Two names, one shortlist, one vendor
For most of the last seven years, choosing a data orchestrator in Europe meant running the same bake-off. Airflow was the incumbent everybody was trying to leave. Prefect and Dagster were the two credible ways out, and they were genuinely different: Dagster made you think in assets, the data you are trying to produce, and Prefect made you think in execution, the work you are trying to run. Teams argued about it in procurement meetings and in engineering channels, and the argument was useful, because it forced a decision about what you actually cared about.
On 13 July 2026 that argument ended, not because one side won, but because Prefect bought Dagster Labs. Terms were not disclosed. Dagster's creator Nick Schrock and Pete Hunt, who built its commercial business, are part of the deal, and many of the Dagster team are joining Prefect. Jeremiah Lowin, Prefect's founder and chief executive, framed the result as a single platform with three layers: Dagster defines the outcomes, Prefect executes the work, and FastMCP governs access.
That third layer is the one most buyers have not priced in.
What was actually bought
The public story is an orchestration merger. The more interesting story is that Prefect already owned FastMCP, the access layer that sits between AI agents and the tools they are allowed to call. Fold Dagster into that and one company now holds three positions in the same stack: what your data is supposed to be, how the work that makes it gets run, and which of your systems an autonomous agent is permitted to reach.
Those three things were separate purchases eighteen months ago. Vendors consolidate for exactly this reason, and it is not a criticism to say so. A single platform across assets, execution and access is genuinely easier to operate than three integrations, and buyers reward that, which is why the deal makes sense for Prefect.
It is a criticism to say that the buyer's negotiating position is worse afterwards and that almost nobody adjusts for it. When your orchestrator, your asset layer and your agent gateway come from one supplier, the cost of leaving stops being the cost of replacing a tool and becomes the cost of replacing a stack. That number is not on the invoice, but it is the number that sets your price at renewal.
"Nothing changes" has a shelf life
Prefect's announcement is explicit and, as far as it goes, credible: Dagster and Dagster+ remain unchanged, existing and new customers can continue exactly as before, no action is required, and both products get long-term maintenance and investment.
Take that at face value, then read what it does not say. It does not say pricing is held. It does not say the two products will be maintained as independent alternatives rather than converged into one. It does not commit to a date beyond which support is guaranteed. Every acquirer of a competing product says nothing changes, and every acquirer means it on the day they say it. What ends the arrangement is not bad faith, it is arithmetic: the moment maintaining two overlapping engines costs more than migrating the smaller user base onto the larger one, the roadmap decides for everyone.
The honest version of this deal for a Dagster+ customer is therefore not "nothing changes" and not "you are about to be forced off." It is that you have a window, probably measured in one or two renewal cycles, during which you have more leverage than you will have later.
Use the window
Three things are worth doing in the next quarter, and none of them require you to distrust the vendor.
Read the Dagster+ contract for the clauses that only matter in this scenario: what happens on end of life, how much notice you get, whether pricing is capped at renewal, and whether you have any right to migration assistance. If those clauses are thin, the time to strengthen them is now, at a renewal where the acquirer wants to demonstrate that the acquisition was safe for customers. That argument is available to you for a limited period.
Second, confirm what your self-hosting position actually is. Both Dagster and Prefect have real open-source cores, and for a European team with sovereignty or data residency constraints, the ability to run the orchestrator yourself is the escape hatch that makes the commercial relationship voluntary. Verify it works rather than assuming it does, because an escape hatch you have never opened is a hope, not a plan.
Third, keep the access layer separate in your architecture even if you buy it from the same supplier. MCP is a protocol, not a product, and the value of agents being able to reach your tools through a standard interface is that the interface can be re-implemented. If FastMCP becomes the only thing that knows how your agents authenticate to your systems, that is a coupling you chose, not one the protocol required.
The pattern behind the deal
This is the second consolidation in a month where the thing being bought was not the product but the position. A carrier bought the connectivity glue between clouds. Now an orchestration vendor has bought the other orchestrator and holds the agent access layer as well. The AI stack is compressing from a set of components you assemble into a set of platforms you adopt, and the compression is happening faster than most procurement cycles can register.
For an owner the implication is not to avoid platforms. Platforms are usually the right buy. It is to notice, when a supplier acquires the thing you would have switched to, that your leverage moved and to spend it before it decays. The best moment to negotiate the terms of your exit is when the vendor is still trying to convince you that you will never need one.
Read next: The Email Layer Built to Dodge Lock-In Got Bought | The EU Product Passport Drives Its First Takeover
