The Twilight of AI Vampires - by Ciprian BORODESCU

Chip On Shoulder 🙃

SubscribeSign in

The Twilight of AI Vampires<br>Do AI coding agents genuinely enhance developers’ cognitive abilities or do they merely increase output at the cost of cognitive degradation?

Ciprian BORODESCU<br>Jun 22, 2026

Share

Generated with Gemini 3.1 Pro<br>In this essay I’m exploring the use of AI coding agents as a contemporary form of cognitive augmentation among developers. Drawing on the distinction between systemic and personal perspectives on cognitive artefacts, I’ll argue that although the hybrid developer–AI system may increase execution speed and economic performance, these benefits do not automatically entail cognitive enhancement for the individual user. To investigate this tension, I’ve conducted a survey (89 respondents) and the findings indicate significant extrinsic benefits while also highlighting relevant intrinsic costs, including mental exhaustion, occupational stress, and a reduced capacity to recover after working hours. I’m proposing that, when developers are empowered to retain control over their collaboration with AI coding agents, and thus stay in a position to comprehend and actively guide the final output of the hybrid developer–AI cognitive system, this mode of operation can become a sustainable form of cognitive support.

Introduction<br>A study conducted in 2025, involving more than 48,000 people from 47 countries, shows that AI has already become an important part of everyday work, with 3 in 5 employees intentionally and regularly using AI to carry out their tasks, and nearly one third using it weekly or more often (Gillespie et al., 2025). Adding to this, the Anthropic report published in March 2026 places computer developers first among the occupations analyzed, with approximately 75% observed exposure to AI use (Massenkoff & McCrory, 2026). Then, in a recent interview, Marc Andreessen, co-founder and general partner at Andreessen Horowitz, co-creator of the Mosaic web browser and co-founder of Netscape, as well as author of “The Techno-Optimist Manifesto” (Andreessen, 2023), explained the term used in Silicon Valley to describe the new reality of developers who use AI coding agents - “AI Vampires”:<br>“You stay up all night coding with AI, because you are so productive, you are getting so many things done, that you cannot stop. The opportunity cost of sleep becomes too high, because if you go to bed, you will no longer be there with the 20 AI coding agents, keeping them moving across all the projects you have put them to work on. Compared with six months ago, these developers look terrible: they are sleep-deprived, they have dark circles under their eyes. It is very clear that they are no longer taking care of themselves. And, at the same time, they are absolutely ecstatic, because they can produce 5, 10, 20 times more code than before.” (Rogan, 2026)

Such anecdotes suggest that the benefits developers obtain in terms of execution speed and volume come at a high price (no, not talking about token costs here): an intensification of the cognitive effort required to maintain control over AI coding agents, the possible degradation of the quality of cognitive processes and, ultimately, harm to the individual’s quality of life.<br>In this context, the human–AI partnership raises profound dilemmas concerning the long-term sustainability of the quality of human cognitive processes and psychological balance (Rice, 2026). I’m proposing a critical examination of the claim that the use of AI coding agents automatically leads to cognitive enhancement and, in doing so, I adopt as a foundational premise the desideratum that “users must be empowered, their autonomy respected, and their decision-making capacities strengthened, for technology to be considered a genuine enhancement” (Voinea et al., 2020).<br>In support of the analysis, I’ve also conducted a study to measure developers’ perceptions of the impact that AI coding agents have on them; the raw (anonymised) answers can be accessed publicly at (in case anyone wants to do a further analysis): https://github.com/cborodescu/ai-vampires and a more friendly exploration of the insights can be found here: https://cborodescu.github.io/ai-vampires/.<br>Cognitive Augmentation Using AI Coding Agents<br>When we speak of cognitive enhancement, the primary aim is to increase the speed or capacity of information processing among normal, healthy individuals. This allows the acquisition of information (perception), its selection (attention), representation (understanding), retention (memory), and the use of information to guide behaviour (Heersmink, 2015, citing Bostrom & Sandberg, 2009) to be carried out more quickly and with fewer errors. It may also make possible tasks that would otherwise be beyond the capabilities of the human brain.<br>It cannot be argued that the augmentation of human cognition is a new phenomenon or that we began to consider the issue in such terms with the emergence of...