Getting the Airflow 3 DAG Authoring Certification

A few years ago I took the Astronomer DAG Authoring certification for Airflow 2. Airflow has moved on a lot since then, so in November 2025 I sat the updated Airflow 3 DAG Authoring exam.

Here is what changed, and a money-saving tip if you want to take it yourself.

What’s different in Airflow 3

If you are coming from Airflow 2, most of your DAG-authoring muscle memory still works — the TaskFlow API, operators and dependencies are all there. But there are a handful of changes worth knowing before the exam:

• Datasets are now Assets. Data-aware scheduling got a rename and an upgrade. You now think in terms of Assets (and the @asset decorator), and you can schedule DAGs on combinations of assets much more expressively.
• DAG versioning. Airflow 3 keeps track of DAG versions, so the UI and the metadata can show you which version of the code a given run actually executed. No more guessing whether a run used the old or the new logic.
• catchup defaults to False. A small change that bites people who upgrade: new DAGs no longer backfill by default. If you relied on the old behaviour you now have to opt in explicitly.
• logical_date can be None. For asset-triggered and manually triggered runs there isn’t always a meaningful logical date, so code that assumed execution_date/logical_date was always present needs a second look. Several legacy context variables were removed too.
• A new, faster UI. The web interface was rebuilt (React-based) and is noticeably snappier, with the new versioning and asset views baked in.
• The Task Execution API / Task SDK. Under the hood, task execution is decoupled from the scheduler. You mostly don’t author against this directly, but it’s the foundation for remote execution and a cleaner separation between your code and the Airflow internals.
• SubDAGs are gone. They were deprecated for ages; in Airflow 3 they’re removed. Use TaskGroups (or assets) instead.

The exam reflects these: expect questions on assets/data-aware scheduling and on the new defaults, not just on the classic operator/dependency mechanics.

Tip: hunt for coupon codes in Astronomer’s webinars

The exam isn’t free, but you very often don’t have to pay full price for it. Astronomer runs regular free webinars and Academy sessions, and they frequently hand out coupon / voucher codes — sometimes for a discount, sometimes for a completely free exam attempt.

So before you book:

• Sign up for a couple of Astronomer webinars or Academy live sessions and actually attend them.
• Watch the chat and the closing slides — the codes tend to show up at the end.
• Keep an eye on their events page and newsletter for “certification week” type promotions.

I got my voucher this way, and it’s the same trick that worked for the Google Cloud certs: the training providers would rather give you the exam cheaply (or free) than have you not take it at all. I used: start-2026-like-a-pro

Worth it?

Yes — especially if your team is moving from Airflow 2 to 3. Going through the material forced me to actually learn the asset model and the new defaults instead of discovering them the hard way in production. And with a coupon code, the only real cost is the time you spend studying.