Agentic AI
HashiCorp founder launches Dome Systems to govern enterprise AI agents at scale
David McJannet, who led HashiCorp through its IPO and IBM acquisition, is now building infrastructure to help enterprises control AI agents the way platform teams once learned to govern cloud infrastructure.
The New Stack
After steering HashiCorp for nine years and overseeing its 2021 public offering and eventual 2025 sale to IBM for $6.4 billion, David McJannet has turned his attention to a new problem: how enterprises should govern AI agents. His new company, Dome Systems, launched from stealth in April with $14 million in seed funding and is now opening its platform to self-service users—a departure from typical enterprise infrastructure sales cycles.
The governance gap#
Traditional enterprise applications follow predictable paths. A user authenticates, a query runs, and the system returns a predetermined answer. AI agents operate differently: they assess initial results, decide what additional context they need, make multiple tool calls, feed information through language models, and potentially take a different route if run again. What happens along that path—which systems the agent accesses, what data it sees, which actions it takes, and how much it costs—matters as much as the final result to risk-conscious organizations.
McJannet argues that enterprises face the same adoption-versus-control tension they once encountered with cloud infrastructure. Developers build compelling applications powered by agents, but operations and security teams balk at deploying systems whose behavior is difficult to predict or constrain. "You've got this queue of cool apps that developers build that the ops and security teams are just not comfortable letting flourish in their environments," McJannet says.
A platform built for control#
Dome's architecture reflects McJannet's thesis: agents are built from three components—code, a language model, and backend tools or integrations. The company created an integrated platform to govern all three together: an agent registry tracks agents and verifies callers; an MCP gateway controls which tools agents can access; and a model broker/router determines which models agents use and how requests route between them.
By bringing these elements behind a single gateway, Dome can enforce unified policies—permissions on each call, response inspection guards, spending quotas, and audit trails. McJannet believes this integrated approach is essential: "If you don't have an integrated platform, you can't enforce controls across everything that the agent is doing."
Today, large enterprises often buy these capabilities piecemeal—a model router from one vendor, a tool gateway from another, and a custom agent registry built internally—creating integration headaches and fragmented visibility.
