LS-Labs

Research

Research

ResearchOriginal research

When AI becomes the user: understanding Agent Experience through affordances

An AI books a hotel but cannot confirm it worked. A 32-run booking experiment shows why agents need affordances, authorization and verifiable results.

  • Affordance
  • Agent Experience
  • AI agents
  • Web accessibility
李昇
Read research

ResearchOriginal research

Why does AI still get it wrong after picking the right tool? The TCC tool-contract framework

Asked to void an invoice, an AI prepares to call a payment tool. The cause may be how the tool describes its capabilities, not the AI itself.

  • #TCC
  • #TSO
  • #ToolMetadata
  • #AIAgent
李昇
Read research

ResearchOriginal research

Can editing only the Tool Description change an AI’s tool choice?

The tool is listed and the API works, yet the AI never calls your service. The problem may lie not in the tool but in how the model understands it.

  • TSO
  • Tool selection
  • Tool retrieval
  • MCP
李昇
Read research

ResearchResearch review

When AI search cites AI: synthetic sources and the knowledge feedback loop

A research review of Synthetic Sources?: its 712-query audit, the detector-based 16% finding and its limits, plus circular reporting and citation trust.

  • Synthetic sources
  • Research review
  • Citation reliability
李昇
Read research

ResearchOriginal research

How AI search finds local businesses: websites, third-party sources and MCAC

A six-restaurant Taipei case study with 21 Sonar observations, separating mentions, citations, reference agreement and actionable information.

  • Original research
  • MCAC
  • Local business
  • AI search
李昇
Read research