It had been shown previously that so-called agent-based traffic micro-simulations could be used for dynamic traffic assignment, that is, iterative route adjustment, until either a Nash equilibrium or some steady state distribution between alternatives had been found. It was also shown that the same approach could be extended to (departure) time adjustment; that is, time adjustment and route adjustment could exist in the same iterative approach. In this paper it is shown that the approach can be extended to mode choice by forcing every synthetic traveler to consider every available mode. The implementation is verified with a test case for which an approximate solution can be analytically derived and for which it is shown that simulation and theory are consistent. It is then applied to a large-scale real-world example, the metropolitan Zurich, Switzerland, area, with about 1 million inhabitants. For this example, it is shown that the adaptive scheme, albeit seemingly simple, can outperform a more traditional approach that first computes mode choice on the basis of aggregate data and then runs the assignment for car traffic only. Sensitivity tests show that the model reacts in meaningful ways, in particular concerning the interaction between the time structure of activities and mode choice.