One valuable goal of instructional technologies in K-12 education is to prepare students for future learning. Two classroom studies examined whether Teachable Agents (TA) achieves this goal. TA is an instructional technology that draws on the social metaphor of teaching a computer agent to help students learn. Students teach their agent by creating concept maps. Artificial intelligence enables TA to use the concept maps to answer questions, thereby providing interactivity, a model of thinking, and feedback. Elementary schoolchildren learning science with TA exhibited "added-value" learning that did not adversely affect the "basic-value" they gained from their regular curriculum, despite trade-offs in instructional time. Moreover, TA prepared students to learn new science content from their regular lessons, even when they were no longer using the software. (Contains 8 figures.)