There is a conversation that plays out in almost every enterprise AI project, usually a few weeks in.

The AI agent has been built. The use case is clear. The team is ready to deploy. And then someone asks: “Does the agent know how our CRM is actually configured?” Or: “Does it understand that we have a custom integration between our ERP and the warehouse system that was built six years ago by a contractor who no longer works here?”

The answer is almost always no. And that is where things slow down.

On March 31, 2026, Whirl AI emerged from stealth with $8.9 million in seed funding led by ICONIQ, with backing from enterprise veterans who founded or led Okta, Splunk, and VMware. The company was founded by Sunny Bedi, who spent two decades as CIO at NVIDIA and Snowflake. The problem it is built to solve is one that every organization scaling AI deployments eventually runs into: AI agents are only as useful as the context they have about how your systems actually work.

The institutional knowledge problem

Every business that has operated for more than a few years has accumulated layers of customization, workaround, and accumulated decision-making inside its technology stack. A Salesforce instance with five years of custom fields, automations, and integration logic. An ERP that has been connected to a dozen other systems by developers who may have since left the company. A data warehouse built incrementally over time, with naming conventions that made sense in 2019 and were never fully updated.

This knowledge lives nowhere a machine can find it. It is in the heads of the people who built the systems. In scattered documentation. In tribal knowledge passed down through onboarding conversations. In Confluence pages that nobody updates and everyone forgets exist.

When a human engineer takes on a new project, they spend weeks reading the codebase, asking questions, and building up a mental model of how everything fits together. AI agents do not have weeks. They need that context upfront, and most organizations have no good way to provide it.

The result is AI projects that move slower than expected, agents that make mistakes because they did not know about a specific business rule, and IT teams that spend as much time briefing and correcting AI tools as they would have spent doing the work manually.

What Whirl AI is building

Whirl’s platform is designed to help enterprise IT teams research, design, develop, and test changes to core business applications and integrations, with the institutional context built in. The idea is to compress what normally takes weeks of investigation and development work into hours, by giving AI agents access to the actual knowledge of how a company’s systems operate.

The founding team brings direct experience with this problem at scale. Bedi spent years at NVIDIA and Snowflake managing the IT complexity that comes with rapid growth and constant system changes. The ICONIQ-led funding round also includes backing from individuals who have lived through building and scaling enterprise software platforms.

What This Means for Business

This story matters because it identifies a problem that most businesses running AI projects have encountered but not named clearly: the gap between what an AI agent is technically capable of and what it can actually do inside a specific organization’s environment.

Generic AI models are impressive. But the reason AI projects often stall is not capability limitations. It is context limitations. The agent does not know your business well enough to be trusted with consequential decisions inside your systems.

This is one reason why the teams that do best with AI deployments tend to invest heavily in documentation and context-building before deployment. Writing down how systems actually work, capturing the institutional knowledge that exists only in people’s heads, building structured briefings for agents.

Whirl AI is betting that this problem is large enough to warrant dedicated infrastructure. Given how many enterprise AI projects are quietly stalling for exactly this reason, they are probably right.

For businesses deploying AI agents: the technical build is often the easy part. The harder work is documenting what your systems actually do, how your processes actually run, and what your agents need to know to operate safely inside your environment. That groundwork is also what makes every future AI project cheaper and faster to deploy.

Want the practical version of this? The free Working With Claude field guide covers the full Claude ecosystem, Claude Code, and how to roll it out across a real business. Download it here.