Maya

Maya is a member of the Applied AI team at Anthropic, known for her work on proactive agent workflows using Claude Code. She introduced herself in the context of presenting infrastructure tooling designed to make Claude Code operate as an autonomous, proactive coding agent.

Core Thesis: Proactive over Reactive Agents

Maya's central position is that coding agents should not sit idle waiting for user input. She argues that "coding agents shouldn't wait for you to press enter to get started" and summarizes the stance simply: "proactive agents beat reactive agents." This frames her broader advocacy for infrastructure that enables Claude Code to act as a "proactive teammate" rather than a prompt-response tool.

The Infrastructure Problem with Current Approaches

Maya identifies a significant friction point in building proactive agents today: doing so requires maintaining a full infrastructure stack outside of prompts, including hosting, data persistence, and authentication. She is explicitly critical of Cron-based scheduling as a stand-in solution, sarcastically asking the audience to keep their hands up if they've "enjoyed building all of that infra and maintaining that job" — implying the answer is universally no.

Routines as the Solution

Maya advocates for Routines as the abstraction layer that resolves this infrastructure burden. Her position is that "we built routines so you don't have to focus on maintaining all of this infra", with Routines specifically handling hosting, session state, and connectors to allow Claude Code to operate proactively at scale. She points to adoption signals as validation: weekly PRs for Claude Code have increased 200% since the beginning of the year.

Context Quality and the Generator-Critiquer Pattern

Maya emphasizes that the quality of inputs to a routine fundamentally constrains agent performance: "the context provided to Claude in a routine is the ceiling of how successful Claude will be." To raise that ceiling in automated settings, she advocates for the generator-critiquer pattern — an agent-on-agent review approach where one Claude instance generates output and another critiques it — arguing this improves the quality of Claude's outputs in automated routines.