---
title: Software engineers must slow down to build skills in the AI era
url: https://www.dataloco.com/en/software-engineers-must-slow-down-to-build-skills-in-the-ai-era
published: 2026-10-30T21:22:40+00:00
language: en
section: Developers
source: https://www.infoq.cn/article/Qwr9oqOSAdmi1YibHVWr?utm_source=rss&utm_medium=article
organizations: Craft, InfoQ
publisher: Dataloco
---

# Software engineers must slow down to build skills in the AI era

Chelsea Troy stated that software engineers must slow down their skill development pace to effectively learn in the era of artificial intelligence. Troy made these remarks during a presentation at the Craft conference titled Leveling up as a Programmer in the Age of AI. The core argument is that while tools and AI can accelerate the delivery of work outcomes, they do not necessarily support the cultivation of deep technical skills. Engineers must understand how systems operate to truly learn, and senior developers can guide junior colleagues by applying specific learning techniques.

Troy described software engineering as containing two distinct modes: figuring stuff out and doing stuff. When the goal is to do stuff, teams typically accelerate their pace. However, Troy argued that figuring stuff out requires slowing down. If execution depends on prior understanding, attempting to speed up the process can lead to stagnation. Engineers must shift into a lower gear, meaning they must slow down, to continue making progress. Activities that promote learning are often inefficient by design. They require high levels of attention and engagement, often forcing individuals to struggle with tasks they are not yet proficient at. Troy explained that developers must confront frustrating challenges and work through them. This discomfort, characterized by confusion and frustration, is essential for building cognitive capabilities.

Using tools as catalysts to accelerate outcomes generally does not benefit skill development. Troy emphasized that even with AI available, developers must understand how the systems they are modifying function. She referenced the neuroscience concept of spaced repetition, noting that skill development at its core requires highly focused attention interspersed with high-quality sleep. Troy suggested that spreading the learning process across several high-quality sleep cycles can help junior colleagues deepen their memory retention.

Troy also discussed the concept of outcome prediction, where attempting to guess an answer before seeing it helps internalize information. Learning truly occurs only when the brain is already engaged in thought. She advised that when learning how something works, individuals should pause at each step to ask what they think the next step should be and why. After making this prediction, they should continue learning to see the actual answer. This practice significantly increases the probability of internalizing the information. Additionally, Troy highlighted format diversification, which suggests that learning is more effective when information is encountered in multiple different formats rather than repeatedly in the same format. People often identify as kinesthetic learners because they read, listen, or watch content first, and then attempt to break things. It is this second or third exposure to the information that truly anchors the knowledge.

Generative AI has changed the landscape of software engineering skill development. However, Troy pointed out that the biological and psychological mechanisms of human problem solving have not changed as rapidly. She concluded that while understanding how tools work is helpful, it is more critical to understand how humans operate. Developers must continuously train themselves and guide junior colleagues using methods and strategies that align with how humans learn. In a subsequent interview, Troy noted that cognitive barriers to learning remain largely unchanged because human cognitive structures and potential have not fundamentally shifted. However, the expectation for an ever-accelerating work pace increases the risk of these barriers. Learning a discipline typically requires slowing down and immersing oneself in the material. When this is no longer feasible in the workplace, developers are forced to take shortcuts and never return to achieve their learning goals. This is a systemic barrier that managers have a responsibility to address by setting guardrails. Troy also noted that learners often confuse completing work with achieving learning goals. She used the example of assigning a compiler project to students, explaining that the goal was not the compiler itself, but the first-hand experience of designing it to better understand broader design issues. AI could write the compiler, but that would miss the point, as the compiler is merely a toy problem for learning how to think about more difficult issues. Reflecting on how one wants to change as a practitioner is the key performance indicator, not the quality of the practice outcome. Senior developers must observe junior developers' actual work, not rely on stereotypes, to provide effective guidance.