From Events to Answers: Real-Time Intelligence and Fabric IQ from zero
One invented oud house, one live demo, and one business rule that Fabric keeps in two places.
By Dr. Estera Kot, Microsoft MVP, CTO at Clouds on Mars, and former Microsoft Principal Product Manager for the Apache Spark runtime in Fabric. Published . Product facts checked against Microsoft Learn on November 3, 2026.
An AI agent is only as good as the business meaning it is given. Real-Time Intelligence is the Microsoft Fabric workload for data in motion. It turns a stream of events into a signal. Fabric IQ is the Fabric layer that gives agents business meaning. It turns a question into a governed answer. Both must use the same business definition, or the signal and the answer disagree.
Dr. Kot taught both from zero in a two-hour community session in Doha, with one live demo end to end. This overview covers the story, the two halves, and the gap that matters most.
In short
In the demo, “winner” means first choice above 60. The rule is written twice: as an Activator condition and as an ontology rule.
The copies already differ, because only the Activator condition requires 20 testers. Microsoft Learn states that the new ontology experience has no direct Activator integration.
An ontology can bind to Eventhouse data in preview. The session materials do not show that join working live.
A free trial covers the stream, the Eventhouse, and the lakehouse. Copilot and the data agent need a paid F2 or higher capacity.
Figure 1. The whole build on one map. Two paths share OneLake and end at one person. The number 60 sits in two engines. The dashed join between them is the preview step the demo planned but did not prove live.
Why an AI agent needs business meaning
An agent cannot answer a business question that rests on an undefined word. A user asks which formulas are winners. No table has a winner column. The trial table holds first_choice_pct, a share from 0 to 100. Until somebody defines the word, the agent guesses with confidence.
The third slide of the session gave the answer: “A business rule belongs in the dictionary, not in the code.” In Fabric, the dictionary is the ontology. In Dr. Kot's view, the most important rule in a company often sits in a spreadsheet, a report macro, or one person's memory.
The demo story: an invented oud fragrance house
The demo follows an invented oud fragrance house that tests new blends, called formulas, before a launch. All data in the demo is synthetic.
Each test is a paired blotter test. A tester gets two paper strips, called blotters: the new formula and the current reference. The blotter the tester reaches for first is the vote. The share of votes for the new formula is called first choice. Above 50, the panel prefers the new formula. The house calls a formula a winner only above 60.
The data is small on purpose:
6 formulas, OB-101 to OB-106.
10 tables as CSV files, including report passages and optimizer output.
5 study reports as PDF files, 3 in English and 2 in French.
7 past trials in the tables, plus 1 live trial.
40 live events from a launch evening, one per tester, from 4 sampling booths.
The live trial retests OB-106, which reached 53 percent in its last study. The running share passes 60 at tester 31 with 61.3 percent and ends at 65.0 percent.
Real-Time Intelligence and Fabric IQ in one demo
The demo runs in two halves: events first, answers second.
The path of one event. A booth sends one JSON event per tester. Eventstream carries it into an Eventhouse table, and a KQL function recomputes the running share. Activator runs a query on a schedule and sends a message when the share passes 60 with at least 20 testers. Part 2, Real-Time Intelligence in plain words, follows that event stop by stop.
The path of one question. A user asks for the cost per kg of the winners and what the study recommended. In the design, the ontology resolves the word winner. The data agent writes SQL against the tables and reads report passages stored as a table. A language model writes the sentence. Part 3, What an ontology is, and why a data agent needs one, explains both items.
Microsoft Learn states that a data agent does not read PDF files directly, which is why the passages sit in a table. On stage, one question in French got a correct answer from a French report. Learn lists non-English languages as not supported by the data agent today, so treat that as a demo result, not a feature. Part 4, The agent writes the sentence, the table holds the number, covers both points.
On Learn, Eventstream, Eventhouse, Activator, and Real-Time Dashboards are generally available GA. The data agent is generally available too GA. The ontology and its use as agent context are in preview Preview. The Copilot dashboard page shows no release label TBC.
Sixty, twice: one business rule in two engines
In the demo, the number 60 lives in two engines. Activator holds a condition: first_choice_pct above 60, with at least 20 testers. The ontology holds a rule: a formula is a winner only when first_choice_pct is above 60. The number is the same, but the sentences differ. A small trial above 60 would be a winner for the agent and silent for the alert.
