Glean has developed a new agent harness designed to improve the efficiency and capabilities of enterprise AI agents. This harness utilizes programmatic tool-calling, where the AI model writes Python code to interact with Glean's tools and data, rather than issuing direct tool calls. This approach reduces token usage by 24% compared to previous methods by minimizing serialization and deserialization overhead. The system also incorporates a sandbox filesystem to manage intermediate data and tool outputs, with truncated previews shown to the model to maintain focus and efficiency during complex, long-running tasks. AI
IMPACT This development could lead to more token-efficient and capable enterprise AI agents by optimizing how models interact with tools and data.
RANK_REASON The item describes a new technical approach to building AI agent harnesses, focusing on implementation details and efficiency gains, rather than a novel model release or a significant industry-wide product launch.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
