Teaching Microsoft Fabric from zero in two hours

Five rules for teaching Fabric to beginners

The plan sets five rules for level 0 to 200, L1 to L5, and writes the reason next to each one. Every term gets a picture before a demo uses it. No query language is taught. Each beat holds one idea and ends with one question. Topics that need level 300 context stay in backup files. Limits fit in three sentences.

1. L1. Every term gets a picture and one example before it appears in a demo. Reason: a beginner who misses one word loses the next ten minutes.
2. L2. No query language is taught. Copilot writes the queries, and a query is opened only to show where a number came from. Reason: the room came for the idea, not the syntax.
3. L3. One beat, one idea. Stop after each beat and ask the room one question. Reason: recall keeps people with you for two hours.
4. L4. Cut the comparison of three builds, access control, and the operations agent. Keep them in the files as backup for Q&A. Reason: they need level 300 context.
5. L5. Keep the limits honest but short: three sentences, not a block. Reason: beginners need to know a feature is preview, not the full gap list.

The rules follow the order of a university lecture: define the term, show it, then let the room use it. Dr. Kot brings that order from her role as Assistant Professor, Institute of Computer Science, Kozminski University, Warsaw.

L2 is the rule technical speakers break most often. A query on screen feels like depth, but to a Power BI user at 6 PM it is noise. The script opens one generated query, in D2, for ten seconds.

A run of show for a two-hour Fabric session

The plan splits 120 minutes into eight blocks, B1 to B8. Three blocks of slides in plain words take 40 minutes. A five-minute break follows. The live demo takes 40 minutes, the longest block. The last 35 minutes cover the announcements from FabCon Barcelona, first steps, and questions. On October 14 the final clock moved 17 of those minutes into a hands-on block, B5a.

The presenter script adds three clock rules. If the session is more than three minutes late after B3, the break shrinks to two minutes. B5 is never shortened. Beat D0 starts at 1:03 whatever the state, which handles plan risk P6. The presenter window flags any gap of more than one minute against the plan.

The data was timed to the script too. The simulator sends one tester every 24 seconds. Tester 31 pushes the share past 60 at minute 12 of the demo, the start of beat T3. The Activator rule checks once a minute, so the alert lands in the beat that explains it. Microsoft Learn gives five minutes as the default and lets you lower it. At five minutes the alert would arrive during T4 or D0.

Explain every Fabric term before the demo uses it

The glossary has eleven terms, G1 to G11, from OneLake to Preview. Each term gets one slide, one picture, a line in plain words, and one example from the story. A term must reach its slide before any demo beat says it. Checked against the slide and beat headers of the presenter script, all eleven terms pass. Some words outside the glossary do not.

The best picture carries more than one word. Slide S5 draws Fabric as an office building with OneLake in the basement and a price tag on the door. The spoken line defines the tag: “You rent the building by size. The rental is called a capacity.” It was added after the walkthrough heard the word four times with no definition.

Distance matters too. OneLake is explained at minute 6 and said again at 1:03. The demo map S22 keeps six words on screen for the whole demo to cover that gap. This review of the script and the final deck found four gaps: one word with two meanings, and three uses with no slide first.

1. One word, two meanings. S8 defines preview as a release status. T1 then says “the preview shows the last rows”, meaning the data pane. The walkthrough caught the same collision for “reference” in T2. Say “the data pane” in T1.
2. Agent instructions appear in D1 and in version B of D3. No slide introduces them.
3. Story words arrive with no gloss: maceration in D1 and D4, wear test in D1, and intensity target in D2.
4. Workspace appears on the sign-in slide S21 at 0:40 and in every hands-on step from 1:25. Its gloss, “a project folder”, comes on S25 at 1:47.

Run the check again after every late change. The fourth gap arrived with the hands-on block on the day.

How to write a presenter script for a live demo

The presenter script is the source, and the deck is built from it: “Build the deck to this script, not the other way round.” Each slide entry carries an On screen line, which is the slide spec. Spoken lines are written the way they are said. In the presenter window, each slide shows notes in five blocks, Goal, Say, Do, Ask, and If, followed by a Handoff.

A table called “How to read this file” fixes six markers: a quoted line to say, an italic stage direction, CLICK, Ask, Handoff, and TBC for anything not verified. A checklist of 14 items runs before the first slide. It starts with a capacity resumed 30 minutes early and ends with an alert inbox that makes a sound at T3.

The plan asked for about 27 slides built to the On screen lines. On the day the deck gained a cue card for each of the eleven demo beats: the sentence to type, the question to ask, and the expected answer. Here is one short beat, S20, as the script writes it.

S20 · Sixty, twice · 1 min · 0:38

On screen: two boxes side by side. Left: the Activator rule sentence
with 60 highlighted. Right: the ontology rule sentence with 60
highlighted. One line under both: "Same number. Two places. Tonight."

Say:      Sixty, twice.
          In the first half, sixty is a rule in Activator.
          It sends a message.
          In the second half, sixty is a rule in the ontology.
          It decides what the word winner means.

