The field of CSCW has long contended with the question of how to design technologies to support collaboration between actors with partial knowledge and distributed roles. In this talk, I discuss how CSCW theories about shared artifacts, mutual awareness, and formalization of social requirements can inform the design of systems that coordinate human-agent activity. I present a series of novel systems that incorporate these lessons: Cocoa, a system for flexibly delegating agency between humans and agents around a shared computational notebook-like artifact; Ream, a workspace where humans and agents can be mutually aware of each others' activity; Compass, a system for continuously aligning social media feed algorithms to a user's changing preferences; and Social-RAG, an approach for continuously aligning an agent into a team's social norms. I conclude by arguing for more research that draws upon ideas from institutional design more broadly, including the design of governments, communities, and organizations, to inform future human-AI institutions. I provide one case study involving our work taking inspiration from case law towards building precedent aligned AI systems.