Recently, increasing attention has been given to the problem of how to produce goal-directed behaviour in the real world. The approaches taken vary along a spectrum from trying to extend the scope of classical plan-ning techniques in systems such as Cassandra and BURIDAN via including execution in plan construction systems such as Sage (Knoblock 1995); through develop-ing systems such as RAPS that execute ready-made plans (Firby 1989; 1996); to the design of systems that compile plans into sets of reaction rules. It seems reasonable to suppose that eventually much of this work will converge some-where near the centre and will result in systems that seamlessly integrate plan construction with plan execution, thus allowing planning, re-planning and acting to be interleaved. Indeed, this is the theme of much recent research as evidenced by many of the papers included in these Working Notes. There is, of course, disagreement about whether this eventual aim is best approached by starting from the plan construction end and adding in more flexibility, or by starting from the plan execution end and adding in more deliberation. In this brief paper I describe an example of the latter approach and consider how much flexible goai-directed behaviour can be achieved with what appear to be minimal deliberative capabilities.