Green Dot Builds Meta:Context, a Business Schema for AI Agents in dbt

The problem: right math, wrong meaning

Green Dot runs an embedded-finance payments platform with a bank subsidiary. According to dbt Labs, it acts as program manager and issuing bank for a dozen-plus enterprise partnerships, including Apple Cash, Intuit QuickBooks, Amazon Flex, and Crypto.com, all processing high-volume debit transactions in a heavily regulated business.

Binkly's starting point was a specific failure mode. A semantic layer gets the query and the math right, but it does not tell an agent what a number means. In one internal test, an agent found that Partner B's cost-efficiency ratio was roughly twelve times worse than Partner A's and recommended that Green Dot help Partner B improve it. The two partners run on entirely different payment rails, so the gap was expected and the recommendation was wrong.

Binkly framed this as a context problem: understanding a metric's healthy range, what is driving a change, and what to do about it. Feeding an agent scattered docs, emails, and spreadsheets does not fix it, he said, because the model still has to work out which pieces are relevant to which metric.

How Meta:Context works

The schema lives directly in the meta block of dbt's semantic layer YAML, next to the metric it describes. It has five categories: descriptions, expectations, investigation, relationships, and decisions. Binkly nicknamed the set "dried" as a memory aid.

Keeping context in the same YAML as the metric has two effects, according to Binkly:

• One semantic layer API call returns both the number and its context.
• The context goes through the same review process as the metric: pull requests, CI, and the dbt manifest.

Binkly said the fields are typed rather than freeform, which makes them checkable. Teams can test whether a healthy range is populated or stale, whether an agent followed the documented investigation path, or whether it fabricated an escalation step. The structure also makes context coverage measurable across a project's metrics.

Test results

Binkly compared how an AI agent handled the same question about a program's decline rate with and without a completed context card. Without the card, the agent correctly avoided a false conclusion but could not verify its hypothesis. With the card, it identified the actual cause, flagged what it could not prove, and named the query that would confirm it.

He also asked seven different models to write a context card from the same raw materials, with all analytical judgment stripped out. All seven flagged an untested claim appropriately, but none reconstructed the one rule that came from an analyst's actual investigation. Binkly used that result to argue that the card is "the residue of an investigation," not a smarter model's writing.

Status and advice for other teams

Green Dot is piloting conversational BI on top of the schema and has no enterprise usage numbers yet. Binkly said two data leaders at regulated European organizations, a public-transport agency and an energy-grid operator, reached out independently after he shared the approach in the dbt community Slack. Both started building their own context cards using the same five-layer structure.

For teams that want to try it, Binkly recommends starting with one KPI and three fields: purpose, healthy range, and business rules. The team can then compare how an agent's answer changes before and after. Because the schema is plain YAML in a dbt project, Binkly said it is portable to other tools if a team later wants to move it.

The session, along with every keynote and breakout from dbt Summit 2026, is available on demand.