The Fraud Boxer Podcast What are AI Agents actually doing on your site?
Aug 18, 2025
Nate Kharrl, Co-founder and CEO of Spec, and Shawn Colpitts, a fraud and data specialist at Spec, dive deep into the world of AI agents and bots. They clarify the differences between helpful agents and harmful bots, highlighting real-world examples like an AI agent's clever shopping habits. Discover why not all bots are bad for business and how blocking them indiscriminately could cost you sales. The duo also shares tips on tracking agent behavior and suggests strategies for distinguishing between legitimate traffic and malicious activity.
AI Snips
Chapters
Transcript
Episode notes
Agents Behave Human But Scale Like Bots
- Agents emulate human sessions with clicks, swipes, logins and persistent profiles yet run continuously and react to defenses instantly.
- Their speed and technical sophistication make them distinguishable from naive bots but hard to catch with traditional signals.
Three Distinct Agent Interaction Types
- Agents touch sites in three ways: scraping, RAG (retrieval-augmented) single requests, and full virtual web sessions.
- Each interaction type appears differently in logs and requires cross-team correlation to detect.
Good Agents Drive Real Revenue
- Many merchants already get a meaningful share of legitimate purchases from agents, sometimes 25% of good transactions.
- Blocking all bot traffic risks cutting off real revenue and customer journeys.