Translytical task flows, generally available since the March 2026 Power BI update (Microsoft Learn, 2026), let your Power BI report write to a Microsoft Fabric database or call an external API by passing report context to a Fabric user data function. In Atlan’s context-layer reading, each click is an agent tool call in miniature, and it inherits the same open question: what context justified the write? Write-back is GA for SQL databases, warehouses, and lakehouse files, while Microsoft’s Cosmos DB guide (Microsoft Learn, 2026) still says preview.
Microsoft gave the feature its own FabCon Europe 2026 session: Nadim Abou-Khalil’s “From Insight to Action: Translytical Taskflows and AI in Microsoft Fabric,” which Microsoft’s session preview (Microsoft Fabric Blog, 2026) says will “close the loop between analytics and action.” The FabCon Europe 2026 guide and FabCon sessions by role cover the rest of the week.
| Attribute | Detail |
|---|---|
| What it is | A Power BI report action that runs a Fabric user data function |
| Status | GA since March 2026; the Cosmos DB guide still says preview |
| Core mechanism | Report filters, selections, and input values passed as function parameters |
| Underlying layer | Fabric user data functions, serverless Python 3.11 |
| Not to be confused with | Fabric’s “Task flows” canvas, or Forrester’s “translytical data platforms” |
| Best for | Simple write-back and approvals driven by report context |
What are translytical task flows, and where does the name come from?
Translytical task flows are Microsoft’s name for report-driven actions in Fabric, borrowing an older database term. Per Microsoft Learn’s overview (Microsoft, 2026), they automate “end-user actions like updating records, adding annotations, or creating workflows.”
The feature moved from preview to GA in ten months, per Microsoft’s Power BI update archive (Microsoft Learn, 2026):
- May 19, 2025: public preview announced (Microsoft Fabric Community, 2025).
- September 2025: enabled by default in Power BI.
- March 18, 2026: generally available, announced in the Power BI March 2026 Feature Summary (Microsoft Fabric Community, 2026).
- May 2026: optional parameters and numeric input added.
Microsoft’s standalone GA overview post followed on April 20, 2026; the release itself shipped in March.
“Translytical” joins transactional and analytical; Microsoft’s SQL database guide (Microsoft Learn, 2026) describes translytical apps as needing “both transactional and analytical access to the same data.”
Is this the same as Forrester’s “translytical data platforms”?
No. Forrester’s Q4 2024 Wave (Forrester, 2024) evaluates 15 providers in a category it has tracked since its Q4 2017 Wave (Forrester, 2017): database engines that, in analyst Noel Yuhanna’s words, bring “transactional and analytical workloads into a single, unified platform” (Forrester, 2025). Microsoft’s feature sits a layer above, a report calling a function on existing Fabric sources.
Not to be confused with Fabric’s “Task flows” feature
Fabric’s “Task flows” is a workspace canvas for mapping how items relate, GA since May 5, 2025 (Microsoft Fabric Community, 2025). Its connectors “don’t represent the flow of data” (Microsoft Learn, 2025), and it never writes a record.
| Term | Coined by | What it actually is | Status |
|---|---|---|---|
| Translytical task flows | Microsoft | Report-to-function write-back and API actions | GA (March 2026); Cosmos DB guide says preview |
| Task flows | Microsoft Fabric | Workspace canvas that maps items; moves no data | GA (May 2025) |
| Translytical data platforms | Forrester | Database engines running transactions and analytics together | Analyst category |
Of the three, only the first changes a record, which makes it the one that needs a context check before it runs.
How do translytical task flows work?
Every translytical task flow runs on Fabric user data functions, a serverless layer that a report button, a pipeline, or an external caller can invoke.
The three-component model
Microsoft’s Cosmos DB how-to (Microsoft Learn, 2026) names three parts: an operational data store, a data logic layer of user data functions, and a Power BI visualization layer. Microsoft’s GitHub sample (Microsoft, 2026) wires them together.
Fabric user data functions: the engine underneath
According to Microsoft’s user data functions overview (Microsoft Learn, 2026), one function can be called from pipelines, notebooks, Activator rules, translytical task flows, and external systems. Each function “automatically exposes its own unique REST endpoint,” secured with Microsoft Entra ID.
That endpoint is also how an agent could reach the function, alongside Fabric’s own MCP servers and other agent interoperability protocols. An AI control plane can hold that endpoint to the same policy as any agent tool, the problem teams face when they build MCP servers for enterprise data.
Three mechanics matter in practice:
- Reports pass filter and selection context in the request payload.
- Updates show immediately in Direct Lake or DirectQuery reports, after an automatic refresh in import mode.
