After nine years as HashiCorp CEO, Dave McJannet now wants to “unblock” enterprise AI agents
The New Stack Paul Sawers
Dave McJannet’s new startup wants to control enterprise AI agents the way HashiCorp controlled cloud software. The pitch: one layer for identity, tools, models and spending, not a patchwork of fixes.
Based on reporting by The New Stack, Paul Sawers — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Ask a normal enterprise app for a customer address and it follows a script its builders already know. Ask an AI agent, and the path is messy: one tool call, then another, then a model, then maybe a third system before it answers. That unpredictability is fine for getting work done. It is much less charming when the enterprise wants to know which systems were touched, what data was seen and how much was spent along the way.
That problem is what David McJannet is betting Dome Systems can sell. McJannet left HashiCorp after the IBM sale closed in 2025, and he started Dome with Marc Holmes, a former HashiCorp chief marketing officer. The company came out of stealth in April with $14 million in seed funding. McJannet’s pitch is that agents are not just software with a new coat of paint; they are a different architecture, closer to a self-driving car than to a classic business app. They make judgments in the moment, which is exactly why they need a control layer built for that kind of behavior.
Dome’s answer is to pull the whole stack into one place. It registers the agent, manages who can call it, brokers the models it can use, and puts an MCP gateway in front of the tools it reaches, whether that’s Zendesk or something else. Once the pieces are behind a single gateway, the company says it can enforce permissions, inspect responses, apply quotas and keep a common audit trail. McJannet argues that buying those parts separately leaves enterprises stitching together a model router here, a tool gateway there, and still not really knowing what the agent can do.
Now Dome is opening itself to self-service users, with published prices, a free tier and sign-up through a credit card. That is an unusual move for enterprise infrastructure, but McJannet clearly wants practitioners to try it before procurement gets involved. He also thinks the bigger market may be the smaller builder who can ship an agent but then gets stuck when an enterprise customer asks about identity, data access and auditability. In other words, Dome is selling not just control, but permission to get the thing into production.
There’s a familiar pattern here. Cloud needed provisioning, credentials and networking before it became boring enough for the enterprise, and McJannet thinks agents need the same treatment. The pitch is sensible, maybe even overdue. But it also underlines how much of the AI boom is now running headfirst into the boring parts of software, which is usually where the real money starts hiding.
My take — AI-written commentary, not fact-checked reporting
McJannet is reading the room correctly: enterprises don’t want another clever demo, they want a way to say yes without having a nervous breakdown in security review. The market is already full of point products that promise to tame agents one piece at a time, which is a fine way to create a very expensive mess. A single control layer sounds less sexy, and that’s exactly why it has a chance.
Read more about this at: The New Stack