More about Why Symbolics Failed
I just came across “Symbolics, Inc: A failure of heterogeneous engineering” by Alvin Graylin, Kari Anne Hoir Kjolaas, Jonathan Loflin, and Jimmie D. Walker III (it doesn’t say with whom they are affiliated, and there is no date), at http://www.sts.tu-harburg.de/~r.f.moeller/symbolics-info/Symbolics.pdf
This is an excellent paper, and if you are interested in what happened to Symbolics, it’s a must-read.
The paper’s thesis is based on a concept called “heterogeneous engineering”, but it’s hard to see what they mean by that other than “running a company well”. They have fancy ways of saying that you can’t just do technology, you have to do marketing and sales and finance and so on, which is rather obvious. They are quite right about the wide diversity of feelings about the long-term vision of Symbolics, and I should have mentioned that in my essay as being one of the biggest problems with Symbolics. The random directions of R&D, often not co-ordinated with the rest of the company, are well-described here (they had good sources, including lots of characteristically, harshly honest email from Dave Moon). The separation between the software part of the company in Cambridge, MA and the hardware part of the company in Woodland Hills (later Chatsworth) CA was also a real problem. They say “Once funds were available, Symbolics was spending money like a lottery winner with new-found riches” and that’s absolutely correct. Feature creep was indeed extremely rampant. The paper also has financial figures for Symbolics, which are quite interesting and revealing, showing a steady rise through 1986, followed by falling revenues and negative earnings from 1987 to 1989.
Here are some points I dispute. They say “During the years of growth Symbolics had been searching for a CEO”, leading up to the hiring of Brian Sear. I am pretty sure that only happened when the trouble started. I disagree with the statement by Brian Sear that we didn’t take care of our current customers; we really did work hard at that, and I think that’s one of the reasons so many former Symbolics customers are so nostalgic. I don’t think Russell is right that “many of the Symbolics machines were purchased by researchers funded through the Star Wars program”, a point which they repeat many times. However, many were funded through DARPA, and if you just substitute that for all the claims about “Star Wars”, then what they say is right. The claim that “the proliferation of LISP machines may have exceeded the proliferation of LISP programmers” is hyperbole. It’s not true that nobody thought about a broader market than the researchers; rather, we intended to sell to value-added resellers (VAR’s) and original equipment manufacturers (OEM’s). The phrase “VARs and OEMs” was practically a mantra. Unfortunately, we only managed to do it once (ICAD). While they are right that Sun machines “could be used for many other applications”, the interesting point is the reason for that: why did Sun’s have many applications available? The rise of Unix as a portable platform, which was a new concept at the time, had a lot to do with it, as well as Sun’s prices. They don’t consider why Apollo failed.
There’s plenty more. To the authors, wherever you are: thank you very much!
December 21st, 2007 at 5:37 pm
This paper appears to be a final paper for course 6.933J/STS.420J, The Structure of Engineering Revolutions, at MIT in 1998.
December 21st, 2007 at 11:58 pm
[...] More about Why Symbolics Failed I just came across “Symbolics, Inc: A failure of heterogeneous engineering” by Alvin Graylin, Kari Anne […] [...]
December 31st, 2007 at 5:58 pm
Although I agree with the disconnect between projects undertaken and their relevance/business analysis of market demand, I do have to take exception to some of the points of historical note.
I notice a lack in the mention of the Graphics division. That would have taken away some of the argument of a lack of focus on non-research products. The Graphics product line was definitely a deliverable to a non-researchy, very practically-oriented market.
Also, note that the Sunstone project did address many of the competitive concerns, especially the continual mention of Sun in this analysis. The Sunstone project included a chip design for a platform meant to run Unix and C, as well as Lisp. It was a safe C exploiting the tagged architecture, for example, to allow checking of array bounds. And the Sunstone project was being produced on-time. But to back up the analysis of Symbolics’ priorities, it was cancelled as we were getting the first chips back from LSI Logic.
An interesting historical note, I took my first ASIC class with the first Sun chip designers, at the time when Sun was just as small. That was the class that helped us produce the G machine chip set. In fact, because of our software expertise, we had an edge at the time in producing ASIC hardware faster than our competitors.
Internally, we continually found ways like this efficient ASIC design process, that showed how the Lisp machine could be used to produce practical, market-demanded products.
