GitHub records expose the agents' trap and show that code-writing licenses double technology managers' burnout
The promised benefit of AI tools was clear in its simplicity: eliminate boilerplate code, delegate routine tasks, and get home earlier. Yet tracking field engineering activity tells a completely different story from what platforms promote. When we examined the contribution records of Orestis Yuanou, Vice President of Technology at Bold.org, via GitHub over six years from January 2019 to late 2026, the numbers revealed a unexpected reality: his current work pace exceeds any prior period, and his mental load is heavier than when he was a full-time programmer at the start of his career.
The recorded data show a structural shift in the production curve. While serving as a principal engineer from August 2022 to July 2025, his monthly contribution rate settled at 114 contributions, marking his lowest annual total in 2023 with 1,189 contributions, after his role shifted toward planning and building technical architecture rather than writing code directly. This quiet pattern continued after his promotion to Vice President of Technology in August 2025, with an average of 111 contributions per month and a low of 50 contributions in December. However, the introduction of smart agents into his daily work at the start of 2026 reversed the trend; by 27 October 2102 he had logged 2,102 contributions, surpassing the total contributions of any full year between 2021 and 2025, and in August 2026 alone he reached 427 contributions, the highest monthly figure in his career compared with his previous peak of 319 contributions in April 2020.
The difficulty arises here from the lack of natural friction that used to regulate the flow of ideas, not from the complexity of any single programming task.When code writing became almost free thanks to the models, Yuanou found himself running several agents simultaneously: one that implements a code change, another that inspects an issue in the runtime environment, a third that reviews the changes, and a fourth that gathers research for a later project. Although none of these agents requires continuous intervention, each periodically returns with a request for plan approval, a review of a line-difference, or an explanation of a test result. This intermittent feedback forces continuous mental navigation between different code repositories and multiple technical contexts, draining cognitive energy quickly; the models’ generation capacity has grown far faster than the human mind’s ability to review and contemplate code consequences.
This flow generates repeated cycles of excitement followed by sharp depletion. The records show three peaks and troughs in 2026, starting in January and February with 277 and 233 contributions as agents were trialed and tasks delegated, then gradually calming to a low of 77 contributions in June as mental fatigue accumulated, before exploding in August with 427 contributions after the summer break, with daily contributions ranging between 14 and 43. The pattern repeats precisely: a newer model or workflow opens new possibilities, and because execution is cheap the engineer launches dozens of experiments at once, causing review requests to pile up and forcing a mandatory slowdown and a sense of weariness with the AI tools themselves.
This shift carries a direct practical implication for technology leaders and development teams in the Gulf, Egypt and the Levant, where companies, banks and startups race to mandate the use of coding agents assuming they will cut work hours or shrink engineering teams. Field experience shows that lowering code-generation cost moves the bottleneck straight to the architectural review and human verification stage without eliminating it. In regional tech hubs in Riyadh, Dubai, Cairo and Amman, this pattern threatens to turn senior engineers and technology managers from strategic leaders into exhausted draft reviewers facing an endless stream of agent outputs.
The most valuable skill in modern programming environments is no longer limited to running more agents, but to having the courage to refrain from building what is unnecessary.If you lead a technical team in the region, protecting your focus and your engineers’ energy requires consciously limiting what is allowed to be built, rather than measuring achievement by the number of code contributions that flood your servers without adding real value to the work.