What I learned at Digital Techniques, Inc. Burlington, MA 1984-1989
From 1984 to 1988, I was a co-op student at DTI of Burlington, Mass., whilst I attended Northeastern University. This was an excellent little company that made software and hardware well ahead of the industry, but which suffered from a marketplace that changed faster than anticipated and not in the way that was envisioned. This resulted in a bankruptcy and reorganization. The company's excellent technical legacy was carried through the Touchcom line which was acquired by today's G4S in 2008 (2012 press release).
The below commentary is my personal perspective, a quarter century later.
Back to 1984
The Touchcom system had an over-engineered chassis, suitable for practically military grade. This might have been an asset at Kappy's installations, but everywhere else it pushed expense and limited flexibility. Industrial-strength CPU and controller cards were used with a custom-built graphics display card (GDC) and other custom components. The use of the industrial STD bus, while possibly an asset in the early 1980s, became an albatross even by 1985.
Software was based on industry-standard MS-DOS but not commodity IBM PC compatible, slowing development and also limiting increased capability. Low-level software was wisely (for the time) separated into an assembly-language coded "TCOS" but coding in a higher level language like C would have offered 90%+ of the speed and much better expansion potential at faster/lower cost. TCOS was poorly documented even to internal systems coders, greatly diminishing the utility of the approach. Many high-level applications reworked TCOS functionality in far slower scripting code.
When opportunity to redesign the hardware came, instead of using standard components and building just a custom GDC, which could have been an opportunity to expand market for DTI's advanced graphics ... instead, a single, massive, all-in-one CPU/I-O-controller/video card was built as an expansion card (!) for the IBM PC system bus. This then necessitated an even more ununusal chassis. The amount of money and time spent on this curious exercise, which did little to make the system more functional or marketable, astonishes me even today.
To be sure, by the time I graduated Northeastern in 1988 and left DTI as a contractor in 1989, all these shortcomings were in the process of being rectified, but the window of opportunity had been lost.
In both hardware and software, follow the UNIX, and LEGO, philosophy of building modules. Build and make money on the smallest possible unique element where you add value ("As small as possible, but no smaller" -- [after] Einstein). Leverage other people's investment wherever you can, by following and building onto standards, and by creative re-use of common components.