TIDE: Proactive Multi-Problem Discovery via Template-Guided Iteration

Researchers introduce TIDE, a template-guided iterative framework that enables AI agents to proactively discover multiple hidden problems within user contexts rather than responding only to explicit requests. The system uses iterative discovery and thought templates to uncover coexisting issues with supporting evidence, demonstrating significant improvements over single-shot approaches in personal workspace and software repository settings.

TIDE addresses a fundamental limitation in current AI agent deployment: reactive problem-solving. While existing agents excel at responding to explicit user queries, they miss opportunities to identify latent issues that users haven't consciously articulated. This research tackles the realistic gap between stated and actual user needs, a critical consideration for enterprise and developer-facing AI applications.

The framework's dual mechanisms represent meaningful innovation in agentic AI design. Iterative discovery prevents the model from anchoring on obvious cases by conditioning each round on previously identified problems, progressively expanding coverage. Thought templates, derived from historical problem-solving patterns, provide structured schemas that ground predictions in recognizable problem classes rather than producing generic recommendations. This design mirrors how human experts systematically work through complex domains.

For AI developers and product teams, TIDE's validation across personal workspaces and software repositories demonstrates broad applicability. The consistent improvements over parallel multi-agent baselines suggest efficiency gains alongside better coverage—critical metrics for resource-constrained deployments. The framework's reliance on reusable templates also hints at scalability potential as organizations accumulate domain-specific problem patterns.

The research signals a shift toward proactive, context-aware AI systems rather than purely reactive ones. As agents become more integral to professional workflows, the ability to surface hidden problems autonomously becomes a competitive differentiator. Implementation challenges remain around template quality and computational overhead from iterative rounds, requiring careful empirical evaluation before widespread adoption.

• →TIDE enables AI agents to discover multiple hidden problems through iterative rounds rather than single-pass predictions, improving problem identification across diverse contexts
• →Thought templates derived from previous cases anchor predictions in recognizable problem classes, reducing generic recommendations and improving actionability
• →Framework validation shows substantial gains over single-shot and parallel multi-agent baselines on task coverage and resolution metrics
• →Proactive problem discovery has immediate applications for personal workspace assistants and software repository analysis tools
• →Template-based approach creates reusable domain knowledge, suggesting scalability potential as organizations accumulate problem-solving patterns