I actually don’t think lack of a single goal was as much of a problem as matching efforts in creating products to what was needed in the market place. Under the appropriate manager, most of the goals suggested as separate at the top of the list in this paper could be organized as a coherent, yet diversified, approach to address the value added by our technology. That technology was fueled initially by a Lisp machine, but it meant so much more than just a machine or even the Lisp language.
January 16th, 2008 at 6:10 pm
[...] story short, the game changed and Lisp fell from prominence. A number of factors torpedoed the Lisp machine companies, the DoD pulled funding for AI research, competing languages improved and incorporated [...]
February 6th, 2008 at 5:53 pm
I have read, financial story, bad story, with bad management.
The lispmachine are and were a heritage of national awareness to protect their unique characteristics. I have read more story of success lisp implimentation and now read power application on CM5, more of this run with SW emulation of lisp software on Aplha for Military application. Some is AI implimentation for Expert system and management theater. But the question is: Why Not Coprocessor Lisp Engine, Why Not continuos support of old client with new implimentation of Coprocessor Board like Ivory for Mac.
More of research around on Lispmachine is real startup for company on 90′year.
This indicates that the lisp language training is a platform absolutely essential in the training of designers. More language 4G is made prototype with lisp.
Now is market is ready for lisp and AI application without limit of management of information.
This was confirmed by the continued success of the company that compilers lisp invest in the stability of the product and support to the end user on different platforms.
For customers for long-term investments need stability of the product and evolving look apple slowly, but continuously evolving. Lisp and burned lispmachine have the advantage because they were not ready hardware platforms. Today can compete with the parallelism of the language that other languages you dream, I can compete on the ability to manipulate information of different types and large volumes may compete with the calculation and management of distributed resources. The lisp has gone through 50 years of history with its adaptability remaining intact is winning the challenges of the time.
September 11th, 2008 at 9:52 am
Reminds me of Tandem Computers Inc or Atari for that matter.
A lot of these early pioneers developed purpose built technologies
for a specific market perhaps not realizing the full marketing
potential of their technologies. This allowed the competition to
supersede in an era where ‘networking’ and multitasking’ was the
new buzz. Maybe ‘multimarketing’ could of saved a lot of these
companies from bankruptcy!
September 11th, 2008 at 10:55 am
Tandem didn’t actually go bankrupt: it was bought by HP, and as far as I know the technology is still being sold and maintained.
Atari did develop special hardware, but that’s how their field went: the Sony Playstation and Microsoft X-box and such are just as much special hardware and purpose-built devices.
December 18th, 2008 at 5:27 pm
On Tandem:
It’s not just being sold and maintained, but actively developed. HP/Intel included lockstep instructions in Itanium, and Tandem products are getting ported there.
(For a little while HP adopted the “Nonstop” name for all their business-critical servers, which is why my book on web servers for OpenVMS refers to “the non-stop webserver” in the title. They seem to have dropped that. But in any case, in the big redo that went with the Compaq acquisition, they decided that in the data-center range, OpenVMS, HP/UX, Linux, and NonStopOS (Tandem) were strategic, and that MPE, Precision Architecture, Tru64, and Alpha, weren’t, and started working to get everything strategic onto Itanium as well as to make Itanium acceptably functional.)
April 16th, 2009 at 2:15 pm
Thanks for an interesting blog item, Dan:
I came to Symbolics in the beginning of 1986 and worked in the International Sales in Cambridge (and in Concord some time).
The International Dept. was not in the main stream, perhaps it should have been. Thus, I do not know everything what went on in the Engineering.
One thing was for sure, money was spent in lavish sales meetings, etc.
I was fortunate to qualify to the Australian Sales Award trip and that was tops. My wife was eternally grateful, but I could have been motivated with spending less money.
Summa summarum, I really enjoyed my time at Symbolics and enjoy finding Symbolians now in social networks.
all the best,
September 30th, 2009 at 11:55 am
Dan,
This was a very interesting post. I think your analysis of the failure of Symbolics is the closest to being correct; that is, it is largely consistent with my view.
Something about me… I was working in computer vision at an industrial research lab when I was hired in February 1982 by the marketing organization of Symbolics. I was employee #32. I came to Symbolics because I’d have access to a lisp machine (“who could pass up that opportunity”) and I expected to get rich while there. To me, getting rich was that after 3 years I could walk away with $300,000, which was enough to buy a decent house in a nice neighborhood in LA. I left in November 1982 to do research in knowledge-based systems and programming environments at Xerox and ISI afterwards, where I still got to use lisp machines (Xerox then Symbolics). So why did I leave Symbolics after 10 months? I left because it was obvious to me that I would not get rich at Symbolics. I expected Symbolics to go belly-up after 5 years. Instead, it took 10 years simply because the software environment was so good – still better than what you can get today. How sad.
