From Managing People to Orchestrating Human and AI Teams: The New Leadership Challenge

In April, Shopify’s Tobi Lütke sent a memo to his employees that was quickly shared publicly. Before asking for more headcount or resources, teams would have to demonstrate why they could not get what they wanted done using AI. Using AI effectively became a baseline expectation, included in performance reviews.
In the same month, Microsoft’s Work Trend Index described the emergence of what it called the Frontier Firm, and introduced the idea that every employee may become an “agent boss”, managing AI agents as part of their daily work. It suggested that companies will need to think about a human to agent ratio.
A month later, Klarna’s CEO said the company would again invest in human customer service, after a period in which it had leaned heavily on AI. Quality, he acknowledged, had suffered.
Three signals in two months. Together they describe the new leadership challenge better than any framework.
Work is being redistributed
Generative AI and early autonomous agents are not simply new tools. They change who, or what, does the work. A team of eight people today may soon be a team of five people and a dozen agents, each handling a part of the process: research, drafting, analysis, reconciliation, first line customer response.
That changes the nature of management.
Delegation. Leaders will need to decide which tasks go to people, which to machines and which to combinations. That requires understanding the work in much more detail than most managers do today.
Accountability. An agent can execute. It cannot be accountable. When an AI system makes a mistake, a human must own it. Every human and AI team needs clear lines of who answers for what.
Performance management. How do you evaluate a person whose output depends partly on how well they direct AI? How do you measure the performance of an agent, and who is responsible for improving it?
Quality and judgment. The Klarna example shows that efficiency gains can hide a decline in quality that customers notice before management does.
Where human judgment remains essential
The most important design question is not what AI can do. It is what humans must continue to do. In my work with portfolio companies, the answer usually falls into three categories.
Decisions with irreversible or high stakes consequences. Credit, safety, legal exposure, relationships with key customers.
Work that depends on context the system does not have. Tacit knowledge, organizational politics, the history behind a client relationship.
Work that builds the next generation of judgment. If juniors no longer do the routine work, how will they learn? This is a question very few companies are asking yet.
Beyond technology adoption
The implications reach far beyond IT. Job design, organizational structure, career paths, skills development and culture all change when part of the workforce is not human. HR, operations and technology functions will need to work together in ways they rarely have.
For boards, the question is whether management understands this as an organizational transformation or treats it as another software rollout. The answer predicts most of the eventual value.
Leadership question: What does leadership mean when an increasing proportion of organizational work is performed by intelligent systems rather than people?


