The problem is that the layers in data lake design map to technical components. This doesn't work - change is orthogonal.
Data engineers are siloed and stuck between teams with different incentives and priorities. 
In a nutshell
We need to slice the data problem around business domains
Pipelines still exist, but they are an implementation detail of those domain data products.
Think of domain data as a product, and create a great product experience. But for that, they need to be discoverable, accessible and trustworthy. APIs have SLOs, and data needs the same.
The KPI for the product is the time to discover and consume the data for the domain. Can we decrease it?
Shared concerns become shared, self-serve infra. Things like setting access controls - right now they are different for every system.
We need common way for data products to talk to each other. Like APIs talk in HTTP. Enable interoperability and discoverability.
New ownership model
What it looks like in reality
Data products hide the pipelines and provide ports.
