About Andy Agent Lab
Andy Agent Lab researches AI behavior, evaluation, security, and the technical conditions that shape how models and agentic systems perform.
Andy Agent Lab is an AI research and applied systems company founded and operated by Tamar Peretz.
Its work investigates how AI systems reproduce patterns associated with human cognition, emotion, and social behavior, and how training processes, model mechanisms, and system architectures give rise to those patterns.
The research begins with the human capability a system appears to approximate. It then examines what the observed AI behavior demonstrates, what it does not demonstrate, and which technical conditions shaped it. This avoids treating a convincing response as sufficient evidence of understanding, reasoning, or reliable task completion.
By connecting observable behavior with its technical basis, Andy Agent Lab identifies human-derived biases, capability gaps, and failure modes that can affect the reliability and security of models and agents.
Research and applied systems
The research is tested in proof-of-concept systems, where questions about model behavior can be examined in the context of an actual AI system rather than as isolated outputs.
This work informs the design, evaluation, and security of models and agents. It includes questions such as:
- how model behavior should be evaluated beyond fluent or persuasive output
- how prompts, context, sources, tools, and system architecture shape an AI system’s behavior
- where the gap lies between apparent capability and verifiable performance
- how human-derived reasoning patterns, bias, and social behavior can create reliability or security risks
Security analysis is mapped, where relevant, to OWASP risk frameworks for LLM and agentic applications. The focus is on identifying where a system’s behavior, authority, or information boundaries can become unclear, misleading, or unsafe—and on defining controls that can be inspected and tested.
Professional training and organizational work
The lab translates research findings into professional training, lectures, and selected work with organizations.
This work supports more controlled use of AI in professional settings: defining the task clearly, setting appropriate evidence and source boundaries, evaluating outputs, and preserving human oversight over consequential decisions and workflows.
The objective is not simply to make AI use more efficient. It is to make the reasoning, evidence, limitations, and system boundaries behind AI-assisted work visible enough to evaluate.
Public knowledge platform
This site is Andy Agent Lab’s free public knowledge platform. It shares part of the lab’s research and applied work with people who use, build, or evaluate AI systems.
- Articles examine AI behavior, model training and evaluation, prompt engineering, agent architecture, and AI security.
- Guides provide practical procedures for structured AI-assisted work.
- Prompts offer reusable task specifications and control patterns.
- Policies define requirements for evidence, sources, verification, review, and output.
- Reference material provides technical models, diagrams, terminology, and platform guidance.
The platform is designed to help readers understand AI critically, evaluate its claims, and adopt it with appropriate evidence, oversight, and control.
Founder
Tamar Peretz is the founder of Andy Agent Lab and an AI researcher. Her work examines the relationship between human capabilities and AI behavior, with particular attention to the cognitive, emotional, and social patterns that AI systems can reproduce.
Her research includes Theory of Mind and the ways these patterns can affect reliability and security. She develops proof-of-concept systems and translates the resulting insights into evaluation methods, professional training, lectures, and selected organizational work.