On Microsoft Learn, an ontology rule is a natural-language statement on the Rules page of the ontology. It links to entity types, properties, or relationships, and the ontology does not run it against data. The new ontology experience has no direct Activator integration. The old experience supported Activator alerts on rules, and it retires on January 31, 2027.
Figure 2. Sixty, twice. The live trial crosses 60 at tester 31, and the Activator condition also waits for 20 testers. The ontology rule has no such guard. At 60, two past formulas are winners. At 50, five formulas pass. Data from the demo kit.
The number appears in more than two places. The dashboard prompt asked Copilot for a line at 60. The fallback version of the live edit, not needed on stage, would have put the rule into the agent instructions.
Can a Fabric ontology bind to Eventhouse data?
Yes, in preview. Microsoft Learn lists eventhouses and KQL databases as ontology binding sources, next to lakehouses, warehouses, semantic models, SQL databases, and mirrored databases. The ontology agent can bind entity types to KQL tables, including time-series bindings.
Three details matter for a live demo. New rows in a source appear after a manual refresh of the ontology item. A data agent picks up ontology changes within 10 minutes, unless you refresh its context by hand. A known issue on Learn: a data agent does not work with an ontology bound to semantic models.
The demo planned a join on this. With the Trial entity bound to the Eventhouse, “which formulas are winners” would also return OB-106 at 65 percent. On stage, Dr. Kot did not claim the join. She used the fallback line: “Tonight sixty is the same sentence in two places.” The join was named as the next step, in preview. The conference demo of the new ontology is in Ontology in Fabric IQ.
Define a business rule once, bind every engine
Dr. Kot's position: define a business rule once, and bind every engine to it. Fabric does not do that for you today. Activator keeps its own condition, and the ontology keeps its own sentence. A data agent can keep a third copy in its instructions.
That third copy is a trap. If the agent instructions also carry the rule, an ontology edit changes nothing. The agent answers from its instructions. The kit removed the rule from the instructions on purpose, so the live edit could change the answer.
Until one definition drives every engine, treat each copy as a controlled duplicate:
R1. Write the rule once in the ontology, in business words, and name one owner for it.
R2. List every engine that copies the rule: Activator conditions, agent instructions, and dashboard lines.
R3. Test every copy against the same cases. In the demo, a threshold of 60 returns OB-104 and OB-105, and a threshold of 50 returns five formulas.
R4. Change all copies in one change, or change none of them.
A demo that hides this gap teaches the wrong architecture.
What the demo deliberately leaves out
The demo shows mechanics, not quality at scale. The session materials list six gaps. Five documents stand in for a corpus of thousands. Report passages were prepared in advance, so PDF extraction is skipped. The optimizer result is a table, not a model endpoint. The build has no vector search, records no user feedback, and does not monitor the agent. Nothing in it measures cost per 1,000 questions. Part 5, From demo to production, takes each gap and compares three ways to build the same layer.
How a room of beginners got hands-on access
Every participant got a seat in Fabric. Fifty accounts in a dedicated workshop tenant had Fabric (Free) licenses. A demo workspace ran on an F32 capacity and a playground on an F16. The room worked through five handout steps for 17 minutes. The checks found one trap: a new account's data agent answered only after its first portal sign-in. Part 6, Fabric for 50 people in one evening, covers the setup and the teardown.
How the session was designed for beginners
The session was written for people who have seen Power BI or Excel but never Fabric. Every term got one picture and one example before the demo used it. No query language was taught. Each demo beat carried one idea and one question to the room. Part 7, Teaching Microsoft Fabric from zero in two hours, explains the method.
How to start with Real-Time Intelligence and Fabric IQ on Monday
You can repeat the first half of the demo on a free trial. The session script sizes each step at about an hour.
Start a Fabric trial, then create a workspace and a lakehouse. Microsoft Learn states that the trial lasts 60 days.
Load the demo kit: 10 CSV files, 5 PDF reports, and one notebook. Check one number: the highest first choice is OB-104 at 68 percent.
Create an Eventstream with a custom endpoint source, run the simulator from the kit, and watch the first event land in an Eventhouse.
