Kath Korevec, Product Director at Google Labs, presents a transformative vision for coding agents: moving from reactivity to proactivity. She uses the analogy of domestic mental load (having to remind someone to do the dishes) to describe the current state of AI agents. Even when they execute tasks, they force developers to supervise them, breaking their concentration and "flow". Humans are serial processors, not parallel ones; supervising 16 agents running in the background is not a viable developer experience.
The solution lies in proactive agents capable of "doing the dishes without being asked". Korevec introduces Project Jules, an autonomous agent developed by Google Labs, designed to observe, learn, and anticipate. She describes three levels of evolution: 1. The Attentive Assistant ("Sous-chef"): It cleans up behind the developer, adding missing tests or fixing imports without explicit intervention. 2. The Project Manager: It understands the overall context, the frameworks used, and team preferences, acting as a partner who knows the "house". 3. Collective Intelligence (Target level): Jules collaborates with other specialized agents (such as "Stitch" for design and "Insights" for data) to understand the consequences of changes. For example, identifying that a drop in conversion is due to a performance regression and proposing a global fix (code + design).
Technically, this translates into features such as persistent memory (which the user can edit), a "critic" agent that challenges generated code, and the capacity for autonomous verification (e.g. writing and running a Playwright script to prove the code works). Korevec announces updates for December, including proactive detection of "TODOs" in code to turn them into automatically executed tasks. The ultimate goal is to free developers from maintenance tasks (friction) so they can focus on pure creation.