n8n vs Airflow: a DAG orchestrator, a canvas, and the open-source system for open-ended work

What Airflow is built to run

Airflow's own documentation calls it a platform for orchestrating batch workflows, and describes it as an open-source platform for developing, scheduling and monitoring workflows. You define the work in Python. A DAG holds tasks, the dependencies between them and a schedule; the framework parses the script, schedules the tasks, and runs them in the order you set. Operators connect a run to Google Cloud, AWS, Azure and other services.

The same docs are direct about who it fits. Workflows with a clear start and end that run on a schedule are a good fit for Airflow DAGs, and the framework is built for people who would rather code than click. People who would rather click than code will find it a poor match. It runs anywhere from a single process on a laptop to a distributed system, and a web UI shows DAG status, task logs and run history, so you can backfill old runs or rerun only the tasks that failed. Airflow's own site announces stateful tasks and multi-language support for its current release (checked October 2026).

When a DAG orchestrator is the right tool

A DAG is the right shape when the steps are known before the run and the same path should repeat:

• A nightly load that lands warehouse tables in order.
• A chain of dbt or Spark jobs with dependencies between them.
• An ETL job that moves records between two systems on a schedule.
• A model-training pipeline that reruns when its inputs change.
• A scheduled report that pulls from several sources and writes one file.

These jobs need ordering, retries, backfills and a record of what ran, which is what Airflow's scheduling and execution semantics exist for. Expressing the pipeline in Python keeps it in version control, testable, and reviewable as code.

When the job is open-ended

The mismatch appears when you cannot write the task list before the run. Research a market and write a one-page brief; read a bug report, reproduce the bug and open a fix; reconcile a month of records where each row needs a judgment. A DAG is a graph the parser builds before the run executes, so the route is fixed before any data is seen. Work like this needs a machine to run tools on, and a plan the agent writes as it goes. That is a different layer from scheduling.

What Kortix runs instead

Kortix is the open-source AI Operating System: your agents, their skills, your company memory and every connector live in one git repo you own (Read the docs). Each session runs one agent on its own cloud computer, one isolated sandbox per session, on its own branch. The agent plans, calls connectors as tools, and when the work is done opens a change request; session work reaches main through a change request a person reviews as a diff. Any model provider with your own keys. Connectors reach 3,000+ apps in a click, plus MCP, OpenAPI, Postman, GraphQL and raw HTTP, and connector credentials are brokered server-side and never enter the machine. When a job should run on a schedule, a cron trigger or a signed webhook starts a session with nobody present.

The two run side by side

A team rarely swaps one for the other. Keep Airflow on the pipelines whose dependencies and schedule are fixed, and give Kortix the work whose route is decided while it runs. The same split applies to the canvas tools: the n8n vs Kortix comparison sets a visual automation tool against Kortix, is n8n obsolete covers when a fixed flow still fits, and the n8n alternatives guide sorts the wider field by the work each tool assumes.

Where to start

Take the one job your current setup handles worst, the one a person still does by hand because the steps change each time, and run it in Kortix. kortix sessions new starts the first session, and the agent opens a change request when the work is done.