Agentic AI Risks Repeating Management Consulting's Dependency Trap

The rapid adoption of agentic AI is creating a dependency trap that mirrors—and may exceed—the one management consulting created over the past four decades. As organizations rush to delegate cognitive work to AI systems, they risk hollowing out the internal expertise needed to evaluate whether those systems are delivering real value.

This parallel comes from Chinmay Kakatkar, writing in Towards Data Science, who draws on economists Mariana Mazzucato and Rosie Collington's 2023 book The Big Con to illuminate how structural incentives, not bad actors, drive organizations toward unsustainable dependence on external solutions.

Why it matters

Unlike previous automation waves that replaced manual tasks, agentic AI targets judgment, strategy, and creative work—the cognitive functions that define organizational identity. When companies outsource these capabilities, they lose not just the work but the knowledge needed to steer it. The consulting industry took decades to create this trap; AI is compressing the timeline to years.

The mechanism of dependency

The consulting industry's core trick was simple: organizations that repeatedly outsource strategic functions gradually lose the ability to evaluate the advice they receive. This creates a self-reinforcing cycle where less internal knowledge means more dependence on external expertise, which further erodes internal capability.

Agentic AI accelerates this dynamic. AI vendors currently price services below true delivery costs to capture market share—the same subsidized-pricing playbook that made cloud migration irreversible for many companies. Early wins in drafting text, writing code, and automating routine processes make adoption feel rational at each step. But as Kakatkar notes, when delegation becomes so pervasive that humans can no longer critically evaluate AI output, the user isn't augmented but replaced.

Individual cognitive costs

Research is documenting measurable harm. Recent randomized controlled trials found that participants who used AI assistance not only performed worse when working independently afterward but also gave up trying sooner. Another study of nearly 2,000 professionals found that while most agreed AI "did most of the thinking," the resulting ideas didn't feel fully their own. Brain imaging studies show measurably lower neural connectivity among people who used AI to write essays.

Kakatkar recounts MIT professor Micah Nathan's account of a student whose creative writing assignment devolved from a grammar check to a complete AI rewrite through incremental surrenders. Each step felt small; collectively they represented total abdication of authorship.

The vendor conflict

The same management consultancies that drove labor outsourcing are now collecting fees to advise on AI adoption strategies. McKinsey's projection of $2.6 to $4.4 trillion in annual value from generative AI conveniently justifies large-scale advisory engagements. The structural conflict is clear: advisors paid to assist with AI transformation have no incentive to help clients build the internal capability that would make future engagements unnecessary.

Vendors are also shifting their narrative from augmentation to replacement, reframing AI in terms of labor cost reduction to access larger budgets. The natural starting point is outsourced work, but strategically central roles may follow as positions are decomposed into automatable tasks.

The risk isn't that AI lacks value—used judiciously, it can free human attention for higher-judgment work. The risk is that the structural incentives driving adoption optimize for vendor revenue rather than sustainable organizational capability. As Kakatkar warns, by the time dependency becomes visible, reversing it may be prohibitively costly.

These details were first reported by Chinmay Kakatkar in Towards Data Science.