The big promise of AI is that it will make us more creative. The concept is simple: give your tedious work to AI so you have time to be creative and generate big ideas. Sik Chung, Assistant Professor of Management and Human Resources at the Wisconsin School of Business, is assessing whether this promise will actually deliver.
“Thanks to generative AI, many people now fully automate repetitive, tedious tasks without pausing to reason about it. My research suggests that doing so can quietly cost them the expertise that makes their ideas useful.”
Chung sought to find an answer by studying the code review process. For each line of code, other developers identify mistakes and offer recommendations before it is pushed to production. The process can be tedious, making developers eager to hand it off.
Analyzing over 23,000 GitHub projects, Chung asked: when a bot takes over the line-by-line checking in the code review process, will developers generate more and better project ideas?
A Surge in Ideas, But Not in Quality
At first glance, the numbers say yes. Idea suggestions from a project’s own developers jumped about 15% after the project adopted an automated code-checking bot and averaged roughly 5% higher over time than in comparable projects that kept the traditional process. But more ideas are not the same as better ideas. To judge quality, Chung used a real-world verdict: whether the project leader actually accepted or rejected each idea. Through that lens, a twist emerged. In the first months, the promise seemed to be delivering: useful ideas surged right alongside the rest. Within four to six months, however, the boost in useful ideas disappeared entirely, while unusable ideas remained elevated a full year later. The lasting legacy of automation, in other words, was more ideas that went nowhere.
“What surprised me is that even such basic automation of tedious tasks weakened people’s ability to generate ideas that are actually useful,” Chung said.
Notably, the data ends in mid-2022, before AI coding assistants like GitHub Copilot spread, so the results isolate the effect of basic automation alone. Chung cautions that generative AI, which automates far more of the reasoning itself, may amplify the effect.
Breadth Without Depth
Why is this the case? The answer lies in what the bots changed about human conversation during the review process. Innovation research has long held that generating good ideas takes both breadth and depth of knowledge. Breadth is the fuel: conversations across many topics expose developers to more possibilities and new ideas. Depth is the filter: discussing code details helps developers stay current on what the project needs and what will realistically work. Analyzing the text of review discussions before and after bot adoption, Chung found that once a bot handled the code checking, developers’ conversations shifted: comments became shorter and shallower, back-and-forth exchanges declined, and discussions drifted toward bigger-picture topics such as new features and project direction. In other words, the bots tilted the balance toward more breadth but less depth. With more fuel but a duller filter, developers generated more ideas yet became less able to filter out the useless ones. Useful ideas faded while unusable ones kept coming because developers’ understanding failed to keep pace as projects evolved.
A Competitive Advantage in the AI Era
For employees, this is an opportunity. While it may appear that everyone around you is generating more ideas, many are likely inapplicable. If you can be the one who takes the extra time to learn constraints and maintain a deep understanding of a project, your ideas will be the ones that can be implemented. Chung shared, “Thomas Dohmke, the CEO of GitHub, has stressed that the ability to reason about and deeply understand systems will become even more important in the AI era, and that those abilities are developed through manual code reviews, running tests, and ensuring adherence to standards.”
Building Better AI-Enabled Workflows
The findings do not tell an anti-AI story. Rather, they point to the need for a more strategic approach to AI. Tedious tasks play a nontrivial role in building understanding, but that does not mean this is the only way.
“To be clear, I am not arguing that we should avoid AI for tedious tasks in creative work,” Chung says. “Adopting AI is a must because the efficiency gains are real and substantial. The question is how to adopt it well: how to design workflows that capture the efficiency of automation while preserving the hands-on engagement through which people build the deep understanding that keeps their ideas grounded.”
By automating tedious tasks, AI can increase idea generation. However, to generate more useful ideas, leaders will need to find new ways to build the expertise that foundational tasks have historically provided.