The IT department: Where AI goes to die
Ethan Mollick’s article argues that many companies are making a fundamental mistake in how they approach artificial intelligence. Instead of treating AI as a strange and transformative technology, they are trying to make it fit neatly into the same management logic used for ordinary enterprise software. In his view, this instinct to “normalize” AI may feel practical, but it strips away the very qualities that make the technology strategically important.
The article begins by pointing out the unusual nature of AI systems. A tool built to predict the next word in a sentence can also write code, generate business ideas, support decision-making, and even respond with a surprising level of emotional sensitivity. Because these capabilities do not fit traditional categories, organizations often respond by simplifying AI into something more familiar: another workflow tool, another efficiency system, another software rollout.
Mollick believes this “de-weirding” of AI is where the real problem starts. Making AI easy to use is not the issue; that is good product design. The deeper issue is that companies begin to think of AI only in terms of smoother interfaces, adoption targets, and routine productivity metrics. Once AI is reduced to a standard software implementation, leaders stop asking bigger questions about what it could change inside the business.
He illustrates this with the typical corporate behavior of assigning AI usage goals across the workforce. When companies require employees to use AI regularly, the result is often shallow or low-value activity: meeting transcripts, extra slide decks, more memos, and a flood of mediocre content. In other words, organizations may achieve adoption numbers without creating meaningful innovation. They use a powerful new capability to generate more administrative output rather than new strategic possibilities.
A second major argument in the article is that this mindset pushes firms toward automation instead of augmentation. When executives hear that AI can increase productivity by 30 percent, many immediately interpret that as a reason to reduce headcount by 30 percent. Mollick argues that this is the easiest and least imaginative response. The more valuable question is what becomes possible when workers can do dramatically more than before.
He gives the example of software development: if one programmer can now produce far more code than in the past, the most important strategic issue is not simply labor reduction. It is whether the company can now create products faster, experiment more broadly, or serve markets it previously could not reach. These kinds of opportunities require rethinking the organization itself. They cannot be discovered through vendor demos, standard consulting frameworks, or traditional IT procurement processes.
This leads to one of the article’s strongest claims: in many firms, AI is sent to the IT department, and that is often where it stagnates. Mollick is careful not to criticize IT professionals personally. His point is structural. Most IT functions are designed to reduce risk, control systems, and maintain security and reliability. AI, by contrast, requires trial and error, experimentation, and a willingness to tolerate uncertainty. Giving full ownership of AI to a department optimized for control creates a deep mismatch.
As an alternative, Mollick proposes a three-part model: Leadership, Crowd, and Lab. Leadership means the company’s top executives must personally own AI strategy. They cannot outsource the matter to middle management or technical teams alone. Senior leaders need to define how AI changes the identity and future direction of the business, not just how it improves current processes. They also need to create incentives that make experimentation safe and legitimate.
The second part, the Crowd, refers to employees across the organization. Mollick believes workers themselves are often the best source of useful AI ideas because they understand their own domains and daily problems. When they are given access to tools and real permission to experiment, they often discover practical and creative use cases that leaders, consultants, and even AI vendors did not anticipate. In this view, AI works best when experts in the business can actively test it in their own context.
The third part, the Lab, is a dedicated cross-functional team focused full-time on generative AI. This group should include both technical and non-technical people and should be responsible for pushing boundaries, inventing workflows, and feeding lessons back into the broader company. Mollick suggests that many large organizations still do not have such a team, which leaves them unable to build internal knowledge. Without a lab-like function, firms remain dependent on external hype instead of learning through their own practice.
The article closes with an important cultural warning: when companies fail to design the right incentives, employees may hide their AI use. Some worry they will be punished, monitored, or replaced. Others may already be working much faster and see no reason to report it. This creates an internal information gap where managers cannot see how AI is truly affecting performance. Mollick’s overall message is that AI should not be forced into old organizational maps. It is weird, powerful, and uncertain, and only companies willing to confront that reality directly will be able to unlock its real value.