I’ve been in the software industry for 30 years.  In many ways that makes me an industry dinosaur… and those all died out. So I guess you’ll have to figure out if I am wise or shortly going extinct.  The ”I” word that I wanted to write about is not the I in AI, and not the I in Innovative but instead is about : Iteration.  In fact I’d posit that most innovation is the result of constructively applied consistently improving iteration.  Most people encounter innovations at the break through moments- when they amazing innovation appears with  fantastic impressive results - but those innovations are most often the result of countless iterations. 

My observation is that one common element in the recipe for failure in building software (not the only one) - is when teams fail to iterate and instead go for the Big Bang, the brand new shiny complicated solution - instead of the iterative release.  The alternative is to build the solution that iteratively layers on prior learnings. What is the smallest change necessary to release the next viable feature to create value and to be in production.  From my earliest days at WebLogic - we built software daily, tested nightly and we shipped often.   (The speed of releases in those days of Enterprise software was glacial compared to today)


When I or my teams have attempted or I have seen other teams attempt the big leap - that requires years of effort and is released in one big push without iterative releases without built in feedback - they have largely failed. 


Looking back at big milestone innovations - we often miss seeing the iteration behind them.  Ascribing their existence from ex nihilo. Lets look at a few examples :


The lightbulb : Edison and his team conducted thousands of tests on many different materials.  One that focused on bamboo filaments, involved 2,774 attempts. Edison famously said : “I have not failed 10,000 times. I have successfully found 10,000 ways that will not work.”


The Internet : burst onto the scene in April 21, 1993 with the release of the first World Wide Web browser : Mosaic; followed by Netscape Navigator in October of 1994.  Which unleashed a .com Internet E-commerce revolution.  (Which is still on going).  But this all had its roots going all the way back to 1969 when the ARPANET network first sent a message between UCLA and the Stanford; and then lots of iterative programs and products : TCP/IP in 1983, the concepulation of the WWW with HTTP and hypertext linking in 1989 before actualization in 1993 of a browser.  


The iPhone : amazing in form and function when released in June of 2007. However this brought together many iterative innovations into a single form. Built upon the release of the iPod in 2001 that handled storage and audio.  And the first Personal Digital Assistant (PDA) the Apple Newton that was released in 1993 and at the iPhone release was dominated by the BlackBerry.  Combined with cellular technology, chip and radio transmitter minutarization as well as cellular network expansion, the iPhone stood on the shoulders of many giant predecessors. 


While here we may as well pause to talk about Apple Vision Pro : did it fail because of price point, because of features, because of infrastructure? Maybe all of them, probably worthy of a much more focused deep dive -but an interesting juxtaposition to the success of the iPhone.


And lastly the most recent innovation : AI - Artificial Intelligence.  ChatGPT in 2022 : seemed like a innovation that came out of no-where.  However ChatGPT has foundations that depend on all the machine learning developments, on the development of neural networks and large data processing that identify patterns and “learn” as they are exposed to more data.  And other developments in natural language processing applied to large data sets : Large Language models.  These LLMs have billions of parameters and become the backers for chat bots that in response to questions (prompts) can generate statistically predicted text results.  They are artificial because they were made by man - and labeled intelligent because their output appears/reads/sounds very intelligent. 


And each of these innovation products came from many many iterations and as you look at their adoption and impact they in turn also relied on broad infrastructure contexts within which to operate.  The lightbulb was only as useful and impactful as the capacity to generate electricity - and the electrical infrastructure (wires) to connect to people's homes.  The Internet was as only as connected and as useful as the power of the computers connected, and the infrastructure connecting them - which ranged initially from physical wires, to utilizing the ubiquitous phone lines and now use high speed of fiber optic lines or the wireless infrastructure that the entire cellular network relies on from the first brick phones of the 1980s to the smart phones of today that started with the iPhone.  And finally AI is reliant on massive compute capabilities - which is why Nvidia stock has skyrocketed because their GPU chips offer parallel processing power that is required by computational intensive tasks required for training and running LLMs.  


So when my teams come to me with bright shiny new ideas - I always try to ask - how can we release iteratively. Including how can we get feedback on designs and conceptual prototypes even before we actually write any software.  How can we break down our full North Star product vision into phases - release software early and often and keep iterating.  


Here is a small example of something that I worked on at Amazon that is now comparatively very very big.  Fulfillment By Amazon has shipped 80,000,000,000 units since 2006.  When we launched FBA in April of 2006, here are all the things that FBA Sellers COULD NOT do :


  • Sellers could sign up for FBA but it was not an automated registration system.  The registration workflow kicked off a series of 25+ tickets that teams had to manually configure to enable a seller for FBA. 

  • Sellers couldn’t see any FBA orders in Seller Central - in either a UI form or in a report. 

  • Sellers couldn’t request removal of your inventory from FBA - except by calling or emailing our operations specialist : Chad Goelzer

  • Sellers couldn’t see in Seller Central - the quantity or state of their FBA inventory 

  • Sellers did not automatically get paid if Amazon lost or damaged their inventory

  • Our systems did not have the ability to charge any FBA fees - we had a public rate card - but at launch we did not charge (nor could we have) any fees


But what Sellers could do was create an inbound shipment, have that inventory received at an Amazon warehouse. Sellers could create offers that were designated as FBA, and when inventory was received those offers because buyable on the website, customers could buy those offers on website and Amazon would pick, pack and ship it. We prioritized launching the key features that enabled us to test with sellers and learn. And then we prioritized releasing feature after feature to expand function and entitlement - something that the FBA team is still doing today. And all of that iterative innovation of FBA itself relied upon the innovation that Amazon had undertaken to build incredibly sophisticated operations and fulfillment capabilities.  FBA would not exist if Amazon had not first built a great fulfillment network to open up to Sellers. In 2005 we launched with 5 sellers.  In 2006 - we ruthlessly prioritized the critical features to launch and got adoption, got feedback and kept going. In 2024 - there were over 55,000 independent sellers.


If you want to innovate - you have to iterate.