A new consumer of the data stack
Agents are already good at the simple work data teams used to do: writing queries, making charts, much of the dashboard work. What they need is correct, data-grounded answers, for people and for autonomous agents alike. His example is a head of sales asking an assistant in Slack for the best ideas to improve pipeline conversion: open-ended, and a pattern he says to encourage.
In the last six months a pattern has emerged. Ramp built an interactive data analyst on its modeled data. Stripe built a knowledge agent on its data platform. OpenAI and Anthropic both built in-house data agents on their data platforms and wrote about it. What they share is that each points its AI at the same stack that has served people for years.
“They are using as a source of truth for AI agents for questions about data.”00:03:18
It works because agents behave so much like people. The work that makes data usable for an analyst (pulling every system into one database, joining across them, modeling it to fit the business) is the work that makes it usable for an agent.
“Modeling is fundamentally an exercise of simplification.”00:05:23
Evidence: transcript 00:01:22, 00:02:56, 00:03:18, 00:05:23
Why not just connect it to everything
The obvious move is to give the assistant MCP connections to every system of record. None of the companies he named do that, and he says it doesn't work. Ask a human analyst to answer a business question by logging into each system and clicking around: they could, but slowly and with many mistakes.
“It would take them like a year, and they would make a lot of mistakes.”00:07:04
An agent going through tools hits the same walls. The source data is messy (every company has a messy Salesforce), while the warehouse copy has been cleaned by the business's rules. Each system's query language is poor: Salesforce's own SQL dialect, Jira's query language, or for Slack just the search box. And the decisive one:
“And you can't even do joins across sources, right?”00:07:04
MCP itself isn't the villain. It's still the right tool for taking actions and for fetching a single record that just changed, and at Fivetran every client (a Slack bot, Claude, an autonomous support agent) reaches the warehouse through an MCP server. The problem is what the tools point at.
Evidence: transcript 00:07:04, 00:09:47
A map, kept in the database
Connect an agent to a database and tell it to look around, and the answers are poor; better than six months ago, still poor. It needs a map. Vendors sell context catalogs for this. His point is that you don't need one, because a database can describe itself.
“Metadata is a kind of data.”00:11:18
His proposal, framed like AGENTS.md but for databases: a schema called agents with a table called root, one column of markdown blocks describing the rest. At Fivetran the tables under it come from their Looker metadata; in the demo there are three, saying what the tables are, what the columns mean, and how to join them. The agent's system prompt only says to read agents.root first, so when the data changes, the map changes with it.
Why it fits: agents interpret open-ended questions and explain answers but forget everything, like the man in Memento writing himself notes. Databases remember and calculate but can't interpret.
“They have exactly the mirror image of each other's strengths and weaknesses.”00:11:18
Evidence: transcript 00:11:18, 00:15:24
Everything in one place
The new domain is unstructured data. Agents read so fast they can go through a thousand customer-call transcripts, so text belongs in the stack too. You don't need a separate vector database: the major data platforms have built-in vector indexes now, and text goes in ordinary text columns.
“And it's really convenient to have everything in one place.”00:14:29
His demo uses Formula 1: structured race data from a public API, plus the rules, glossary and post-race press-conference transcripts. Connectors written with Fivetran's connector SDK (he suggests letting an AI write them) land it all in BigQuery. dbt models normalize it and build the embeddings and vector index, and a Gemini agent answers questions, checked by a small set of evals in CI. A typical agent query is hybrid: a vector search for talk of tire problems, filtered to 2025 races. He built it in about five hours. Asked who the most improved driver is, the agent reads agents.root, probes the data, then writes a long multi-join query, much as a human analyst would.
Evidence: transcript 00:06:38, 00:14:29, 00:15:24, 00:22:31
The data team owns the context
His close: this simple pattern gets better as models get smarter, because it puts the model in charge instead of boxing it in with an over-engineered context system. The work belongs to the data team, whose skills at delivering analytics are the ones that deliver context. Use standard components, avoid anyone's walled garden, and keep definitions in code so agents can write code too.
“The data team should own your context.”00:26:44
Evidence: transcript 00:26:44