A computer only turned up on my desk in 1997. For my career's previous 11 years most calculations were done with no analysis aid other than a calculator. As a result, you constantly had to make choices. What checks were important? Shall we assume it is one-way spanning (much easier)? Where exactly is this beam bearing?

And most of my early years were spent designing steel frames. A kit of parts fabricated off site and assembled bit by bit in (mainly) simply supported spans. The concrete on metal decking was assumed to span simply as well. All easily understood and separately calculable. The engineer was in charge. But slowly we have been given more and more analytical power, and we can build vast and complicated models with everything in them, that do all we ever dreamed of. And often everything is connected to everything else, and, not surprisingly forces, go everywhere in ways we don't understand - but we feel we need to reinforce for them. And perhaps for insitu reinforced concrete that's not far from the truth.

I've been reminded recently, several times, that it is important that we remember how things are put together. A simple example: A precast walkway spans between two precast beams and has a precast 'twinwall' panel running beneath its third edge - the wall of a tank. All elements will be joined together by an insitu stitch, so why not model it as a continuous concrete structure? It is supported on three edges surely! Why not save that bit of rebar?

But the reality and economics of Design for Manufacture and Assembly means that the site team need to be able to land the precast walkway quickly and reliably. The beams provide easy bearing points, but the face panel of the twinwall is narrow and not necessarily at the same level as the beams. So, for ease of construction the right answer is to design the walkway as simply supported between the beams, using the stitch to the wall only to help stiffen the wall panel horizontally in the final state. If you were calculating by hand, I think you'd do this naturally, thinking about how the elements go together. Instead we got a 'wished in place' continuous analysis, and the first walkways on site were a pain to erect (we sorted out the later ones).

For small items, poured or milled as a single piece, a continuous finite element model may be the truth (or close to the truth). However, our buildings and infrastructure are all of a scale where they take time to build, have intermediate stages of stress, shrink, expand, and are increasingly made up of assemblies of offsite manufactured components.