Developer advocates deliberate friction to counter agentic coding skill erosion

Developer and writer Vicki Boykis has published an essay titled "We should be more tired than the model," in which she argues that agentic code generation undermines long-term skill retention by bypassing essential cognitive processes. Boykis contends that while these tools offer immediate speed, they function similarly to slot machines, providing quick rewards without the mental engagement required to solidify programming foundations.

Boykis describes a personal experience of feeling a loss of control over her code and experiencing "brain fog" after relying on agentic generation. She notes that while the outward signs of having written code remain, the internal processes that occur during manual coding are absent. To counteract this passive consumption model, she advocates for reintroducing "deliberate friction" into development workflows, such as manually rewriting code generated by AI, to ensure a deeper understanding of the underlying logic.

The essay draws on cognitive science concepts, distinguishing between short-term memory, which processes information quickly like RAM, long-term memory, which acts as database storage, and working memory, which synthesises information to create solutions. Boykis posits that default agentic user experience affordances are antithetical to skill retention because they replace the internal cognitive processes of reading and writing code with a passive model akin to social media feeds.

She references specific advice from a user named "Oz" on X (formerly Twitter) to manually rewrite portions of code to counteract the passive nature of AI generation. This approach was further inspired by a paper on slowing down, thoughts on using AI to write code more slowly, and "Mitchell’s adoption journey." These elements collectively suggest that adding friction negates short-term speed gains but improves long-term proficiency by solidifying the developer’s own foundation rather than relying on foundation models.

Boykis acknowledges that it takes concerted effort to move from simply generating answers to using the tool deliberately. By choosing to work more slowly and manually, developers can maintain agency over their code and prevent the erosion of foundational skills that comes with over-reliance on automated generation. The piece aligns with broader industry discussions regarding the impact of artificial intelligence on developer experience and the potential risks to cognitive load in software engineering.