Handoff:  Five minutes. Water, phones, stretch.
          When you come back, the testers start.

Two more lines promise the join at the end only if the preview allows it. The beat holds one idea, the same number in two places, and plants the question that D5 answers 45 minutes later. Figure 3 shows a longer beat in all six blocks.

The If block is the reason for the format. A recovery line written before the evening is calm. A line invented on stage sounds like an excuse.

Walk the script as a participant before the dry run

Before any dry run, the script was walked on paper as one participant. The stand-in was a Power BI user who had never opened Fabric, with English as a second language. Every spoken line was read in order, and every unknown word, break, and attention drop was noted. The walkthrough found 17 problems: 2 high, 8 medium, and 7 low.

High breaks trust or the demo. Medium loses the beginner for a minute or more. All 17 were applied on October 11, 2026, three days before the session.

• The screen and the speaker disagree. In G1, rated high, the script said tester one chose the test blotter first. The simulator printed the reference blotter. Both now read the real event, a vote for the current formula.
• A word arrives before its slide: “the dictionary” 25 minutes early (G4), “ontology” 8 minutes early (G5), capacity and running share never defined (G6, G7).
• A demo step can fail. In G13, also high, the Copilot prompt asked for a running share from raw events, and Copilot may chart a count. The prompt now names the saved view.
• Attention drops after forty minutes of slides (G9). Three real events now print in a terminal at S9, minute ten.
• Numbers and silence. Nine numbers read aloud became three (G11, G12). Five asks got an if-silence line (G16).

The plan also asks for a timed dry run with one beginner. Dr. Kot's position: do both. Paper finds what the script says. A dry run finds what a person hears.

Live demo risks, cut order, and recovery lines

The plan names seven risks, P1 to P7, each with its handling. The script adds five cuts for a late demo and six recovery lines for live failure. Two beats are never cut: T3, where the alert fires, and D3, where one edit to the ontology changes the answer. D0 starts at 1:03 whatever the state, so the second half always runs.

Together the cuts save 14 minutes. Six recovery lines, R1 to R6, pair a sentence with an action. A wrong number: open the query. A preview that will not load: play the backup recording. A silent stream: ingest events by hand. A late alert: “The rule checks on a schedule. The message will arrive in the recap.”

A recovery line is only as good as its fallback. At 17:00 Doha time on October 14, the build record noted that the fallback dashboard for R5 did not exist yet. On stage, Copilot built the dashboard live, so the gap did not matter that night.

The rule in D5: claim only what worked

D5 runs only if the binding of live events to the ontology worked in the dry run. If it did not, the script gives a different line: “Tonight sixty is the same sentence in two places. Activator and the ontology.” The binding is named as the next step, and as preview. Nothing that failed in the dry run is claimed on stage. On stage in Doha, the fallback line ran.

Microsoft Learn supports that line today. The new ontology experience has no Activator integration, and the ontology does not run its business rules against data. Learn also says a data agent picks up ontology changes every 10 minutes unless you refresh the context by hand. Put that refresh in the Do block of any live edit. And Learn states the data agent does not currently support languages other than English. Dr. Kot's position: present the French question in D1 as a test result, not a feature.

Room questions, preview badges, and three limits

Each beat ends with one question to the room. Five carry an if-silence line, because a formal venue at 6 PM can stay quiet. The deck closes with three questions that each have a one-word answer: Eventhouse, Ontology, and Activator. Every feature on screen carries a GA or Preview badge, and the limits fit in three sentences.

At S16 the question is “Who owns the dictionary in your company today? IT or the business?” The silence line: “In most companies the answer is nobody. That is the problem.”

1. Q1. “Every event lands in one box, sorted by the time it arrived. Name the box.” Eventhouse.
2. Q2. “The business dictionary, where winner gets its meaning. Fabric's word for it.” Ontology.
3. Q3. “The rule that watched the stream and fired at tester 31. Fabric's name for it.” Activator.

Preview badges on every slide where they apply

Slide S8 teaches two badges before any feature appears. Preview: you can try it, it can still change, and you do not promise it to your boss. Generally available: Microsoft supports it and you can build on it. Every feature slide then carries one badge, and S23 repeats them in a table. The statuses below were checked on Microsoft Learn on November 3, 2026.

The data agent itself is generally available. For N2, the Learn how-to page carries no label, while the Copilot pane shows Preview, so it is marked TBC. For each announcement in depth, see the FabCon Europe 2026 series.

Three honest limits in three sentences

S24 in the script, “Three honest limits”, shows three lines. Five documents: the demo shows mechanics, not quality at scale. Prepared passages: a notebook cut the PDF text in advance. Preview: the dictionary can change before your first project ends. Learn backs the second line, since the data agent does not read PDF files directly.

Hands-on design for a beginner Fabric session

The hands-on block has three parts. A sign-in slide stays up during the break. A printed handout carries the sign-in screens, a table of screens you may see, and ten numbered steps. A self-check table lists the expected answer to each question, so a participant can tell a wrong agent from a wrong click.

