Ramp's Internal AI Agents Pass One Million Sessions

Ramp's internal AI push has reached a notable milestone: its remote agent platform, Inspect, recently passed a million sessions, according to data platform engineer Jay Sobel, who detailed the effort at dbt Summit 2026.

Sobel described "internal AI" as getting the whole organization to use large language models productively, rather than building AI into Ramp's product. He sees it as a natural next step for analytics engineers, who he argues have an edge over software engineers thanks to their comfort with a model's opaque behavior and their willingness to write the prose agents need.

Ramp Research, the company's data agent, launched about a year ago and drew a mention in dbt Summit's own keynote. Inspect presents each sandboxed coding agent session as a ticket-style interface. Glass, an internal prototype Sobel nicknamed after Anthropic's assistant before teammates renamed it, went viral and is now pre-installed on new hires' laptops, connected to internal systems and popular with non-engineers like sales. Rampplify hosts internal apps for quick shipping and teardown, while WaveDash renders every dashboard as a SQL file in a repo so agents can read and edit them; Ramp is moving off Looker onto it.

Sobel shared a chart showing data-agent query volume climbing from roughly 10,000 to 15,000 questions a month to a level that now dwarfs the human-driven queries it is compared against. He cautioned that Ramp avoids locking itself into a single AI lab's models, citing swings such as Cognition's reported $45 billion valuation as reason to keep options open.

Asked about small data teams without resources for custom platforms, Sobel advised leaning on a design system, a chart library and a "vibe-coded" internal repo instead of a from-scratch build. He said he hires analytics engineers with some software engineering instincts, and was candid that Ramp does not rigorously measure ROI on these AI investments today, calling it "black boxes and vibes." He credited the security engineering team for enabling the push and complained about AI-generated writing bloating internal documentation.

Source: dbt Labs Blog