For Fabric
Data Workers runs the detect, diagnose, fix, review and verify loop on a Microsoft Fabric estate when nobody is in the workspace. A failed pipeline overnight is diagnosed, a fix is proposed against the offending activity, a named human approves anything irreversible, and every applied change carries a receipt with its blast radius and a rollback.
For a data team on Fabric or Synapse: OneLake, lakehouse and warehouse items, Data Factory pipelines, semantic models in Power BI, and Purview for governance.
Last updated: September 10, 2026 · Dhanush Shetty, founder, Data Workers
None of these is a Fabric defect. They are the failures a busy estate produces, and the hours between one of them starting and somebody verifying a fix are what this platform is for.
A Data Factory pipeline failed overnight and the report still renders
Power BI does not go blank when the pipeline behind it fails. It shows yesterday's numbers with today's date on the page. Somebody presents from it at 09:30 and the error is found in the meeting, not in the monitoring hub.
A semantic model in Direct Lake mode silently fell back to DirectQuery
The report keeps working and gets slower. Nobody files a ticket about slow, so the fallback sits there for weeks consuming capacity, and the investigation starts only when somebody asks why the F SKU is saturated.
A capacity throttled and the jobs that lost are not the ones you would have chosen
Fabric capacity is shared across workloads, so a heavy ad-hoc query and a nightly refresh compete. Which one loses is an accident of timing. Finding out which refresh was smoothed or delayed means reconstructing an hour from the capacity metrics app.
A lakehouse table in OneLake with no owner, no description and three shortcuts pointing at it
Shortcuts make the blast radius of a change genuinely hard to see, because the consumer may live in a different workspace owned by a different team. The person who could tell you what the table means left, and Purview records that it exists rather than what it is for.
A schema change upstream in an on-premises source arrives without warning
A column is renamed in a source system, the gateway keeps syncing, and the mapping quietly drops it. The downstream measure reads zero rather than failing, which is the worst of the two outcomes because zero looks like a business result.
Alerts are not fixes. Detection tells you the first of those sentences. The rest is the work.
Runs the loop
Autonomous Data-Conductor
The Autonomous Data-Conductor watches Fabric for the conditions that start an incident: a failed or delayed pipeline run, a refresh that did not complete, a table that missed its expected arrival, a semantic model no longer meeting its freshness contract. It then runs the whole loop rather than stopping at detection. You set the autonomy dial per class of change: read-only, propose-and-wait, or apply-then-report for reversible work such as a retry or a refresh. Anything irreversible always stops for a named human, whatever the dial says.
Does the work and writes
Data-Agents Swarm, 20 specialised agents
The Data-Agents Swarm is 20 specialised agents. On Fabric the ones that earn their place first are incidents, which does statistical anomaly detection and graph-based root cause across the pipeline and the model behind a report; schema, which diffs metadata and detects renames rather than reporting a drop and an add; quality, which scores tables on five weighted dimensions against a rolling baseline; and governance, which scans for PII in lakehouse tables and proposes masking. The open-source core reads and recommends. Writing is the paid tier.
What the agents reason on
Data Context Wizard
The Data Context Wizard is the graph the agents reason on, and on Fabric it matters more than on a single-warehouse stack because a Fabric estate is rarely only Fabric. It holds what OneLake and Purview record, plus what they do not: metric definitions as your business uses them, ownership that reflects who actually answers, past corrections and why they were made, and column-level lineage that crosses from Fabric into Snowflake, BigQuery or Databricks without stopping at the boundary. Every fact carries its provenance, so an agent can say where it learned something and you can disagree with the source.
Where humans approve and audit
Spellbook Data Catalog, in preview
Spellbook Data Catalog, in preview, is where a human approves an agent's proposed change and where the audit trail lives afterwards. It does not replace Purview. Purview stays your Microsoft governance surface; Spellbook is the approval and audit plane for changes an agent proposed, which is a thing no catalog was built to be.
What each module has to do to qualify as this kind of product, stated generically so you can use it on any vendor: the autonomous agentic data platform.
The first three steps need no form, no account and no key we issue. The open-source core is Apache-2.0: 11 agents and 160+ MCP tools that read, analyse and recommend.
The full list of who we are not for, including the cases where the honest answer is to buy something else: should you use Data Workers.
Does this work on Synapse as well as Fabric?
The agents read through the warehouse and metadata interfaces rather than a Fabric-only API, so a Synapse dedicated pool or serverless endpoint is reachable the same way. Fabric is where the newer surfaces such as OneLake shortcuts and Direct Lake semantic models are, and those are the ones the context graph models explicitly.
Does this replace Purview?
No. Purview stays the Microsoft governance and catalog surface. The Data Context Wizard reads what is there and adds what a catalog does not hold: metric definitions, acted-on ownership, past corrections and cross-cloud column-level lineage, provenance-stamped. Spellbook, which is in preview, is the human approval and audit plane for agent-made changes.
We are on Fabric and something else. Does that help or hurt?
It helps, and it is the case the context graph was built for. The graph spans warehouses rather than being native to one, so a change that crosses from Fabric into another warehouse stays legible to the agent instead of stopping at the boundary. Warehouse-native agents answer no to that question by design.
Can it change production in Fabric without us?
Not for anything irreversible. The write path is propose first, dry run where one exists, and a named human approves before an irreversible change lands. Every applied change carries a receipt: the diff, the approver, the blast radius across downstream tables, models and reports, and the rollback path.
Two next steps
Run the read-only agents on Fabric yourself, free: clone the open-source core.
Or bring one real incident to a 45-minute session and watch the approval gate stop the agent: book a time. Pilot Program $7,500 one-time, then Scale from $1,000 per month, no usage meter, as published on 10 September 2026. Pricing.