Power BI Developer to Fabric Engineer: A Roadmap

The move from Power BI developer to Fabric engineer is usually described as learning a new product. It is closer to changing what your job is: from building an artefact to operating a platform other people build on.

That reframing explains why some very strong report developers stall on the transition, and it determines what you should learn first.

What transfers, honestly#

The people who move fastest are usually the ones who already thought in star schemas. The people who struggle are usually the ones whose strength was visual design.

Stage 1 — SQL, properly (3–4 weeks)#

Not "I can write a SELECT". Window functions, CTEs, query plans, and the ability to look at a slow query and form a hypothesis about why.

This is non-negotiable because the whole platform layer is expressed in SQL and Spark. Power Query lets you avoid learning it; Fabric does not.

the level to reachsql
-- Running totals, ranking within groups, and gap detection
-- should feel routine, not clever.
SELECT
    customer_id,
    order_date,
    amount,
    SUM(amount) OVER (
        PARTITION BY customer_id
        ORDER BY order_date
        ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW
    ) AS running_total,
    LAG(order_date) OVER (
        PARTITION BY customer_id ORDER BY order_date
    ) AS previous_order_date
FROM fct_orders

Stage 2 — Data engineering concepts (4–6 weeks)#

Idempotency, incremental processing, slowly changing dimensions, orchestration, testing, lineage.

Learn these on something free before touching Fabric. Building a small local stack — DuckDB, dbt and an orchestrator — teaches every one of these concepts and removes the platform as a confounding variable. When you then open Fabric, you are learning one product rather than a product and a discipline simultaneously.

Stage 3 — Fabric specifics (4 weeks)#

Now the product. Lakehouse versus warehouse, medallion layering, Direct Lake and its fallback behaviour, deployment pipelines, capacity and throttling.

Fabric is much easier to learn once you have opinions about how a data platform should be shaped, because most of its choices are then recognisable rather than arbitrary.

Stage 4 — The part people skip (ongoing)#

Platform engineering is substantially about enabling other people. Documentation, naming conventions, a governed metric layer, onboarding, and saying no to requests that would compromise the model.

A realistic timeline#

Three to four months of consistent effort alongside a full-time job gets you to competent. The people who take longer usually skipped Stage 1, and the SQL gap surfaces later disguised as a Fabric problem.

What to build while learning#

One end-to-end project beats six tutorials. Ingest a real public dataset, layer it properly, build a semantic model on top, and write up the decisions you made and why. That write-up is worth more in an interview than a certification, because it demonstrates the thing the role is actually about: judgement under trade-offs.