Last edited on Oct 26, 2026
Jay Sobel, a data platform engineer at Ramp, used his dbt Summit 2026 session to describe what he calls "internal AI": not building AI into Ramp's product, but setting up the whole organization to use large language models productively. Jay frames it as the natural next step for analytics engineers rather than a separate discipline.
He argues analytics engineers have a real edge over software engineers here, both in their comfort working with a model's "black box" behavior and in their willingness to write the prose that agents need, which he half-jokingly claims is harder than Python.
Ramp Research, the company's data agent, launched about a year ago and became widely shared enough that dbt Summit's own keynote referenced it. Inspect is Ramp's remote agent platform: a ticket-style interface where each "ticket" is actually a sandboxed coding agent session, and it recently passed a million sessions.
Glass started as an internal prototype Jay nicknamed after Anthropic's assistant before teammates renamed it and it went viral. It's now pre-installed on new hires' laptops, pre-connected to internal systems, and especially popular with non-engineers like the sales team.
Rampplify is Ramp's internal app-hosting platform, built for quickly shipping and just as quickly tearing down internal tools. And WaveDash, the newest piece, is a dashboarding tool where every dashboard is a SQL file living in a repo and rendering to Markdown, so agents can read and edit dashboards the same way engineers do; Ramp is actively moving off Looker onto it.
Jay shared a chart of data-agent query volume that has gone from roughly 10,000 to 15,000 questions a month to a volume that now dwarfs the human-driven queries it's compared against. He also offered a caution: Ramp deliberately avoids locking itself into a single AI lab's models, pointing to swings like Cognition's (maker of Devin) reported $45 billion valuation as a reason to keep options open.
In the Q&A, Jay's advice for small data teams without the resources to build custom platforms was to lean on a design system, a chart library, and a "vibe-coded" internal repo rather than a from-scratch build. On hiring, he said he looks for analytics engineers with some software engineering instincts.
On measuring ROI on these AI investments, he was candid that Ramp doesn't do this rigorously today, calling it "black boxes and vibes." He also credited Ramp's security engineering team for making the internal AI push possible, and vented a little about AI-generated writing bloating internal documentation.
Watch every keynote and breakout from dbt Summit 2026 on demand. Watch dbt Summit 2026 on demand →
Get started in dbt
Join the analytics engineers building data infrastructure that actually scales.
Install dbt Wizard CLI
Get started with an agent purpose-built for analytics engineering. It knows which tool to call, which context to pull, and checks its own work before surfacing anything to you.
Latest posts
From prompt to PR: how Careem’s AI writes and validates dbt models before a human ever sees them
on Oct 26, 2026
How DocuSign cut analytics engineering time in half by pairing Claude Code with dbt v2, dbt MCP, and agent skills
on Oct 26, 2026
How Virgin Media O2 cut compute costs 30% by giving dbt a memory
on Oct 26, 2026
Join the largest community shaping data
The dbt Community is your gateway to best practices, innovation, and direct collaboration with thousands of data leaders and AI practitioners worldwide. Ask questions, share insights, and build better with the experts.