AI coding tools may be blocking junior devs from learning
The pitch is familiar: AI coding tools act as a personal tutor, closing the gap between junior and senior developers. A cluster of recent studies says that pitch is wrong, and a developer essay pulling them together lays out why the problem compounds over time.
What the research shows
JetBrains highlighted a study titled “The Widening Gap: The Benefits and Harms of Generative AI for Novice Programmers” that analyzed live coding sessions with varying levels of AI assistance. The headline finding was blunt:
“Participants thought it was like having a personal tutor. From the data in our study… we observed that they did not, in fact, use GenAI tools like a personal tutor. In fact, it was quite the opposite.”
Participants who leaned into heavy AI assistance often skipped crucial planning stages and finished with what the researchers called an “illusion of competence rather than true understanding.” The novice developers who performed best were the ones who greatly mitigated or outright ignored AI coding assistance.
In a separate large-scale 2025 study from UPenn, “Generative AI without guardrails can harm learning”, 1,000 students using an LLM to learn mathematics ended up performing 17% worse than students using only a textbook. The students using AI assistance thought they were excelling. The same study tested a “Tutor” mode where students asked for help and then solved problems independently. That group performed 127% better during AI-assisted practice sessions, though they scored roughly the same as the textbook group on the final test.
Anthropic’s 2026 study “How AI assistance impacts the formation of coding skills” reached a similar conclusion:
“For novice workers in software engineering or any other industry, our study can be viewed as a small piece of evidence toward the value of intentional skill development with AI tools. Cognitive effort—and even getting painfully stuck—is likely important for fostering mastery.”
The core problem
The essay’s author frames this as a structural paradox: AI coding tools demand expertise to use well, but they actively short-circuit the friction that builds expertise in the first place. The more experience a developer has, the more benefit they get from these tools, because they can steer, audit, and verify outputs accurately. The less knowledge they have, the more the tools can mislead them.
The essay cites David Cramer, co-founder of Sentry, on the compounding quality problem:
“You want to flex that you can generate all of your code and have hundreds of things going in parallel, I will flex and show you how broken the code is 100% of the time.”
Joel Spolsky’s 2002 Law of Leaky Abstractions gets cited here too: “The abstractions save us time working, but they don’t save us time learning.” The essay calls LLMs the ultimate leaky abstraction.
The distinction that matters
The essay draws a line between two modes of AI use that are easy to blur in practice:
- Cognitive offloading: Delegating the mechanical and tedious. Acceptable.
- Cognitive debt: Abdicating judgment and decisions. The trap.
Used as Socratic sparring partners rather than answer generators, the essay notes that studies have shown dialogic AI systems can meaningfully stimulate reflective, critical, and independent thinking. Used as production machines, they stall the formation of developer intuition entirely.
Six questions to check your own usage
The author offers a self-audit for any developer trying to stay on the right side of that line:
- Without the AI tool, could you still accomplish this task?
- Are you using the model to deepen understanding, or to expedite the answer?
- Could you adequately explain the generated output if asked to audit it?
- When learning something new, have you done enough research to know the right questions to ask?
- Have you cross-referenced the approach through documentation, search, or community sources?
- Is this a genuinely rote task, or does it require executive decision-making somewhere in the process?
The closing argument is pointed: if domain knowledge atrophies across the developer population while AI tools handle production coding, Sam Altman’s stated vision of selling intelligence back to users on a metered basis becomes a structural dependency, not just a business model.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
