Runtime intelligence startup FriskAI Inc. launched today with $3.6 million in pre-seed funding to give enterprises a record of what their artificial intelligence agents actually do once they go into production.

FriskAI is aiming at a problem that comes with agents. Given different inputs, a different tool set or a shifting objective, the same agent can take a route nobody wrote down in advance. Conventional software does not improvise. The observability and security tooling most companies run was built on that assumption.

The software runs alongside the agent and logs what the agent does with its tools. Each entry holds the arguments the agent sent. The response is in there, and a timing figure. From that record, FriskAI builds a behavioral profile of each agent broken down by task and by tool, then flags when a new version or a new deployment starts acting differently from the last one.

Instrumentation is handled through Python and TypeScript software development kits, with prebuilt adapters for LangChain, Claude Agent SDK and Strands. Anomaly detection runs without preset rules. The platform watches for shifts in the scope of an agent’s activity, the systems it reaches and the volume of calls it makes, and alerts in real time when something looks off.

“AI agents are quickly becoming core operational infrastructure, but most organizations still don’t understand what those agents are actually doing once they’re deployed,” said founder and Chief Executive Neel Palrecha. “Companies can monitor servers, applications and employees, but they lack runtime intelligence for AI agents.”

FriskAI is pitching the platform at healthcare, insurance and financial services buyers, where teams have to be able to explain an agent’s decisions to auditors. Health benefits provider Sana Benefits Inc. is an early user. Staff software engineer Jason Moore said the software gives his team “the operational visibility we’ve been missing,” and has sped up troubleshooting.

Money is moving quickly into the same problem. Israeli startup Zenity Ltd. announced a $125 million round on Aug. 3 to build a security layer for AI agents, and a string of smaller rounds has gone to companies selling visibility into agentic workflows.

MaC Venture Capital Management led FriskAI’s round. Wischoff Ventures joined, along with New Enterprise Associates Inc. partner Rick Yang and a group of angel investors. Marlon Nichols, managing partner at MaC Venture Capital, likened the opportunity to what Amazon Web Services Inc. was to cloud and Cisco Systems Inc. was to the early internet and said the FriskAI team has built “the infrastructure layer that lets companies see what their agents are doing.” The company’s website also lists Detroit Venture Partners among its backers.

The pre-seed money will go toward engineering and go-to-market hiring, customer deployments and further work on the behavioral analysis side of the product. FriskAI, based in Los Angeles, is signing up users through an early access program.

Image: FriskAI