Why Traditional Management Falls Short with AI Agents: New Leadership Strategies Needed

Agentic AI is creating unprecedented management challenges that traditional business approaches can't handle. Research from MIT Sloan Management Review and Boston Consulting Group (BCG) reveals that AI systems that act as both tools and coworkers are breaking down conventional management logic.

The Core Problem: AI Defies Traditional Categories

Unlike standard technology that serves as either a tool or worker replacement, agentic AI functions simultaneously as both. This dual nature creates conflicts across departments:

• IT leaders need predictable, scalable systems with clear specifications
• CFOs require measurable ROI models and depreciation schedules
• HR executives need performance frameworks and supervision protocols for what feels like digital employees

Key Management Shifts Required

Investment Strategy Overhaul

Traditional tools require large upfront costs but deliver predictable returns. Agentic AI demands substantial initial investment plus ongoing variable costs for model training and updates. BCG's Sylvain Duranton predicts companies will spend more on technology than people as "the relative share of tech cost versus people cost will shift."

Supplier Relationship Management

Duranton recommends executives "strategize in terms of managing their portfolio of technology providers" since these costs will increase over time. The relationship with AI vendors may require union-like negotiations.

Timing and Deployment

Conventional replacement schedules don't work when AI systems simultaneously depreciate through model drift while appreciating through fine-tuning. Leaders must balance long-term capability building with short-term returns.

The Bottom Line

Organizations need new frameworks that treat agentic AI as neither purely tool nor worker, but as a hybrid requiring entirely new management principles. Companies that adapt their leadership approach early will gain competitive advantages.🔗 Read the full article on Computer Weekly