The introverts’ edge: How AI is leveling the developer floor

The introverts’ edge: How AI is leveling the developer floor A nervous junior developer comes to work feeling neglected and spends time playing ping-pong rather than asking a senior developer for help for fear of looking bad, which only makes them fall further behind. The emergence of AI coding assistants has made that scenario less frequent — but the full picture is more complicated than that. Neel Sundaresan, GM of Automation and AI at IBM Software, was part of the team that built the original Microsoft GitHub Copilot before joining IBM and led the team that delivered IBM’s Bob coding assistant. Sundaresan didn’t need prompting on this emerging question from The New Stack. He’s clearly been sitting with it. His observation is that the entry-level developer experience at most companies is already broken, independent of AI. “Junior developers come in, they’re put on testing projects, documenting, taking on existing code and maintaining — that boring job because you don’t trust them so much,” Sundaresan tells The New Stack. “If they are not bold, if they are not extroverts, they will waste their time for the first three, four months. Maybe playing ping-pong or just browsing the web. Because the managers don’t have time for them.” “If they are not bold, if they are not extroverts, they will waste their time for the first three, four months. Maybe playing ping-pong or just browsing the web. Because the managers don’t have time for them.” Darko Mesaros, Senior Principal Advocate at AWS, recognized the pattern immediately — he was a junior once himself. “You tend to get shunned by some older colleagues,” he tells The New Stack. The shift AI brings, in his view, isn’t primarily about code generation. It’s about access to a source that doesn’t judge. “No question is a dumb question,” he says. “You can start doing the job you like basically from the get-go.” Andrew Cornwall, an analyst at Forrester, puts a finer point on it: AI coding tools help juniors get over the blank-screen problem — that paralysis when you don’t know where to start, he says. And juniors can ask an AI about code in a way they wouldn’t be comfortable asking a senior developer. But Cornwall calls out the trade-off. “Juniors might be getting answers to a specific problem, while talking with a senior developer might give them additional perspective on architecture or development process that they wouldn’t get from a chatbot,” he tells The New Stack. A hidden education system The tool Sundaresan describes isn’t eliminating the broken onboarding pathway or the path to knowledge so much as restructuring it. With Bob, IBM was able to assign FedRAMP compliance work — previously requiring principal engineers — to developers with one or two years of experience who worked alongside the tool. Bob presents multiple solution paths with trade-offs, so developers absorb the reasoning even when they implement only one option. “It’s kind of like a hidden education system,” Sundaresan says. IBM has distributed Bob to 80,000 users internally, and “as we’ve gone on, junior developers think of Bob as a distinguished engineer sitting by their side and guiding them — because I’ve seen this [behavior] at Microsoft and other places,” he tells The New Stack. At AWS, the story looks similar. Amazon’s internal tool of choice is Kiro, AWS’s own coding assistant, available to developers in both the console and IDE from day one (Amazon added Anthropic Claude Code this week, and OpenAI Codex arrives next week). But the more significant shift, Mesaros says, is how Amazon teams now build their codebases and documentation — not just with AI, but for AI. “They build their code and their documentation and everything else around it for AI,” he says, so that coding assistants understand not just the code but the way the team operates it. The maternity leave story Sundaresan tells another story that he says left him emotional. An IBM engineer who returned from maternity leave after 4.5 months told him she felt disoriented and afraid to ask colleagues basic questions, and was seriously considering quitting. She got access to Bob, the coding assistant. “Bob is never going to complain. He’s not going to call me stupid or make judgment,” she told him. She decided to stay on the job. That’s just one anecdote. But Sundaresan says it points at something real about what makes junior developers leave before they should — and it applies beyond juniors to anyone returning to a codebase after time away. What “junior” means is changing The term itself is shifting. “A junior developer 25 years ago was different than a junior developer five years ago,” Mesaros says — not just because of tooling but because of languages, frameworks, and practices. A developer who once would have written 68000 assembly now builds full-stack applications solo. With AI, that baseline keeps moving. The gap AI doesn’t close, Mesaros argues, is systems thinking — understanding how a piece of code fits into a complex, internet-scale architecture. When presented with the idea in a recent ACM paper by Microsoft’s Scott Hanselman and Mark Russinovich — with a caveat. “These junior developers will not be developers who don’t know how to write an error-correcting function,” he says. “They need to be mentored in how to approach it with systems thinking.” “They need to be mentored in how to approach it with systems thinking.” Cornwall reinforces this from a different angle. Code review used to be a primary channel through which senior developers actively invested in junior developer growth. AI-based review tools now help juniors fix obvious issues — such as naming conventions and style — before the code ever reaches a senior. That can be a good thing, focusing reviews on architectural concerns. But it also changes the texture of the relationship. The double-edged sword The risk Cornwall identifies is structural, not individual. Some organizations have concluded that a senior developer with several AI agents is just as productive as one with several junior developers. “In those cases, AI is making it harder for juniors to get the experience they need to become seniors,” he says. That’s the tension the optimistic framing tends to skip over. AI may lower the barrier to becoming a developer while simultaneously narrowing the path from junior to senior inside organizations. AI may lower the barrier to becoming a developer while simultaneously narrowing the path from junior to senior inside organizations. Not everyone shares that concern. Tools like GitHub Copilot, which are free and widely accessible, may be creating a generation of self-taught developers who don’t need an organizational context to start building at all. Some observers argue that juniors bring something AI can’t replicate regardless: they aren’t saddled with how things have always been done, and in an industry defined by novelty, that matters. Experience that used to take six months of project exposure now requires roughly six days An AWS spokesperson puts the speed-compression issue plainly: Experience that used to take six months of project exposure now requires roughly six days. Mesaros, who has used coding assistants since their early days, framed the broader shift this way: “It felt like, ‘Oh, this is just a shortcut for me to do a thing.’ But now, it’s so much more than a shortcut. It can enable me, even as a senior developer, to do so much more.” That suggests that while the ping-pong table isn’t going away, the reasons junior developers end up there will look very different going forward.