Productivity gurus online these days say AI agents should book your travel, manage calendars, and send emails. Lumbra thinks they can do a whole lot more, and so far, a rapidly growing list of customers in the intelligence community and the Pentagon agree. Investors do too.
This morning, the DC-based AI startup announced a $25M Series A to keep scaling its AI agent operating system and agent harness across the defense and national security space and expand into “regulated commercial sectors,” the company told Tectonic exclusively.
- Point 72 Ventures and Ansa Capital co-led the round, with participation from the company’s OG backer, Costanoa Ventures.
Secret agent: Founded by ex-Army Ranger and CIA counterterrorism officer Aaron Brown (CEO) and fellow Ranger—and MIT-trained computer scientist—Michael Hiebert (CTO), Lumbra’s pitch is that AI agents have the potential to be incredibly capable tools for analysts and decision-makers in national security, but they need the right operating framework, or harness, to make their analysis work auditable, verifiable, and trustworthy.
Lumbra’s platform, called LumbraOS, does that with three core offerings:
- Nebula is the agent orchestrator that records actions, traces claims to sources, and enforces data protection. The human running the agents fact-checks those findings and analysis workflows before they’re relayed to a system of record.
- StarChart is a customer organization’s knowledge graph where agents automatically compound knowledge as they work, with “insights and conclusions recorded alongside sources and sensitivity labels,” according to Lumbra.
- Astrolabe is the agent evaluator that tests them before deployment and tracks scores in production using the customer’s benchmarks.
Put simply, “We’re building a coordination layer for intelligence advantage and decision intelligence—it’s the harness, it’s the control panel, the control layer, it’s the consumption layer,” Brown told Tectonic. “We’re putting the controls in place, the data lineage, the explainability, the coordination layer, the benchmark, and the evaluations, plus the harness.”
- The platform is “model-neutral” (as opposed to model-agnostic), meaning it can run a range of frontier and open-source models depending on the performance level required for the task agents are running.
- “To do some of these tasks, there’s a level of ability that the model has to be above,” Brown said. “It just needs to be a model that is at that level of performance, so we’re model neutral with strong model opinions.”
Rolling out: So far, that’s proven to be a hit with intelligence community customers—where Lumbra signed its first contracts—and increasingly with a self-proclaimed “AI-first” Defense Department. Lumbra is already under contract, with its agents in production on top-secret networks across the Pentagon and IC.
- Lumbra was one of two companies—the other being a small company you might’ve heard of called Palantir—tapped by the Pentagon’s Chief Digital and Artificial Intelligence Office (CDAO) for the “Agent Network” Pace-Setting Project under the Department’s AI Acceleration Strategy.
- Under that contract, Lumbra is responsible for implementing “AI orchestration technology operational on US Government systems” to “compress the time it takes to turn intelligence into informed options for commanders around the world.”
The trust element is a big reason for Lumbra’s hot start with defense customers and on classified networks.
“There are a lot of different ways you can slice up these potential agentic systems. For us, especially when it’s leading to a consequential decision, everything needs the ability to be explained,” Brown said. “You need to be able to allow a human, who’s cognitively limited compared to these machines, to be able to consume that quickly and say, ‘All right, now I trust this output because I read it with my own eyes.’”
Growing up and out: With a fresh $25M in the bank, Lumbra is looking to take that validation earned in the most high-consequence environments into the commercial world, especially in highly regulated sectors. According to Brown, the company is already “working with some private companies in a B2B relationship—some very big ones.”
For a startup that itself is “agent-pilled,” as Brown put it, he’s betting that Lumbra will be “one of the first examples of a 100-200 person company that swings like a 2,000-3,000 person company,” he said. “I don’t think anyone who’s touching our product right now would say anything other than we’re at the front edge of this compared to anyone else in the game.”
