From HashiCorp to Dome Systems: McJannet Tackles Enterprise AI Agent Governance

Traditional enterprise software follows a predetermined path: verify the user's identity, retrieve data from the appropriate database, deliver the result. The outcome remains consistent across repeated requests. AI agents operate under fundamentally different rules. They might query one system, determine they require additional context, execute multiple tool calls, process information through a language model, and ultimately produce an answer—potentially via a completely different sequence on the next run.

Within an enterprise context, the path an agent takes carries consequences as significant as its final answer. Which systems does it access? What data does it examine? Which actions does it perform? What does it cost? This distinction between deterministic software and applications that make probabilistic choices in real time forms the foundation for a startup launched by someone with substantial experience imposing structure on emerging infrastructure technologies.

AI agents are hard to govern

Dome Systems emerged at the beginning of the year under the leadership of David McJannet, who spent nearly a decade as chief executive of Terraform creator HashiCorp, overseeing its landmark 2021 initial public offering and the subsequent $6.4 billion acquisition by IBM in 2025. Marc Holmes, formerly HashiCorp's chief marketing officer for more than six years, serves as co-founder alongside McJannet.

In conversation with The New Stack, McJannet articulates his company's framework for managing AI agents, contending that enterprises face the same fundamental challenge they encountered with cloud infrastructure: rapid adoption precedes the difficult work of implementing appropriate controls across security, operations, and financial domains.

The underlying issue, according to McJannet, stems from how agents are architected—a departure from legacy enterprise applications that organizations spent years learning to manage. "It's actually a very different architecture, and that is what unlocks the power of these new [agentic] applications," he explains.

McJannet draws a parallel to autonomous vehicles: a model ingests real-time data and communicates with vehicle systems as circumstances shift, since no developer can feasibly pre-program every scenario a vehicle might encounter. "It's making judgments along the way, as opposed to trying to look up the historical maps of the world and make a real-time decision," McJannet continues.

Enterprise agents function similarly: invoke one tool, evaluate the outcome, determine whether another is needed, and continue until completion. This adaptability enables agents to handle tasks that would be impractical to script exhaustively beforehand—yet it simultaneously complicates enterprise governance of their behavior. This challenge is precisely what McJannet is working to address through Dome.

Table stakes for the agent era

Dome emerged from stealth in April with $14 million in seed funding. McJannet had left HashiCorp in August following the IBM transition's completion.

According to McJannet, an agent comprises three components: code, a model, and the backend systems or tools it engages with. Consolidating these elements within a unified platform represents, in his view, the minimum requirement for enforcing meaningful restrictions on agent behavior.

"If you don't have an integrated platform, you can't enforce controls across everything that the agent is doing," McJannet states.

Dome's architecture centers on these three components. An agent registry maintains records of agents themselves; an MCP gateway manages which tools agents can invoke; and a model broker/router determines which models agents can access and how requests are distributed.

The workflow begins by registering the agent and assigning it an identity, specifying who can invoke it, and linking the backend tools it can reach—such as Zendesk. Subsequently, Dome connects a model provider, organizes available models into a pool with routing and failover mechanisms, then unifies the agent, its tools, and its models behind a single gateway. This gateway serves as the enforcement point for policies governing agent permissions.

Once integrated, teams can establish permissions for individual calls, employ guards to examine responses, implement quotas to manage expenses, and maintain a unified audit log documenting the agent's complete activity.

Currently, enterprises typically assemble these capabilities independently, according to McJannet. A dedicated model broker might manage spending while a separate tool gateway addresses security and operational requirements. Some organizations construct their own agent registry to integrate these systems.

Purchasing these capabilities separately introduces additional integration challenges. A model router might govern one aspect of an agent's operation and a tool gateway another, while the agent itself moves between them.

"If you just provide the tool gateway or just the model router, it doesn't allow you to have this kind of system of control," he says.

