You're looking at Less Wrong's discussion board. This includes all posts, including those that haven't been promoted to the front page yet. For more information, see About Less Wrong.

Daniel_Burfoot comments on Open Thread, Apr. 27 - May 3, 2015 - Less Wrong Discussion

3 Post author: Gondolinian 27 April 2015 12:18AM

You are viewing a comment permalink. View the original post to see all comments and the full post content.

Comments (352)

You are viewing a single comment's thread. Show more comments above.

Comment author: Daniel_Burfoot 27 April 2015 02:48:09PM *  6 points [-]

Disclaimer: this thought is "foxy", in the sense that I don't assert it's definitively true, but I still think it could be a useful lens for viewing the world.

Startups Don't Create New Technology

Contra gurus like Paul Graham and Peter Thiel, successful tech startup companies do not actually create new technology. Good tech startups do one of two things: 1) invent a new technology-dependent business model, or 2) repackage and polish existing technology in such a way as to bring it above the threshold for widespread use.

Consider a couple of recent successful tech startups: Facebook, Twitter, Uber, AirBNB, and Dropbox. None of these can be said to have innovated deeply new technology. Instead, they realized that they could create a new business model based entirely on available technology.

Uber is a particularly illustrative example. The company depends enormously on several powerful new recent technologies: smart phones, GPS, and mapping software. However, Uber itself did not innovate any of those. If one of those technologies hadn't been available, Uber probably would not have been successful. Uber certainly could not have created any of those technologies as part of its business plan.

I'm not suggesting here, of course, that tech companies in general do not create new technology. The point is that startups don't create technology. Instead, deeply new technology is primarily developed by large, established companies. The basic pattern for technology creation is:

  • Invent a new business model that depends on currently available technology (startup phase)
  • Grow the business fast based on profits from new business model (growth phase)
  • Using newly-available resources of finance and talent resulting from initial success, develop deeply new technology (mature phase)

The history of Amazon illustrates this pattern very well. Amazon started by creating a new business model using currently available web technology. It depended on a huge array of technology that was developed by others - web browsers, web servers, databases, the internet, personal computers - but it did not develop any of that technology itself and would not have been successful if it had had tried to do so (imagine trying to innovate the web browser so you could sell books online).

While Amazon did not create new technology in its startup phase, it certainly has created deeply new technology now that it is in its mature phase. The clearest example of deeply new technology created by Amazon is cloud computing (some people might also point to eBooks). Cloud computing could never have been innovated by a startup company - the resources required in terms of finance, talent, and corporate resilience are far too great. While cloud computing could never have been innovated by a startup, it is now becoming a foundational technology for the new generation of startups.

So the lifecycle of entrepreneurial technology development suggests a kind of virtuous circle. A company becomes profitable by building a new technology-dependent business model or repackaging technology developed by others. Then it grows, and when it reaches a certain point, it becomes able to create new technology that feeds the next generation of startups.

Comment author: Daniel_Burfoot 29 April 2015 02:49:03AM *  9 points [-]

To add a bit of empirical analysis to this comment, I analyzed the YCombinator Winter 2015 batch. I categorized the startups into one of three buckets: Tech-Dependent Business Model (TDBM), RePackaging and Polishing existing tech (RPP), and Novel Tech (NT). The list can be found here.

  • CampusJob - TDBM
  • Seed - TDBM
  • NextTravel - TDBM
  • TheMidGame - TDBM
  • eBrandValue - TDBM
  • Standard Cyborg - RPP, maybe NT
  • Rescue Forensics - TDBM (social entrepreneurship)
  • Lumi - TDBM/RPP
  • Undeground Cellar - TDBM
  • Transcriptic - NT?? but not easily evaluated
  • Atomwise - maybe NT but probably RPP/TDBM of machine learning (I doubt they created new ML algos)
  • Spark Gift - TDBM
  • Gradberry - TDBM
  • Industrial Microbes - NT?? but probably TDBM/RPP of existing chemical engineering tech
  • TechList - TDBM
  • Meadow - TDBM
  • ReSchedule - TDBM
  • Diassess - RPP, synthesis of biotech and infotech
  • RazorPay - TDBM
  • DirectMatch - TDBM
  • BuildScience - TDBM/RPP
  • ShiftLabs - RPP, making medical devices cheaper.
  • Valor Water Analytics - TDBM
  • Instavest - TDBM
  • Open Listings - TDBM
  • CloudMedx - TDBM/RPP
  • BankJoy - TDBM
  • TransitMix - TDBM
  • ZenFlow - NT, in biotech space.
  • Final - TDBM
  • Lully - maybe NT but probably RPP
  • Spire - TDBM/RPP
  • AnalyticsMD - TDBM
  • Smarking - TDBM
  • 20N - NT
  • GrubMarket - TDBM
  • CribSpot - TDBM
  • KickPay - TDBM
  • Notable Labs - uncategorizable but brilliant, some kind of legal/biotech/infotech combination play
  • Pretty Instant - TDBM
  • VetPronto - TDBM
  • Akido - TDBM
  • DroneBase - TDBM
  • MashGin - maybe NT in computer vision space but probably RPP/TDBM
  • LabDoor - RPP/TDBM
  • Bonfire - TDBM
  • EquipmentShare - TDBM

The following pattern emerged from this exercise: YC is not funding startups that are developing new computer science technology, with the possible exception of MashGin and AtomWise. The YC startups that are attempting to develop new technology are in the biotech/medtech space - Transcriptic, Standard Cyborg, Industrial Microbes, Zenflow, Lully, and 20N.

Edit I noticed after writing that the list is from Demo Day 2, representing the second half of the Winter 2015 batch. However, it doesn't appear to me that analyzing only half the batch causes a serious bias in the conclusion. The Demo Day 1 batch is available here.