms face competitive pressure to automate. P2 prevents institutions from permanently sheltering broad categories of human work. P3 follows: a small number of people may remain indispensable, while the majority lose access to economically necessary labor.

Hidden Assumptions

• Human judgement remains scarce instead of becoming partially automatable.
• Ambiguity and exceptions remain outside computational reach.
• Oversight, compliance, privacy and quality control create enough durable jobs to offset execution losses.
• Middle managers adapt into higher-value roles rather than becoming an expensive coordination layer.
• Human development, trust-building and conflict management remain economically necessary at mass scale.
• Firms will redesign cautiously through measured pilots instead of accelerating deployment under competitive pressure.
• The people managing AI will control it, rather than becoming servitors operating systems owned by others.
• Role adaptation is the relevant survival test, while ownership and control of AI capital are left unexamined.

Social Function

Classification: partial truth, transition management and ideological anesthetic.

This is not pure copium. It accurately identifies process redesign, permissions, traceability, accountability and the emergence of a digital workforce. It also admits that many middle managers may fail to make the transition.

Its anesthetic function lies in presenting adaptation as the solution. “Balance calculation and judgement,” “orchestrate” agents and focus on uniquely human qualities: this gives professionals a survivable script while concealing the ownership question. The article never asks who owns the agents, who captures their output, or what happens to workers whose productive contribution is no longer required.

The claim that technology can bring the human factor back to the centre of leadership is especially misleading. Human activity may remain at the centre of legitimacy, liability and high-value exceptions without remaining at the centre of production. That is a premium niche, not restored mass participation.

The Verdict

The article is structurally perceptive but strategically evasive. It sees that agentic AI enters the organisation by replacing process steps and reorganizing authority, yet treats the surviving human layer as a stable destination rather than a shrinking bottleneck.

Under DT logic, the manager-as-orchestrator survives only while orchestration remains difficult for machines. Once agents can allocate, monitor, audit and optimize other agents, that role becomes another target for compression. The likely endpoint is not human-centred work redesigned around AI, but AI-centred production with a narrow class of Sovereigns, a thinner layer of indispensable Servitors and a displaced majority managed through transfers and residual niches.