Automating Foundational Work Risks Long-Term Organizational Expertise

The AI revolution is not failing because of a lack of capability. It is stalling because our organizations and educational systems assume that human struggle is the primary engine of value. As Ethan Mollick argues, we are trapped in a coexistence phase. We try to use powerful new tools within systems designed for a world where intelligence was scarce and effort was the only way to measure quality. The hidden consequence is a massive de-skilling event. By automating the junior work that historically forged experts, we are dismantling the pipeline for future expertise. Readers who recognize that this friction, or desirable difficulty, is a feature rather than a bug gain a competitive advantage by intentionally keeping human-led processes where others are blindly outsourcing their own cognitive development.

The Hidden Costs of Efficiency

The most important insight from this conversation is that we are misinterpreting the AI paradox. We see productivity gains in individuals, yet we struggle to find them in corporate bottom lines. Mollick suggests this is because organizations are garbage cans, or bundles of legacy processes, hidden dependencies, and evolutionary dead ends. When you inject AI into such a system, you are not just automating a task. You are disrupting a feedback loop.

> Every technology is both an extension and an amputation. A machine that extends human intelligence is also a machine that cuts off the need for humans to be intelligent.
>
> -- Derek Thompson

When junior employees use AI to bypass the grunt work of their roles, they are not just saving time. They are losing the apprenticeship cycle that creates tomorrow’s senior leaders. This creates a downstream effect where mid-level managers, also relying on AI, lose the ability to mentor or assess the quality of human work. Over time, the system loses its capacity to generate expertise, leading to a brittle organization that can execute tasks but lacks the deep knowledge required to navigate the jagged frontier where AI models fail.

The Myth of the Smart Tool

Conventional wisdom suggests that AI is a tutor that helps us master complex subjects. The reality, as revealed by recent data on math performance, is that AI acts as a crutch. When students use AI to solve problems, their performance on computer-assisted tasks improves, but their performance on unassisted tests plummets.

> If you use AI as a learning shortcut... you have not learned anything. We are very bad at knowing when we are learning then you destroy your learning opportunities.
>
> -- Ethan Mollick

This creates a systemic trap. We are building a labor force that is AI-native in the sense that they know how to prompt models, but they lack the foundational knowledge to verify the output. The competitive advantage goes to those who treat AI as a sparring partner, using it to push their thinking, rather than a replacement for the struggle of learning.

The Rise of Agentic Swarms

We are moving from a prompt-response model to an agentic model, where AI agents proactively solve problems. This shifts the dynamic from humans directing tools to humans being prompted by agents. While this offers immediate convenience, it creates a second-order problem: the swarm effect. When thousands of agents begin negotiating, shopping, and filing requests, they overwhelm the friction-based systems, such as customer service or compliance, that organizations rely on to manage demand.

> We are in a transitional state of working out this coexistence piece... we are negotiating not just with ourselves but with this technology.
>
> -- Ethan Mollick

Companies that rely on friction as a moat, making it intentionally difficult to cancel services or get support, will find their business models under siege by agentic swarms. The winners will be those who redesign their systems to provide value rather than relying on the difficulty of the user experience to protect their margins.

Key Action Items

• Audit your Gym processes: Identify which tasks in your workflow are desirable difficulties that build your expertise. Protect these from automation. (Immediate)
• Decouple learning from output: If you are a leader, create separate channels for AI-assisted output and human-only development. Ensure junior staff still perform the foundational work that builds intuition. (Over the next quarter)
• Shift from Prompting to Managing: Stop viewing AI as a search engine. Start treating it as an agent you must supervise. Practice teaching the AI by having it act as a student, which forces you to articulate concepts clearly. (Next 6 months)
• Build for the Agentic future: If you run a business, assume your customers will soon have AI agents negotiating with your support channels. Simplify your processes now, or risk being overwhelmed by automated requests. (12-18 months)
• Prioritize Taste and Agency: As technical knowledge becomes commoditized, focus on developing human taste, the ability to curate and judge which outputs are valuable. This is the one skill that remains scarce as AI output volume explodes. (Ongoing)