The sign-in slide, S21, shows numbered accounts from 01 to 50, two workspaces, and three rules: no uploads of your own data, close any trial offer and raise a hand, reload once before asking. The shared password stays on the screen only, because the PDF outlives the evening. Handout steps 1 to 5 open the items. Steps 6 to 10 ask the agent and read the winner rule in the ontology.

The handout says how to read the table: a different answer means the agent is wrong, not the data. That sentence tells a beginner what to do with a surprise.

Two gaps are worth fixing. The slide lists three sign-in screens, the handout four. The build steps use words the session never taught, such as Delta table and KQL queryset.

The setup record notes that 50 distinct accounts signed in. Learn sets the floor for any hands-on with an agent: the trial does not support Copilot or data agents, so you need a paid capacity, F2 or above. Part 6 covers the accounts, capacities, and capacity math for 50 people.

Done when: five criteria you can check

The plan defines done in five criteria. Each one is a fact that someone can check after the evening, not a feeling about how it went.

Dr. Kot's position: write done criteria you can check from the folder, then write the evidence into it on the night itself.

What to do before your next session

Start with the glossary, not the slides. Then write the script in fixed blocks, walk it on paper as one participant, and finish with a dry run in front of one real beginner.

1. List every term the demo says aloud, in demo order.
2. Give each term one slide, one picture, and one example from a single story.
3. Write the script before the deck, in Goal, Say, Do, Ask, If, and Handoff.
4. Walk the script on paper as one participant, and rate each finding.
5. Write the cut order, the beats never cut, and a recovery line per demo step.
6. Run a timed dry run with one beginner, and record every word they missed.

For the whole story, start with Part 1, the overview. For a room of 50, read Part 6. The kit is public on GitHub, and the handout was shared in the Fabric Falcons community. The Microsoft Fabric implementation guide goes deeper on the platform. To plan a beginner session with Dr. Kot, book a slot.

This is the last part of From Events to Answers. Before your next session, pick the one word your room is most likely to miss, and give it a picture.

Key facts at a glance

Session facts come from the session folder. Product facts come from Microsoft Learn, checked on November 3, 2026.

Frequently asked questions

How do you teach Microsoft Fabric to complete beginners?

Explain every term before the demo uses it. Give each term one slide, one picture, and one example from a single story. Teach no query language, and open a generated query only to show where a number came from. End each beat with one question to the room, and keep the limits to three sentences.

What is a good agenda for a two-hour Fabric session for beginners?

One agenda used for a community session has eight blocks: 10 minutes to start from zero, 15 on Real-Time Intelligence, 15 on Fabric IQ, a 5-minute break, a 40-minute live demo, then announcements, first steps, and questions. A hands-on block of 17 minutes can come out of the last 35 minutes.

Should a beginner Fabric session teach KQL or SQL?

No. At level 0 to 200 the room came for the idea, not the syntax. Let Copilot or the data agent write the queries. Open one generated query once, for about ten seconds, to show that a number came from a table.

What should a presenter do when a live Fabric demo fails?

Say a recovery line written in advance, then act. If the agent gives a wrong number, open the generated query. If the stream is silent, ingest one batch of events by hand. If the agent will not load, play the backup recording. Claim on stage only what worked in the dry run.

Can a Fabric hands-on session for beginners run on the free trial?

Only in part. The Fabric trial lasts 60 days, but Microsoft Learn states that it does not support Copilot or data agents. The stream, the Eventhouse, the Activator rule, and the dashboard run on the trial. Copilot dashboard generation and the data agent need a paid capacity, F2 or above.

How do you explain preview features to beginners?

Teach two badges before any feature appears. Preview means you can try it, it can still change, and you do not promise it to your boss. Generally available means Microsoft supports it and you can build on it. Put one badge on every feature slide, and check each status on Microsoft Learn first.

What is a paper walkthrough of a presenter script?

A paper walkthrough reads every spoken line of the script in order, as one named participant would hear it. After each slide it records unknown words, breaks in the thread, drops in attention, and open questions, each rated high, medium, or low. In this session it found 17 problems before any dry run.

Sources

Microsoft Learn, checked on November 3, 2026. Session facts come from the session folder: the plan, the presenter script, the participant walkthrough, the glossary in the architecture file, the abstract, the handout, and the deck.

• Try Microsoft Fabric for free
• Generate Real-Time dashboards with Copilot
• Get started with Real-Time Intelligence
• Activator: query frequency for query data sources
• Activator: ingestion from KQL Querysets
• Create Activator alerts for Microsoft SQL (preview): query as often as every minute
• What is ontology (preview)?
• Create and manage business rules in ontology (preview)
• Use the ontology agent (preview) in Fabric
• Use Ontology as context in Fabric data agent
• Fabric data agent concepts
• What is Fabric IQ?
• Fabric IQ in Microsoft 365 Copilot Chat
• Real-Time Intelligence compared with Azure solutions: operations agent (preview)