Capabilities
The institute does research and development on AI systems that adapt to the people and settings they work with, and on methods for testing those systems.
Core competencies
- Adaptive and agentic AI systems
- Design and prototyping of AI agents that plan and carry out tasks over many steps under human direction.
- Evaluation and assurance
- Studies of what an AI system does in practice, with each finding tied to the record of the system's actions.
- AI cybersecurity
- Research on AI agents that find and fix software vulnerabilities, and on attacks against models and agents.
- AI for learning
- Design of tutoring and teaching tools, and measurement of what students learn with them.
- Decision support for defense
- Research on AI that supports analysts and operators when information is incomplete or deliberately misleading.
- Visual communication
- Systems that turn source material into diagrams, animation and narrated video, and tests of how well people understand the result.
Kinds of work
- Applied research
- Studies that answer a stated question and report the methods and data behind the answer.
- Prototype development
- Working software that shows whether an approach holds up outside a paper.
- Evaluation studies
- Independent tests of an AI system or agent against tasks agreed with the sponsor.
- Research partnerships
- Joint work with universities and laboratories, as a lead or as a partner.
Open-source tools
Written by the institute's founder and used in its work.
- Engineering Behavior Observatory
- Records how AI coding agents do their work across eight agent harnesses and ties each behavioral finding to the native record behind it.
- OpenSymphony
- Coordinates software work done by AI agents and people together, with human review steps.
Start a conversation
Write to the institute about a program, a solicitation or a research partnership.
Write to the institute