The downside is, when all you have are Excel sheets, everything looks like rows and columns (and not e.g. objects with behaviors). If Excel had more robust import/export mechanisms that normal users could understand (e.g. built-in REST client with JSON + XML serializers + many database adapters w/ lots of helpful wizards or tools to guide you), it'd be way more powerful. Then again, if someone is at the point where they'd be able to look at some JSON and compare it with their spreadsheet and be able to describe the mappings, they're possibly better off going to some training sessions on ${your favorite programming language} to learn how to do this the easy way.
Ideally, an organization comes to understand that Excel is a fantastic tool at the frontier where the business needs to adapt rapidly, but once a process is fixed, replacing it with a fixed system is worth the tradeoff in reduced operational risk.
To your second para., much of that falls to internal apps to provide decent RESTful APIs across their systems. Some companies are doing this, in the process getting to a point where the Excel frontier is just analyzing and reporting on data, not acting as a source in its own right. Then you have traceability for every data point in the organization, and you're in a pretty sweet spot operationally.
It seems more like a non-functional declarative vm to me.
Any specialist user willing to invest some time in learning their tools can do this. A culture develops around it.
And replying to parent: those efforts around teaching 'civilians' to code are probably misguided. The investment needs to be in adding scripting and programmability into existing line of business tool, not on encouraging people to sit in front of an isolated REPL disconnected from any business value or context.