Engineering Teams Survive The AI Rush By Protecting The Struggle That Builds Experts

Most conversations about AI in engineering orgs fixate on the productivity math: how many hours saved, how much output per head, or how fast a senior developer can now move. That framing hides a harder question about where the next generation of senior developers comes from. Expertise has always been built by doing the work, by struggling through the research and the failed attempts until judgment forms. If AI absorbs the tasks that used to turn juniors into seniors, then the immediate gains may be real while the mechanism that produces expert talent erodes underneath them. The risk is that AI dissolves the learning pipeline that makes people good at those jobs in the first place.

Joseph Spano is an AI Product Developer who leads developer growth and ecosystem engagement at DevHub within NEAR Protocol. He helps build the systems that turn curious developers into productive ones. Across roughly fifteen years shipping software, he's come at reliability from an unusual direction, drawing on performance psychology and the literature on how humans acquire expertise and writing publicly on topics like agent memory and cognitive architecture. That lens leads him to look past AI's output metrics and at the developmental cost underneath them.

"If what makes me an expert is doing the work, and I'm offloading all of the work, then where does my expertise come from?" he asks. The question points at something most productivity conversations skip over, which is what all that work that's being eliminated was building.

The work is about more than output

Spano's central observation is that the labor AI now absorbs was doing double duty all along, producing expertise alongside deliverables. He describes his own recent experience as a shift in kind, not degree: he spends less time doing the active work and more time directing an AI to do it. "I feel like I'm a manager of my AI, because nobody really writes code anymore," he says, describing planning and development across the organizations he works with flowing through AI systems rather than hands on a keyboard.

The catch Spano recognizes is that the struggle he's spared was part of how he learned. He points to the familiar idea that mastery comes from hours of doing, and follows it to its uncomfortable conclusion. Remove the doing and the source of expertise goes with it, even for someone who's already an expert and stops actively practicing.

Skip the work, miss the errors

There's a second cost that compounds the first. When a professional no longer does the underlying work, they also lose the footing to judge whether the AI's output is any good. "You don't know if what the AI is telling you is even truly correct, because you don't have that basis," Spano points out. The longer the offloading continues, the deeper the gap becomes.

The erosion is subtle. Spano describes the loss of the whole cycle of learning: the feedback loops, the disagreements, the failures a person notices and corrects for. When the AI fails, he notes, a team may not even catch it, because the people positioned to do so are no longer building the judgment that would let them. Continuing education stops being something that happens through the work, because the work is gone.

No one knows who to hire

Spano is blunt about the challenge of building a team from the ground up under these conditions. "How does somebody go from becoming that junior developer to a senior developer? What is now the pipeline for someone to become a software architect?" For hiring managers, this turns into a concrete and unresolved problem, and it extends beyond developers. Spano points to the legal field, where junior years were how burgeoning lawyers absorbed the nuance of the craft. "How do you go from being a junior associate to being a partner when you're not putting in the hours? With AI, everything is just being answered for you."

He's watched organizations respond in ways that make the near-term math work while deferring the long-term problem. Some have effectively stopped hiring recent graduates, distrusting that they can validate AI output or even know the right questions to ask. Others consolidate, keeping one senior person to do what used to be three roles by leaning on AI, which raises its own question about burnout that Spano believes we're only starting to see play out.

Heavy AI leverage with a human on top doesn't hold

Spano is skeptical of the popular directive to treat the tool as an assistant and act like a manager, citing published reports of how over-reliance on AI is linked with organizational knowledge decay. "When you take that approach of, 'We're just going to heavily leverage AI and put a human in front of it,' it doesn't go well," he says. The missing ingredient is the same one throughout: a person with enough hard-won judgment to prompt well, verify the output, and know when it's wrong.

This is why he frames the current moment as more disruptive than a simple wave of automation. "AI is not simply replacing jobs. It's replacing a whole thought process." Because the damage to the pipeline shows up a generation out, the incentives all point toward pressing the short-term advantage and discovering the cost later. The reckoning arrives on a delay, and it's one he believes may force real changes in how organizations train people and even in what a college education is for.

In his view, the teams that come through this well will be the ones that treat the learning pipeline as something to protect on purpose, before the results come in and prove them right or wrong. "You now have something that can fully offload your cognitive ability, if you let it. The whole challenge is instilling the importance of doing the legwork anyway," he asserts.