- Service limits (Microsoft Learn, 2026) cap requests at 4 MB, runs at 240 seconds (100 through a public endpoint), and responses at 30 MB, on Python 3.11.
| Data source | Write-back status | Notes |
|---|---|---|
| Fabric SQL database | GA, native connection | Microsoft’s recommended source for heavy read/write |
| Fabric warehouse | GA, native connection | Structured read and write |
| Fabric lakehouse | GA for files | SQL analytics endpoint is read-only |
| Mirrored database | Read-only | Not a write-back target |
| Cosmos DB in Fabric | Separate guide | Still says “currently in public preview” (updated June 2026) |
SQL databases replicate to OneLake in near real time, the latency real-time data for AI agents depends on. Any check you want on a write has to live at the function, because the button is only one caller.
What can a translytical task flow actually do?
Microsoft Learn names five scenarios a translytical task flow can run, and in every documented example a person clicking a report element starts it. Per the overview (Microsoft Learn, 2026), they are adding, editing, and deleting data, calling an external API, and surfacing in-report notifications.
Microsoft’s examples show the range:
- A sales rep requests a 10% discount (Microsoft Learn, 2026) with a justification, and the function posts it to Teams for approval.
- A “Generate AI Suggestion” button sends a prompt to the Azure OpenAI Responses API.
- Microsoft’s GA overview post (Microsoft Fabric Community, 2026) adds data correction “at the point of discovery” and review of AI-categorized data.
One scenario goes undocumented: an agent calling the same function. It would behave identically, which makes giving AI agents access to enterprise data a write decision as well as a read one. That move to an agentic workflow changes the risk and guardrail needs of whichever type of AI agent calls it, even when the code is unchanged.
Translytical task flows vs. Power Apps: when should you use which?
Microsoft’s documentation doesn’t compare translytical task flows with Power Apps, so the working rule comes from practitioners who have built both.
Nicky van Vroenhoven, Microsoft MVP, writing in April 2026 (nickyvv.com, 2026), reserves task flows for “simple, Fabric-native write-back where the report context is the main input.” Power Apps, in his view, suits richer interfaces and writes outside Fabric.
Jon Stjernegaard Vöge, Principal Business Consultant and Microsoft Data Platform MVP at Inspari, welcomed a Fabric-native option because Power Platform licensing had blocked past projects. Writing in September 2025, before GA (Downhill Data, 2025), he was still “likely to pick Power Apps over Translytical Task Flows in their current state.”
Both posts predate the May 2026 update (Microsoft Learn, 2026) adding numeric input and optional parameters, so the input gap is now narrower.
| Dimension | Translytical task flows | Power Apps |
|---|---|---|
| Input model | Input slicer plus report context | Richer forms and custom UI |
| Write targets | Fabric SQL database, warehouse, lakehouse files, and APIs | Can write outside Fabric |
| Licensing | Runs on Fabric capacity | Power Platform licensing |
| Best fit | Simple write-back driven by report context | Richer UI or non-Fabric targets |
Both tools put a write button in front of someone reading a report. The tool decides the interface; whether the row on screen was current stays your team’s question either way.
What are the limitations of translytical task flows right now?
Three documented constraints apply today, per the Limitations section (Microsoft Learn, 2026), and the function layer adds more:
- Functions must return a
strtype to be added to a report. - Power BI Embedded works only in secure embed scenarios.
- Data function buttons don’t rebind across workspaces, so a report promoted through deployment pipelines still calls the original function until someone updates it.
The user data functions service page (Microsoft Learn, 2026) adds that only the item owner can edit and publish function code, and service principals aren’t supported through managed connections. That makes owner-only editing an AI agent governance fact: whoever owns the function owns what it writes.
Rebinding is the likeliest production problem. A promoted button that silently calls last quarter’s function is an agent guardrail problem, and measuring task success won’t catch it if that function still returns a success string.
Keep maturity in proportion: every documented example has a human clicking, and none has an agent calling unattended.
Why translytical task flows look like the action half of an agent loop
Strip away the Power BI report and a translytical task flow is contextual parameters in, a database write or an API call out, which is what an agent’s tool call does. Today’s caller is a person, but the function can’t tell a person from an agent, so the parallel matters before agents arrive. In an agent loop, the acting step is a function call with arguments, the shape that MCP and function calling formalize.
Microsoft built the feature for report users and never uses this vocabulary, but its SQL database guide (Microsoft Learn, 2026) says captured actions can route to Fabric Notebooks “for AI-assisted processing” before the data lands.
Once a function can change real records, three questions come before each call:
- Was the record current? A write against stale data is a confident mistake, so context freshness matters more here than on reads.