So what did I see that caused me to leave?
(1) About a month after I began working at Symbolics I went to a conference to help market Symbolics machines. The night before, I went to a large hotel room to meet the “east coast” folks. (I was one of two technical guys on the west coast.) Just after entering the room with my supervisor, he and one of the “east coast” guys started yelling at each other. The mere sight of seeing each other caused this display. And before and after this I witnessed what I considered to be excessive email flames that were distributed to all employees of the entire company. So I asked myself how can a company function (and succeed) with these kinds of problems? Good management would have had talked to these people and said something like “no matter how valuable you are to the success of this company, you are not so valuable that we can tolerate this behavior.” That was never done. BAD MANAGEMENT!
(2) The company was clearly run by the researchers, who were not profit driven. This allowed all kinds of things to happen. For example, Symbolics produced a laser printer based on a Canon print engine. Why? How could we possibly compete with Canon, Xerox, and others when our printer was essentially the same as their printers? Similarly, we delivered LM-2s without the instruction pre-fetch unit, which was supposed to make the machine run 50% faster. This was because the person that was assigned to do this was busy building 16-bit digital-to-analog converters when only 8-bit digital-to-analog converters were available. (At least this is what I was told.) This was done so that Symbolics could embed high-quality sound within its keyboard. How many machines would this feature sell? If we were lucky, maybe this feature caused the sale of 3 machines. Again, BAD MANAGEMENT that was not MARKET-DRIVEN.
(3) The machine was marketed as a Lisp machine. However, it was more than that. As you mentioned it supported C, Fortran, and Ada. When I was attached to marketing I can’t tell you how often I said we should be marketing to those communities too. There were a lot more C programmers than Lisp programmers. For whatever reason, all this great technology for non-Lispers was hidden. Again, BAD MARKETING. Also, what you did not mention was that Symbolics had built a VLSI design tool that they used to design the 3600 chips. Why wasn’t this sold? It should have been. Again, BAD MARKETING — there was a big and growing need for tools like this (and the mney was available for expensive VLSI design stations).
(4) When I was on the front line, trying to sell Lispms, we had a major problem. I’d try to sell Lispms to friends and former associates that cost a minimum of $90K, and generally $105K out-the-door. They were typically given $120K to buy computers. This left them with a choice of purchasing 1 Symbolics machine or 3 Xerox Dolphins. They generally selected the 3 Dolphins. Our cost for the keyboard was $3,200 and our cost for the monitor screen was $8K (at least that was what I was told). My recollection was that the manufacturing cost of a Symbolics machine was over $40K. Hence, about one-third of the cost of a machine was the keyboard, mouse, and display. I told people within Symbolics to consider using a dumb display as the front-end (something like a Datapoint terminal). I can remember one employee telling me something like “if people can’t appreciate why the machine is the way it is then they’re stupid.” Unfortunately, the potential buyers desired (drooled) to have them but few could afford them. Again, BAD MARKETING that didn’t align the products that Symbolics made with those of what the market would bare.
(5) Like I said earlier, the researchers led the company, and it was obvious that a problem we’d be facing in 2-3 years was the impending explosion in Intel-based PCs. In 1980, I bought a Radio Shack TRS-80 for about $1K than ran Emacs and Lisp (written by Jonathan Allen) incredibly fast, although without a fancy GUI. Again, BAD MANAGEMENT/MARKETING.
In sum, Symbolics developed incredible technology in 1982 (and a lot more thereafter) that is still better than what is available today. It failed because marketing did not identify the products that we should be building and selling (i.e., market-driven requirements) and management did not direct its personnel to build those products. If the problems that I mentioned are obvious to me – who was a reasonably smart tech weenie that wanted to make a few bucks it should have been obvious to marketing and management folks. (It should be noted that the Symbolics sales representative for Northern California left Symbolics to become VP of Marketing for Sun Microsystems in the summer of 1982, after about a half year of employment at Symbolics. So I assume he saw the handwriting on the wall too.) I wonder what would have happened if Sun’s management had been run Symbolics. I expect something much different.