RudderStack occupies a position that is genuinely interesting even if the teaching around it is only adequate. It competes directly with Segment, it takes the opposite position on two questions that matter, and understanding why is more educational than most of the individual articles in its learning centre. The first question is where your customer data lives. The conventional CDP ingests your events into its own store, builds profiles there, and syncs outward.
That means a full copy of your customer data sits in a vendor's infrastructure, with its own access controls, its own retention behaviour, and its own obligations when a deletion request arrives. RudderStack's warehouse-native position is that your warehouse is already the system of record, already governed, already subject to whatever controls your compliance function has established, and the CDP should read from it and write to it rather than becoming a second home for the same data. Anyone who has had to answer a data subject access request across two systems that disagreed about what they held will find this argument lands harder than it reads. The second question is whether you can run it yourself.
The RudderStack server is AGPLv3 and the SDKs and integrations are MIT, which means self-hosting is genuinely available rather than nominally available. For an organisation with data residency requirements, or one operating in a jurisdiction where sending customer events to a third party is a regulatory conversation, that is not a nice feature, it is the deciding one. The learning centre content is a mix. The best material is the general data engineering writing that is not really about the product at all, covering event collection design, warehouse modelling, identity stitching and the practical differences between streaming and batch activation.
That content stands on its own and will still be useful to someone who chooses a different vendor. The weaker material is closer to positioning, comparing approaches in ways that unsurprisingly favour the approach being sold. This unevenness is the normal condition of a company blog pressed into service as a curriculum, and it is the main structural criticism to make: this is a library, not a course. There is no defined path, nothing checks whether you understood anything, there is no certification, and a learner who does not already know what they are looking for will wander.
Compare it with Segment University, which has an actual curriculum and a free credential with a human review at the end, and the gap in pedagogical investment is obvious. The AGPL point deserves care because it is the sort of thing that derails a project late. AGPLv3 is a strong copyleft licence with a network provision, and while for most companies running RudderStack as internal infrastructure it presents no practical difficulty, a substantial number of organisations maintain blanket policies against copyleft code in their estate, and some have genuine exposure depending on how they deploy and what they build around it. The documentation states the licensing accurately.
It does not warn you with the prominence that a reader whose company has such a policy needs. Establish this before you build a plan, not after. The self-hosting cost is the other thing to weigh honestly, and the same criticism applies here that applies to Matomo. Running a customer data pipeline yourself means you now own an ingestion service that must not drop events, a transformation layer, a set of destination integrations that break when third-party APIs change, and the monitoring to know when any of it is failing.
The failure mode is quiet: events stop reaching a destination, nobody notices for a fortnight, and a downstream team has been making decisions on incomplete data the whole time. That is a real operational burden and the material does not price it. Self-hosting is the right answer when the requirement is genuinely regulatory. It is frequently the wrong answer when the motivation is cost, because the staff time exceeds the licence saving at most organisations.
Open core adds the usual complication, where the licence boundary and the practical feature boundary are not the same line, and working out which capabilities you will actually need on which side of it requires more care than the marketing invites. Three point six. A worthwhile free library with a genuinely differentiated architecture behind it, held back by being a collection rather than a curriculum, by uneven article quality, and by underplaying both the licence question and the operational reality of running it yourself.