The setting
's lets data teams build, transform, govern and validate their data workflows, either in a web browser or with command-line tools.
The catalog reflects a new metric definition while AISQL still reasons from the old one.
Fast context refresh protects governed answers, while unnecessary rebuilds add latency and compute cost.
Who you’d be doing this for
“I changed the definition this morning, and the next answer still used last week's logic.”
Petr Markovic · Analytics Engineer · a character in this brief
Maintains governed dbt metrics used by analysts and downstream AI workflows.
What is at stake
Changed metric definitions are reaching the catalog while AISQL answers from stale context. You must weigh faster refreshes against query latency and needless rebuilds.
Why it isn’t already fixed
Every obvious fix costs something else. That’s the part you’d have to decide.
- fresh metric grounding vs query latency
- targeted invalidation vs operational simplicity
- fast release vs verified evaluation evidence
- catalog updates vs retrieval consistency
Why Fivetran
AISQL depends on dbt metric definitions to produce warehouse-native SQL that analytics engineers can trust.
Written with these in mind
Not your kind of problem? 54 more at Fivetran, or browse every organization.
Where the organization context comes fromThe postings this brief was written from, and the reporting read before it.
The organization context comes from public information. The situation and characters are written for the brief.
Job postings it was written from
Public descriptions of the work at Fivetran. They are listed as sources, not as openings. The lines under a posting are its own words, where it meets this brief.
- Staff R&D Software Engineer, Fivetran AI · no longer posted
- Drive the AISQL capability forward: natural language to SQL grounded in dbt metric definitions, executed natively against the warehouse
- Research emerging techniques in retrieval, reasoning, and agentic AI, and decide what’s actually worth pursuing for Fivetran AI’s roadmap — then convince others
- Sr. Principal AI Systems Architect (opens in a new tab)
- This role owns the architecture behind the analyst AI system, including metadata contracts, tool interfaces, context assembly, prompt and skill orchestration, evaluation loops, and the operational patterns that make agents useful in real workflows instead of isolated demos.
- Design retrieval and context strategies that give the model the right information at the right time with strong cost, latency, and quality tradeoffs.
- Senior R&D Software Engineer, Fivetran AI · no longer posted
This is the setup. The work is inside.
Running it puts you in the room: the full situation and its constraints, stakeholders who push back in their own words, and the decisions that are yours to make. What you produce becomes a Day One Plan — work you can show someone instead of describing.
Where similar work appears
Other live briefs connected to this one, and what connects them.
Similar problems at other organizations
Like this one, each of these turns on fixing one thing without breaking another.
Fiscal-period accuracy fell from 97% to 78% across 180 checked questions.
You must weigh a focused model-side correction against a rollback that protects other context-dependent answers.
Three eligibility loads have exceeded the 15-minute freshness SLA, leaving availability data stale.
You need to restore the flow while making the next schema change visible before it delays customers.
Twenty-two percent of yesterday’s hypertension readings are absent from the morning dataset.
You need to restore the data without creating duplicate or misleading history.
The same setting, from another role
Each is set around Fivetran’s dbt platform, written for a different role. They’re separate briefs, not the same scenario.
Enterprise CSMs must reconcile conflicting records in 17 of 40 onboarding questions.
You decide whether AI may infer a remaining step from conflicting notes, or must leave that part of the reply for the CSM.
Six of eight Enterprise customers depend on one person to maintain their dbt models.
You need to weigh proof of independent maintenance against the practice time customers will actually protect.
An enterprise customer may remove two business units from its annual renewal today.
You must weigh preserving ten reporting workloads against committing to savings that a full monthly close has not tested.
Other problems with a similar purpose
In each of these, as in this one, the aim is to build machine-learning systems that earn trust by staying explainable where their errors carry real consequences.
An audit found sensitive fields in 7% of shared artifacts that manual review missed, and the classifiers disagree on the edge cases.
You have to build evidence for a production commitment you cannot easily reverse.
One threshold sends very different numbers of drivers into coaching in each region, and reviewers there agree less often.
You have to balance fair treatment of drivers against enough coverage to improve coaching.
Confirmed takeover messages are reaching guests, while legitimate pre-check-in threads are also being blocked.
You must weigh fraud prevention against a guest losing the instructions needed to enter a booked stay.
