On October 28, 2026, InfoQ published a report summarizing the key points from Chelsea Troy’s keynote speech, “Leveling up as a Programmer in the Age of AI,” delivered at the Craft Conference. Troy pointed out that although generative AI has significantly improved task efficiency, software engineers’ core competencies—understanding system mechanisms and building deep cognitive models—still rely on deliberate, albeit inefficient, yet high-quality learning processes.
She categorizes software engineering practices into two modes: “figuring stuff out” and “doing stuff.” The former requires deliberately slowing down and confronting confusion and frustration, as it is precisely this discomfort that drives substantial cognitive growth; the latter tends to prioritize faster delivery, which, if misapplied in a learning context, can actually lead to stagnation in skill development. She emphasizes that AI tools should serve as “output catalysts,” not “learning substitutes”—even when using tools like Claude to generate compilers, the intellectual training gained through hands-on construction far outweighs the value of such automated outputs.
Drawing on neuroscience and learning science, Troy proposes five key strategies: maintaining highly focused attention; incorporating quality sleep to support spaced repetition; activating the brain’s thinking state through “outcome prediction” (first independently reasoning, then verifying answers); diversifying formats—such as reading, practicing, and explaining—to strengthen information retention; and urging managers to establish institutional safeguards via “upward management” to prevent work pace from encroaching on learning time. She specifically notes that when senior developers mentor junior colleagues, they must base their guidance on actual job performance rather than stereotypes, treating “changes in mindset” rather than “output quality” as the primary measure of learning.