The closing slides in Doha pointed to the Fabric Falcons community for the kit. The kit is now public on GitHub: from-events-to-answers-kit, with the data, the five reports, the notebook, the agent instructions, the simulator, the runbook, and the expected answers.
Before you plan a project. Copilot and the data agent do not run on a trial. They need a paid F2 or higher capacity, or Power BI Premium P1 or higher. The ontology needs two tenant settings from your Fabric admin. Outside the US and the EU, Copilot is off by default until the admin allows processing outside your region.
Real-Time Intelligence and Fabric IQ from zero, on one invented oud house. The demo writes the winner rule twice, in Activator and in the ontology, and Fabric still has no link between them.
Five parts, one job each. Follow one tester's vote through Eventstream, Eventhouse, a dashboard, and an Activator rule, and see why the alert waits for 20 testers before 60 counts.
An ontology in Microsoft Fabric IQ is the dictionary of the business. One worked example: a rule that a winner has first choice above 60 changes the data agent's answer, with no change to the data.
A Fabric data agent should write the sentence and the query, never the number. One demo answer taken apart, the guards that keep figures in tables, four build failures, and capacity facts.
A Fabric IQ demo proves the mechanics, not quality at scale. Six gaps to close, three ways to build, preview risk, and a production checklist for data agents.
A hands-on Fabric session fails at sign-in, not at the demo. Accounts, MFA, a CU budget, and teardown for 50 people, from one evening at Microsoft Qatar.
Beginners leave at the first word nobody explained. How a two-hour Fabric session was built: five level rules, a minute clock, a glossary in pictures, a paper walkthrough, and a recovery line per demo step.
Key facts at a glance
Learn facts were checked on November 3, 2026.
ID
Fact
Source
K1
The session ran on October 14, 2026, in Doha, for two hours at level 0 to 200.
Session materials
K2
The demo holds 6 formulas, 10 tables, 5 PDF reports, and 40 live events.
Demo kit
K3
The live trial passes 60 percent at tester 31 and ends at 65.0 percent.
Event simulator
K4
At a threshold of 60, the past winners are OB-104 and OB-105.
Expected answers
K5
The ontology is in preview, and its old experience retires on January 31, 2027.
What is Real-Time Intelligence in Microsoft Fabric?
Real-Time Intelligence is the Fabric workload for data in motion. Eventstream brings events in, and an Eventhouse stores and queries them. Real-Time Dashboards chart them, and Activator acts when a condition is met. Microsoft Learn shows no preview label on these core parts.
What is Fabric IQ?
Fabric IQ is the layer that gives people and agents business context in Fabric. Microsoft Learn names three layers: unified data in OneLake, Power BI semantic models, and ontologies. The IQ workload is in preview.
Can a Fabric ontology bind to Eventhouse data?
Yes, in preview. Microsoft Learn lists eventhouses and KQL databases as binding sources, including time-series bindings. New rows in a source appear after a manual refresh of the ontology item. The demo in this series planned such a binding but does not show it working live.
Can an ontology rule trigger an Activator alert?
Not in the new ontology experience. Microsoft Learn states that the ontology does not run its rules against data and has no direct Activator integration. You build the Activator rule separately, so the business rule exists twice.
Does the Fabric free trial include Copilot and data agents?
No. The trial lasts 60 days and includes Real-Time Intelligence and lakehouses. Microsoft Learn states that Copilot and AI experiences such as the data agent are not supported on a trial capacity.
What capacity does a Fabric data agent need?
A paid F2 or higher Fabric capacity, or Power BI Premium P1 or higher with Fabric enabled. Your Fabric admin may also need to enable the cross-geo processing and storing settings for AI.
About the author
Dr. Estera Kot is CTO at Clouds on Mars, a Microsoft and Databricks partner for data and AI in Europe and the Middle East. At Microsoft she was Principal Product Manager for the Apache Spark runtime in Azure Synapse Analytics and Microsoft Fabric, and she created the Native Execution Engine. She is a Microsoft MVP and holds a Ph.D. in Computer Science from the Warsaw University of Technology. She presented this series as the community session "From Events to Answers" at Microsoft Qatar HQ in Lusail, Doha, on October 14, 2026.
Sources
Microsoft Learn, checked on November 3, 2026. Session facts come from Dr. Kot's session materials.