To tackle today's most challenging problems, we must explore effective and scalable ways to train the next generation in leading the design and research of high-impact solutions. In addition to core design-research skills (i.e. prototyping techniques and research methodologies), learners must also develop the metacognitive skills necessary for self-directing complex work (i.e. learning to strategically plan, seek help, and reflect on process). Undergraduate research training provides an opportunity to study how we might better prepare the future workforce. However, scaling effective environments to be robust to a growing number of learners is difficult when mentoring resources are limited. Such practical shortcomings can lead students to undervalue research experiences and even lower their self-efficacy in leading complex work, resulting in fewer students receiving the promised benefits of undergraduate research programs. My focus is to develop "Agile Research Studios (ARS)", a learning ecosystem that trains learners in how to execute their metacognitive practices across available community supports to effectively self-direct research. To enable scalability despite limited mentoring resources, ARS implements a dispersed control approach -- an ecosystem composed of socio-technical "component supports" (i.e. individual studio tools, agile processes, social structures) that distribute learning interactions across a community of practicing researchers. To enable effective research training, ARS fosters self-direction, where students focus on the metacognitive skills required to lead design-research work. To achieve this, ARS implements "subsystems" that interweave component supports via "subroutines" (i.e. sequences of learning interactions) that enable students to practice planning, helpseeking, and reflection skills. My thesis contributes three iterations of Agile Research Studios, each motivated by emergent challenges in learning ecosystem design. In Chapter 2, I introduce "Agile Research Studios" through a pilot study that explores the potential of the ARS "learning ecosystem" to scale effective research training. In Chapter 3, I introduce "Polaris," a case study of how to extend "subsystems" within the ecosystem with new "supports" and "subroutines" to address critical gaps in how the subsystems scaffold expert practice. In Chapter 4, I introduce "Compass," a case study of how to augment the ecosystem with "process scaffolds" that train students to manage their process execution as they practice subroutines within a subsystem, and within the ecosystem more broadly. By designing ecosystems that interweave tools, processes, and social structures together to support metacognitive practice in research communities like ARS, we can achieve increasingly scalable and effective learning environments that train students in the skills they need to drive the high-impact solutions of tomorrow. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://bibliotheek.ehb.be:2222/en-US/products/dissertations/individuals.shtml.]