Dome's newest initiative addresses this fragmentation. Following its initial months in early access, the company is now extending platform access to self-service users for the first time, enabling teams to register with minimal friction—essentially just a credit card—and circumvent the typically lengthy enterprise procurement cycle.

Dome goes self-serve

Adopting a self-service model represents a relatively unconventional strategy for enterprise infrastructure software of this type. Dome is making its pricing transparent, providing a complimentary tier, and permitting practitioners to begin without navigating a sales process, while preserving the traditional enterprise sales channel for larger organizations.

This approach reflects McJannet's assumptions about who will adopt the product. Rather than restricting access to procurement-stage buyers, self-serve enables individual practitioners to independently discover, experiment with, and deploy the platform. "We want to make the barrier as low as possible to have people come on board," McJannet says, noting that Dome had already received self-service registrations before the official launch.

The pricing strategy also embodies McJannet's perspective on where market value will ultimately concentrate. He views model routing and tool connectivity as foundational capabilities, with the genuinely valuable component being the controls spanning the entire agent—permissions, data redaction, and spending caps.

While platform engineering teams within large enterprises—collaborating with operations and security—represent Dome's primary target, self-serve also opens possibilities for a different user category: the small organization, potentially just one or two individuals, constructing an agent for enterprise sale. This scenario aligns with the AI industry's vision of the solo-founder unicorn.

McJannet notes that developers can progress substantially in building the application itself, only to encounter obstacles when a prospective enterprise customer initiates its security and operations assessment. How is identity managed? Who accesses the data the agent retrieves? What occurs when it invokes other agents? Can its actions be audited retroactively?

Some builders have inquired whether they can "certify" their agents on Dome because "my agent won't get deployed until I can satisfy these infrastructure elements," McJannet recounts. He emphasizes that Dome does not presently operate such a certification program, though the company could potentially pursue this direction.

"If you register that agent on Dome, all the infrastructure elements are taken care of," McJannet says.

'Unblocking AI agents': Lessons from the cloud era

The tension between developers wanting to launch and enterprise teams concerned about downstream implications mirrors what McJannet witnessed during his HashiCorp years—a parallel he emphasizes.

Throughout his tenure at HashiCorp, the company increasingly positioned itself as a solution for large organizations standardizing cloud infrastructure provisioning, security, and connectivity. This included the 2020 introduction of HashiCorp Cloud Platform (HCP), which delivered its infrastructure tools as managed cloud services.

McJannet's account of early cloud adoption describes developers using personal credit cards to deploy directly to Amazon because cloud infrastructure made previously impractical applications feasible. These applications proved compelling enough that enterprises embraced cloud despite pushback from operations and security teams. What followed was a second phase: organizations required shared services for provisioning, credential management, networking, and other controls before cloud could become standard across the enterprise.

Platform engineering teams emerged as the mediators between these competing demands: empowering developers to build while granting security, operations, and finance sufficient oversight to permit applications into production. McJannet contends agents are now generating identical friction.

"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," he says. "And so, inevitably, it has to go that same direction where the platform engineering team has to figure out [a way] to get to say 'yes'."

Dome's wager is that enterprises will ultimately favor a single system encompassing the entire agent over a collection of gateways, routers, and security solutions. In McJannet's framing, this unified control layer permits enterprises to restrict agent behavior while preserving autonomous operation.

"You have to have this control layer that provides this corridor where we can constrain the behavior of that new type of application architecture," he says. "Because without that, you cannot unblock the deployment of AI applications."

Dome still has substantial ground to cover. The company is not disclosing customers at present; McJannet indicates that enterprises working with Dome have declined public identification, though he reports the company has engaged dozens over the past eight months.

McJannet's conviction is that the cloud era demonstrated new applications achieve widespread adoption only once enterprises possess the controls to permit them. Dome represents his attempt to solve this problem for agents.

"I think that's the part that we're trying to answer — how do we unblock agents at scale?"

Source: The New Stack