The science of leadership in action
Leadership is easier to evaluate when you watch it happen.
What does this leader actually do in a leadership situation?
LeaderSim and ExecutiveSim place leaders inside realistic business situations where they have to gather information, interact with AI colleagues, work through competing priorities, make decisions, and respond as circumstances change.
Why simulation?
That idea is not new.
Leadership is behavior in context.
Performance-based assessment creates an opportunity to observe those behaviors directly.
Decades of research on work samples, assessment centers, situational judgment tests, and other simulation-based methods have established an important principle: job-relevant situations can provide meaningful evidence about how people perform.
Research on assessment centers, for example, has demonstrated criterion-related validity for dimensions including communication, influence, organizing and planning, and problem solving. Assessment center performance has also been shown to add information beyond measures such as personality and cognitive ability.
AI agents change what can be done
Highly standardized assessments are easier to administer but often less interactive.
Rich assessment centers can create realistic interactions and business cases — but require trained assessors, role players, extensive scheduling, and considerable administration.
AI agents make it possible to create leadership situations that are both realistic and responsive.
New research on leading AI agents
In a preregistered experiment, researchers Ben Weidmann, Yixian Xu and David Deming had 249 people lead both human teams and teams of GPT-4o-based AI agents. Because participants were repeatedly randomized across human teams, the researchers could estimate each person’s causal contribution to human team performance and compare it with how they performed leading AI teammates.
In both settings, stronger leaders asked more questions and generated more conversational turn-taking, and had to gather information held by different people and synthesize it into decisions.
What we observe
Our simulations are designed to examine areas such as:
01
Diagnosing the Situation
What does the leader notice, investigate, connect, and prioritize?
02
Inquiry & Information Gathering
What questions does the leader ask? What evidence do they seek? How do they test assumptions?
03
Engaging Others
Whom do they involve? When? How effectively do they use the expertise and perspectives around them?
04
Influence
How do they surface disagreement, challenge assumptions, and move others toward a direction?
05
Strategic Judgment
How do they integrate information, evaluate trade-offs, make commitments, and explain their reasoning? What is their final strategic position and presentation to the board?
06
Adaptability & Execution
What happens when the situation changes? Do they reconsider appropriately, remain anchored when they should, and translate decisions into action?
For consequential uses such as hiring, promotion, and succession, Persona 28 believes AI-enabled simulations should ultimately be held to the same fundamental scientific standards expected of other personnel assessments and to additional standards where AI introduces new risks.
Validity is not something an assessment technology possesses forever.
It is evidence that has to be established, examined, and maintained.
Where the evidence stands today
Simulation-based assessment — work samples, assessment centers, situational judgment tests and other job-relevant methods — rests on a long research tradition. Decades of research have established that job-relevant situations can provide meaningful evidence about how people perform.
The Weidmann, Xu and Deming research provides proof-of-concept evidence that leadership behavior with AI agents can carry meaningful information about leadership with people. It supports the premise behind AI agent leadership assessment.
Opportunity before inference.
We distinguish between behavior a leader did not demonstrate and behavior the simulation never gave them a meaningful opportunity to demonstrate.
Evidence tied to behavior.
Every observation in a report is tied to something the leader actually said, quoted exactly as they wrote it.
Simulation testing.
Thousands of measurement runs have been completed across MicroSim and LeaderSim, including full synthetic simulations to evaluate our measurements and test-taking behavior.
Try MicroSim.