Manager of Agents

The landscape of day-to-day software engineering has fundamentally shifted over the last few months. With the new suite of AI models now available to use in Jetski (Google's internal Antigravity platform)—specifically Gemini 4 Argon, Claude Fable 5.1, and Claude Opus 5.5—my daily workflow has transformed completely.

Instead of sequentially writing code, testing, and debugging, I now operate with several agents running simultaneously in the background. While I focus on higher-level architecture or review, these agents handle the tedious testing, meticulous planning, and boilerplate generation.

What is most striking is the raw efficiency. These agents are usually much faster than a human counterpart, and with the recent model upgrades, they are significantly more accurate than they were even a year ago. They rarely miss a semicolon, they write comprehensive tests, and they can refactor large swaths of a codebase in seconds.

However, human scrutiny remains absolutely crucial. The agents are incredibly powerful, but they lack the overarching business context and the subtle intuition of knowing why a particular design pattern fits the current constraint better than another.

At this point, my job feels less like being a traditional Individual Contributor (IC) typing out every line of code, and much more like being a manager. I spend my days delegating tasks, reviewing pull requests from my AI counterparts, providing directional feedback, and orchestrating their parallel efforts to build better systems faster. It's a new era of engineering, and the leverage it provides is unprecedented.