<p>Increasingly software systems are required to
<br />survive variations in their execution environment
<br />without or with only little human intervention. Such
<br />systems are called &quot;eternal software systems&quot;. In
<br />contrast to the traditional view of development and
<br />execution as separate cycles, these modern software
<br />systems should not present such a separation.
<br />Research in MDE has been primarily concerned with
<br />the use of models during the first cycle or development
<br />(i.e. during the design, implementation, and
<br />deployment) and has shown excellent results. In this
<br />paper the author argues that an eternal software
<br />system must have a first-class representation of itself
<br />available to enable change. These runtime
<br />representations (or runtime models) will depend on the
<br />kind of dynamic changes that we want to make
<br />available during execution or on the kind of analysis
<br />we want the system to support. Hence, different models
<br />can be conceived. Self-representation inevitably
<br />implies the use of reflection. In this paper the author
<br />briefly summarizes research that supports the use of
<br />runtime models, and points out different issues and
<br />research questions.</p>