Enterprises are currently focused on wiring AI into all their business operations. When leveraging AI for data science and analytical workflows, a common failure mode is that an enterprise user asks the same question twice and gets two different answers. Despite the revolutionary capabilities of LLMs for unstructured data and language understanding, they fail at common structured data analysis.

This isn’t an LLM problem. The problem is that the structured data which actually runs the business in BigQuery, Snowflake, Postgres, and other systems lacks meaning and business logic that lives somewhere the LLM can’t see it.

Every organization attaches its own meaning to its data. "Revenue" at one company may exclude trial customers who churned and at another it doesn't. Columns across multiple data sources might have the same name but carry different meanings. Business rules may be embedded in old queries, dashboards, or simply in the experience of the analysts who work with the data every day.

A good data engineer carries all of this in their head. An AI agent has to guess, and when it guesses wrong, it produces a confident, well-formatted, incorrect answer. Do that a few times and the organization stops trusting AI with its data entirely.

We believe the winners in enterprise AI will be the companies that deliver governed business meaning to AI at scale, and that's why we led Credible's seed round.

The Opportunity for Credible

The mechanical work of analytics has become inexpensive in the time of LLMs. An LLM can write any query and draw any chart in seconds. What it can't supply is what the data actually means: how metrics are defined, how entities relate, which rules apply, and who is allowed to see what. Charting is commoditized. Meaning is the product.

Credible is a context engine that captures an organization's business meaning and delivers it to AI agents, applications, and analytics tools at runtime. Built on Malloy, an open-source semantic modeling language, the platform works in three steps:

• Capture shared meaning. The meaning of a company's data is scattered across schemas, queries, catalogs, and the heads of its data engineers. Credible captures that institutional knowledge directly from the data ecosystem, informed by real queries and edge cases surfaced during use.

• Encode it in a semantic model. Credible's AI helps teams encode meaning into a governed Malloy model: versioned like code, easy to evolve, and readable by both humans and machines. The model is a living asset that accrues business context from real work rather than a frozen artifact that rots.

• Deliver it in context. Through APIs and MCP, Credible serves that meaning wherever questions get asked: the AI in a chat window, the coding assistant in an IDE, the agent running in production. No BI tool to log into, and no five-vendor stack to assemble.

On top of that foundation, Credible adds what enterprises need to run this in production: role-based access controls, multi-tenant isolation, auditability, and the scale to handle agent-volume traffic. Because it's built on open-source Malloy, customers get portability instead of lock-in.

Few people are better suited to build this than Kyle Nesbit. Kyle led Business Intelligence and Data Analytics at Google Cloud, where he spearheaded Looker's integration into the platform, and saw firsthand both the power of a governed semantic model and the cost of locking it inside a closed BI tool. Before that, he led Forward Deployed Engineering here at Gradient, so we've watched him operate up close for years. Credible is the product of someone who has lived this problem at the largest scale possible.

Today, Credible announced its $10M seed round led by Gradient, alongside SignalFire, K5 Global, and angel investors including Fivetran co-founder Taylor Brown and G2 co-founder and CEO Godard Abel.

We're proud to back Kyle and the Credible team as they build the layer that makes enterprise AI worth trusting: business meaning, defined once and delivered everywhere decisions are made.