Before you adopt an AI tool, check how you would leave it

Most AI features get evaluated on what they do on day one. The harder question is what you can take with you if you ever need to switch.

A recent BytePreview guide on data portability draws a useful line between three things: the data you put in, the outputs the tool generates, and whatever belongs to the provider or depends on its infrastructure. An "Export" button doesn't tell you which of those you get back, or whether the result is usable anywhere else.

Points worth turning into a checklist:

• Read the exclusions, not just the promise. Microsoft's documentation, for example, lists exportable customer data (text, audio, image and video files, database content) but leaves out resource configuration details and protected proprietary data.
• Retention depends on the data type. In Amazon Location Service, device positions are kept for 30 days and then deleted automatically, while geofence geometries you upload must be deleted explicitly.
• An API is not an exit plan. Check which operations and data it really exposes. Being able to call a feature doesn't give you its underlying model.
• Format, content and procedure are separate checks. Does the export keep the fields and relationships you need, and can you repeat it at volume?

The guide suggests a simple matrix (export, format, retention, deletion, interfaces), writing down open questions instead of filling gaps with assumptions, and saving the version and date of every document you review. Anything critical that isn't documented belongs in the contract.

Full guide on BytePreview: https://www.bytepreview.com/en/articulos/data-portability-in-ai-software-a-guide-to-assessing-lock-in-risk?utm_source=dev&utm_medium=social&utm_campaign=bytepreview