Two Agents, Not Five

Between 15 June and 17 August, the reporter committed code on 64 consecutive days, working across multiple aspects of the ChargEV FleX product. This period felt rewarding and efficient, but upon reflection, the reporter realized the constraints they faced. They had always been a senior developer, capable of working on any part of the stack, but working on all of them simultaneously was not possible due to limited coding hours.

With the ChargEV FleX project, the reporter took on the full stack responsibilities themselves, removing the constraint of time. This shift changed their role, making them the one orchestrating the work, deciding what to build, specifying it, questioning the approach, and reviewing the outcomes. However, adding more agents brought additional work that the single person had to judge, leading to increased workload.

The reporter pulled data from their own history and found that in July and August, they wrote over a million characters of commit messages, added 38 pages to their Confluence wiki, and had 24% of their commits during late-night hours, a significant increase from the previous ten months. The reporter's work intensified during the summer launch run-up, with 87 commits spread across four repositories in a single day.

While the output felt good, it came at a cost to family time, team calm, and sleep. The reporter noticed that most changes were smaller rather than larger, with a median of 149 lines per commit after the agent usage. Most importantly, only thirteen reviews were left by others, indicating that reviews had shifted from being requests to being results.

This change in review patterns suggests that the reviewer's judgment may have improved, as they could catch issues earlier in the development process.