Automatic synthesis of electronics systems may be viewed as a decision making process with lots of decision variables. Huge design spaces and exponential complexity algorithms need to be handled pragmatically with the aid of fast but not very accurate heutistics. We propose a new architecture for design automation tools which allows better design space exploration using estimations as guiding rather than bounding heuristics. The automated design process is explicitly represented as a sequence of design decisions which may be reviewed, deferring ultimate decisions to a later time when enough information is available.