- Was the caller allowed to change that record? Generic connections, per Microsoft Learn (2026), use the item owner’s identity, whoever clicked. That puts AI agent identity and agent access control on the design list.
- Can the action be traced to the data it acted on? The GA overview post (Microsoft Fabric Community, 2026) calls the Teams discount flow a complete audit trail. That trail holds the request and justification; decision traces also record which definitions and data versions it relied on.
Microsoft doesn’t claim to answer these, and no Fabric feature fails by leaving them open. Each asks about the record’s surroundings (freshness, permissions, lineage), which is what a context layer holds.
In harness engineering terms, the pipeline can be trusted to run the call; trusting the call itself takes context. Atlan’s approach to an enterprise context layer is to carry each record’s definition, owner, freshness, and policy to whatever is about to act on it, person or agent.
The read-side counterpart, on ontology status, refresh, and permissions, is Fabric AI agent governance questions. Governing agents on Fabric takes both halves.
What has to be true before a translytical task flow fires
Translytical task flows turn a Power BI report into something that acts: report context goes in, and a record changes. Pipelines, Activator rules, and any REST client with an Entra ID token can call the same function, and an agent is one more REST client.
So the question for your team comes before the click. Can you say whether the data was current and AI-ready, whether the caller could change that record, and which definition the filter relied on?
Teams that prepare enterprise data for AI agents answer those from context, the working core of context engineering. The AI-ready data checklist is a practical place to start.
FAQs about translytical task flows
1. What is a translytical task flow in Microsoft Fabric?
A translytical task flow is a Power BI report action that runs a Fabric user data function. A user selects data, enters a value, and clicks a button, and the function adds, edits, or deletes records, calls an API, or stores an in-report notification.
2. Are translytical task flows generally available or still in preview?
They are generally available. Native write-back covers Fabric SQL databases, warehouses, and lakehouse files. Microsoft’s Cosmos DB guide still carries a preview note, so check that source separately.
3. What is the difference between translytical task flows and Fabric user data functions?
A user data function is the serverless Python code that performs the action. A translytical task flow is one way to call it, from a Power BI button. Pipelines, notebooks, and Activator rules can call the same function, and external apps reach it through its REST endpoint.
4. Can translytical task flows call external APIs, not just write to a database?
Yes. Calling an external API is one of the five documented scenarios, with examples that post a discount request to Microsoft Teams and send a prompt to the Azure OpenAI Responses API. Each call can run up to 240 seconds.
5. Do translytical task flows work with Cosmos DB in Fabric?
Yes, with a caveat. Microsoft’s Cosmos DB in Fabric how-to still says the feature is in public preview, as of its June 2026 update, and Cosmos DB is absent from the native write-back connection list.
6. Translytical task flows vs. Power Apps: which one should you use for write-back?
Use translytical task flows for simple write-back inside Fabric, where report context supplies most of the input. Two Microsoft MVPs recommend Power Apps for richer forms or writes outside Fabric. Licensing differs too: task flows run on Fabric capacity, while Power Apps needs Power Platform licensing.
7. What are the limitations of translytical task flows?
Microsoft lists three: functions must return a string, Power BI Embedded works only for secure embed, and buttons don’t rebind across workspaces after deployment. The function layer adds a 4 MB request limit, a 240-second timeout, and owner-only editing.
8. How is “translytical” different from Forrester’s “translytical data platforms” category?
Forrester uses translytical data platforms for database engines that run transactional and analytical workloads on one platform. Microsoft’s translytical task flows are a report feature that calls a function to write data or call an API. The two sit at different layers.
Sources
-
Previous monthly updates to Power BI Desktop and the Power BI service, Microsoft Learn
-
Power BI March 2026 Feature Summary, Microsoft Fabric Community
-
Translytical task flows (Preview), Microsoft Fabric Community
-
Translytical Task Flows (Generally Available), Microsoft Fabric Community
-
Service details and limitations of Fabric user data functions, Microsoft Learn
-
Translytical task flows with Cosmos DB in Microsoft Fabric, Microsoft Learn
-
Use SQL database as the source data engine for translytical applications, Microsoft Learn
-
Task flows in Microsoft Fabric (Generally Available), Microsoft Fabric Community
-
FabCon Europe 2026: The sessions we’re most excited to bring to Barcelona, Microsoft Fabric Blog
-
The Forrester Wave: Translytical Data Platforms, Q4 2024, Forrester
-
The Forrester Wave: Translytical Data Platforms, Q4 2017, Forrester
-
Translytical Databases Are Fueling Modern AI Apps, Forrester
-
Comparing write-back options for Power BI and Fabric, Downhill Data
