PolicyGuide guides LLM agents through compliant workflows
The system compiles domain policies into workflow graphs to improve agent policy compliance.
Why it matters
The work targets a key deployment gap for AI agents: action-level safeguards may block risky steps but do not ensure agents follow required multi-step procedures. Workflow-based policy guidance could make customer-service agents more reliable across regulated or policy-heavy tasks.
The key points
- 1.PolicyGuide converts domain policies into workflow graphs.
- 2.Mean Pass^4 rose from 0.42 to 0.62 on τ^2-bench.
- 3.Workflows transferred to Claude Sonnet 4.6 and Gemini 2.5 Pro agents.
A paper featured by HF Daily Papers introduces PolicyGuide, a system for customer-service LLM agents that compiles organizational policies into workflow graphs. At user-turn boundaries, a proactive verifier uses persisted graph state to reconcile open requests and return step-specific remediation along a policy-compliant path. Across τ^2-bench airline, retail, and telecom domains with a GPT-5.4 agent and verifier, PolicyGuide raised mean Pass^4 from 0.42 to 0.62, with the largest gain in telecom from 0.19 to 0.61. The same workflows transferred to Claude Sonnet 4.6 and Gemini 2.5 Pro agents.
⚡ Try this today
If you run customer-service agents, map policies into explicit workflow states and verify progress at user-turn boundaries.
Sources & original reporting
This brief summarizes and links to reporting from the publishers below.
Enjoyed this brief? Get the next one in your inbox.
More in Research
Study finds AI signs in a third of post-ChatGPT web pages
Research cited by TechCrunch says AI models are shaping much of the new web.
OpenAI math solutions stir debate among mathematicians
The Verge reports that OpenAI's work on longstanding math problems has raised questions for the field.
Preprint proposes spectral neuron for interpretable ML
The model uses learned symmetric matrices and eigenvalues to add nonlinear expressiveness while retaining transparency.