He designed the distribution model of the Oz programming language
in 1996 together with Seif Haridi and Per Brand.
This is arguably the first system to show that network transparency
can be practical.
This result was also contrary to popular wisdom at the time.
The model was subsequently implemented in the
Mozart Programming System
and first released publicly in 1999.
It provides network transparency with efficient network communication protocols
and fault detection within the language.
The design was made possible due to the structure of the Oz language,
which cleanly separates stateless, single assignment, and stateful
entities, and does not use state by default.
An important part of this work is
the distributed protocol for
single assignment variables,
which does distributed rational tree unification.
The algorithm was published in 1999 with a correctness proof (ACM TOPLAS).
It is the first published algorithm for this operation
that is both correct and complete.