The continuous improvement J-curve reflects Forrester's Law, which identifies a reliable trend to initially resist structural changes required to seek breakthroughs and open the path to unlimited improvement. This resistance initially depresses overall measurable performance, which provides an excuse to shut down any innovative improvements. It's because most organizations fall victim to this that advantage remains for competitors.
 Phases and variants
Very mature engineering proceses hit asymptotic limits after which improvement may not appear to improve measurable performance much but have intangible benefits and so tend to pay off even more financially - see six sigma.
Before this limit is reached, the process may hit one or more design plateaus reflecting the need to deeply integrate the work product with others using standards (so as to remain modular, reusable and recycleable). These plateaus appear in performance metrics, resulting in a phased S-curve similar to that of punctuated equilibrium. A biomimicry view is that this is similar to the equilibrium reached by natural selection as organisms find niches in their natural ecosystems. Only when the ecosystem changes is there opportunity for much creative destruction and thus engineering (or evolutionary) innovation.
Very early phase engineering processes will encounter the continous improvement J-curve phase. Entering the exponential part of this reflects a change of mindset which transforms views of problems, as predicted by the twelve levers management method.
 Personal computer example
A well known example of all the above is the shifts from 1980 to present in personal computer technology. The mainframe mindset under which all information required very strong central control only yielded with great reluctance over the 1980s - as late as 1989 it was possible to say no one ever got fired for buying IBM. After that, it wasn't.
Only after this initial resistance did PCs spread rapidly through all organizations in the 1990s, encountering several design plateaus: desktops, towers and laptops. Innovation focused on these and did not for instance lead to successful mainstream tablets, wearable computers, home theaters until roughly 2006, symbolized by the release of Windows Vista, iPhone and SED monitors respectively. These represented new design plateaus. Meanwhile, wearable theater and biofeedback gear emerged as the likely next plateaus before the very personal computer merged with digital FM solar headbands probably before 2020 - depending on Chinese electronics price-performance curve.
Already by early 2007, however, the power, weight, shock and user interface characteristics of the very personal computer were becoming relatively clear, e.g. the Twiddler one-handed keyboard, the Micro-Optical SV6 visor, and shoulder sling holder for a 1GHz, 30GB device, drawing about 7 watts of power, as worn by Thad Sterner of Georgia Tech, suggested some parameters for an asymptotic limit of improvement of wearable devices.