LinkedIn·Tuesday, 18 August 2026·8d ago

Scalability and reproducibility aren't enough, it's time to add the principle of replaceability to your data stack. Early this year, we…

Martijn Bauters

Founder @ Agoya | CDO @ RVA-ONEM | President @ Rhinos Rugby Oudenaarde

Scalability and reproducibility aren't enough, it's time to add the principle of replaceability to your data stack. Early this year, we were hit at the NEO - National Employment Office with Airbyte, one of our vendors, sending a formal notice: we're stopping our on-prem support next summer, you migrate to our cloud solution or you are no customer of ours anymore. For us, on-prem data processing isn't optional, it's where we pseudonymize and anonymize our data before anything can go to the cloud. Non-negotiable. Luckily within our data platform we've set up an exit strategy for each of the components anticipating potential price structure changes, consolidations or changes in the regulations. So currently every component is replaceable with a second-best option. So when we got notified, we activated our exit strategy. No crisis, just taking out the plan and executing it. In the end we moved to dltHub, an open-source, Python-native tool that fit better with where the team's skills were heading anyway. With the consolidation wave going through the industry, Dagster being bought by Prefect, dbt by Fivetran, the big platforms eyeing the next-best thing in the market, you better understand that the tools you pick today will change under you. The only thing you control is whether you designed your platform for that.