Design Principles for Effective Web-Agent Tools

Identify practical design principles that characterize effective tools for web agents—specifically, programmatic APIs or LLM‑synthesized procedural functions—in order to guide robust tool construction and deployment for realistic web-task automation.

Background

Multiple frameworks for tool-use web agents differ in sources (human vs. LLM‑synthesized), formats (REST APIs, Python functions, JSON action flows), and selector strategies, yet prior studies do not converge on which design choices most reliably yield effective tools.

The lack of comprehensive, controlled comparisons has left unclear which principles—such as complexity, compositionality, or coverage—should guide the creation of effective tools for general web-task performance.

References

As a result, several fundamental questions remain unclear: i) whether tools provide consistent gains for web agents, ii) what practical design principles characterize effective tools, and iii) what side effects tool use may introduce.

— The Tool Illusion: Rethinking Tool Use in Web Agents  (2604.03465 - Lou et al., 3 Apr 2026) in Abstract