Disagreeing With Paul Graham: How Not To Pick A Platform

October 17, 2006

I am one of the many thousands of raving Paul Graham fans out there.  I’ve read most of his content (Paul doesn’t write blog articles, he writes essays).  He is clearly a very gifted writer.  He is also very, very smart (and I rarely use two verys).  But, at least on one point, I humbly submit that he is very wrong.

In the most recent essay, titled “The 18 Mistakes That Kill Startups”, Paul identifies (as you might expect from the title) the common causes of startup failure.

I’d like to focus on point is #17:  Choosing The Wrong Platform

I agree with Paul that picking a wrong platform can indeed sometimes kill a startup, but I’m not yet convinced that this is always the case.  History is replete with startups that picked what were widely considered to be the “wrong” platform and still survived to tell the story (and make a ton of money in the process).  One example would be MySpace and their use of ColdFusion (not that Cold Fusion is a bad platform, but most hacker-types – and particularly those that follow Paul, would likely categorize it as a sub-optimal platform).  There are other examples of startups that succeeded (some modestly, some spectacularly), despite having chosen the “wrong” platform.  One additional example that comes to mind is eBay’s early use of Microsoft’s platform (ISAPI DLL written on top of IIS).

But, this is not my primary point of contention with the article.  Little harm is done by identifying wrong platform selection as a potential mistake that startups should try and avoid (in fact, I think it helps to raise awareness of the importance of this decision).  My issue is with how Paul advises startup founders go about actually picking a platform.

Paul Graham:   “How do you pick the right platforms? The usual way is to hire good programmers and let them choose. But there is a trick you could use if you're not a programmer: visit a top computer science department and see what they use in research projects.” 
I agree with the first half.  A great way to pick a platform (if you’re not a programmer yourself) is to hire great programmers (not just good ones) and let them choose.  But, I don’t think visiting a computer science department and seeing what they use in research projects is an effective strategy.  Here are my issues with this particular approach:
  1. Being a prior computer science student myself, I have a bit of a feel for how platforms get picked for research projects.  Rarely do these coincide with how startups in the real world work.  People in academic research projects are often solving for a very different problem with very different motivations than a startup.  Lots of research projects are a learning exercise.  Most startups are a building exercise.  The desired outcomes are often vastly different.

  1. The platform selection process is sometimes domain and/or user specific.  For example, though Python is a cool language (and I’m sure there are many academics that like it), if you are seeking to build the next big killer desktop application to run on Windows, it will likely prove to be a fatal choice.  The reason is simple.  From a user’s perspective, they expect a Windows application to look and feel like a Windows application.  Chances are, your Python desktop app won’t quite feel “just right” (the user’s dog will bark at it).  This is a case where the users do care about the platform choice because it actually impacts what they experience.  Similar arguments can be made for other target areas like mobile applications.

  1. There may be other dependencies (i.e. integration points) that influence your decision.  As a startup, if you are building an application that will be an extension of an existing application (or consume its services somehow), it often helps to pick a platform that is conducive to that integration.  For example, if you’re building an Outlook plug-in, you probably don’t want to use Ruby for that (even though it might support COM).  

Basically, it seems that Paul thinks that all startups are going after “change the world” strategies and don’t need to concern themselves with user preferences, business domains or the need for integration with existing systems.  Though it would be great if this were true, it’s really not.  

What do you think?  Am I off-base here?  Are all of you writing world-changing software applications that need to use the higher-end languages and platforms from computer science research groups?  Or, are at least a few of you taking a less glamorous (but practical) approach?


Let's Connect

And, you can find me on Google+, Twitter, and Linkedin.