Traditional enterprise software follows a predetermined path: authenticate, query, return. But AI agents operate differently. They might consult multiple systems, make dozens of tool calls, route information through language models, and take a completely different path on the next identical request. In an enterprise context, that unpredictability matters—which systems get accessed, what data flows through, which actions execute, and how much it costs all become critical concerns.
This fundamental architectural difference between scripted applications and probabilistic agents sits at the center of Dome Systems, a startup founded by David McJannet, who spent nearly a decade as CEO of Terraform-creator HashiCorp before the company's 2021 IPO and subsequent $6.4 billion acquisition by IBM in 2025. Marc Holmes, who served as HashiCorp's chief marketing officer for over six years, co-founded the company alongside McJannet.
The governance challenge
McJannet contends that enterprises face the same predicament with AI agents that they encountered during cloud infrastructure adoption: rapid developer-driven adoption followed by the hard work of implementing security, operational, and financial controls. The core issue stems from how agents are fundamentally constructed.
It's actually a very different architecture, and that is what unlocks the power of these new [agentic] applications.
David McJannet
McJannet illustrates this with autonomous vehicles: a model ingests live data and interacts with vehicle systems as circumstances evolve, since no developer can preprogram every possible road scenario. Enterprise agents operate similarly—they call one tool, evaluate results, determine if additional information is needed, and continue iterating until completion. This flexibility enables agents to handle work that would be impractical to script exhaustively, yet it simultaneously complicates governance.
Building the control layer
Dome launched from stealth in April with $14 million in seed funding, following McJannet's departure from HashiCorp in August after the IBM transition wrapped up. The company's foundation rests on a straightforward premise: agents combine three components—code, a model, and backend systems or tools—and integrating these under a unified platform is essential for enforcing meaningful constraints.
If you don't have an integrated platform, you can't enforce controls across everything that the agent is doing.
David McJannet
Dome's architecture centers on three elements: an agent registry tracking agents and their identities, an MCP gateway controlling accessible tools, and a model broker/router governing model selection and request routing. The workflow begins by registering an agent, establishing who can invoke it, and connecting backend tools like Zendesk. Next, a model provider is integrated, available models are grouped with routing and failover logic, and the agent, tools, and models are consolidated behind a single gateway. This gateway becomes the enforcement point for policies dictating agent behavior.
Through this unified gateway, teams can establish permissions per call, deploy guards to inspect responses, implement spending quotas, and maintain a unified audit trail. Currently, enterprises typically assemble these capabilities separately—a standalone model broker for cost control, a separate tool gateway for security, and custom agent registries to tie systems together. This fragmented approach creates integration challenges where different components govern different aspects of agent activity without coordinated oversight.
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.
David McJannet
The self-serve shift
After initial months in early access, Dome is now offering self-service access, allowing teams to register with minimal friction—essentially a credit card and signup form—while maintaining traditional enterprise sales channels for larger deployments. This approach diverges from typical enterprise infrastructure products.
McJannet's reasoning reflects his expectations about who will use the platform. Rather than restricting access to procurement-ready buyers, self-serve enables individual practitioners to discover, experiment with, and deploy the platform independently. McJannet notes that Dome had already received self-service registrations before the official launch.
We want to make the barrier as low as possible to have people come on board.
David McJannet
Pricing reflects McJannet's perspective on where market value concentrates. He views model routing and tool connectivity as baseline capabilities, with the premium value residing in controls spanning the entire agent—permissions, data redaction, and spending quotas. While platform engineering teams inside large enterprises remain the primary target, self-serve also opens possibilities for smaller operators: a solo developer or two-person team building an agent for enterprise sale. This scenario aligns with the AI industry's promise of individual-driven innovation.
McJannet observes that developers often progress far building applications, only to encounter obstacles when prospective enterprise customers conduct security and operations reviews. Questions arise: how is identity enforced? who accesses agent-retrieved data? what occurs when agents call other agents? can activity be audited retroactively? Some builders have inquired whether they could "certify" agents on Dome, as their agent "won't get deployed until I can satisfy these infrastructure elements." While McJannet clarifies that Dome doesn't currently operate a certification program, he acknowledges this as a potential direction.
If you register that agent on Dome, all the infrastructure elements are taken care of.
David McJannet
Parallels with the cloud era
McJannet identifies the strongest parallel to his HashiCorp experience in the tension between developers eager to deploy and enterprise teams concerned about post-deployment implications. During his tenure, HashiCorp positioned itself as the standardization layer for how large organizations provisioned, secured, and connected cloud infrastructure, including the 2020 launch of HashiCorp Cloud Platform (HCP) offering managed cloud services.
McJannet's account of early cloud adoption describes developers using personal credit cards to deploy directly to Amazon, enabled by cloud infrastructure's practicality. Enterprises adopted cloud despite operations and security resistance because the applications proved compelling. A second phase followed: organizations required common services for provisioning, credentials, networking, and controls before cloud became routine. Platform engineering teams emerged as the mediators, balancing developer velocity against security, operations, and finance requirements. McJannet sees agents creating identical tension today.
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.
David McJannet
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'.
David McJannet
Dome's thesis posits that enterprises will eventually prefer a single system spanning entire agents over a patchwork of gateways, routers, and security products. In McJannet's framing, this unified control layer enables enterprises to constrain 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. Because without that, you cannot unblock the deployment of AI applications.
David McJannet
Dome remains in early stages with much to demonstrate. The company is not naming customers; McJannet indicates that enterprises it has engaged are unwilling to be publicly identified, though Dome has conducted discussions with dozens over the past eight months.
That's the part that we're trying to answer — how do we unblock agents at scale?
David McJannet
McJannet's conviction, rooted in cloud infrastructure experience, holds that new applications become widespread only once enterprises possess controls permitting their deployment. Dome represents his attempt to solve that problem for AI agents.
Source: The New Stack