
Yesterday, we rolled out free accounts to LayerVault along with the concept of forking. The release had been several months in the making. But unlike most big launches, the pieces of this release were exposed over time. If anything, it wasn’t so much the big launch but one of the last launches in this thread of thought. After building LayerVault for nearly two years, we have realized that the best way to build a product is to release things piecemeal and to avoid the Big Launch.
Launches carry substantial risks. The perceived rewards from a launch always skew rationality in the most toxic way possible. The thinking is this: “The last time we had a nice bump was when we did a launch. Therefore, we should do an even bigger launch.” Related to this line of thinking is the “People won’t care unless we launch something.” I’ve been there before. We all have.
This type of planning fails to capitalize on the incremental successes of building a great product. Launches seem beneficial because they help you cut through the noise. You’ve got a better chance of getting written about if you grant embargoes and negotiate exclusives. We all want to be the company that launches a great feature to fanfare. But we are basically drunk on launches at this point, dangerously close to dying from launch poisoning.
But launches in general are a poor representation of how great software today is built. It’s a holdover from the days of boxed software, where supply chains had to be managed and masters golded. The same that is true then is still true now: great software is the result of continuous refinement. The only thing different today are the release schedules.
Software today is developed on a continuum. The discrete measure of software progress is a commit. But commits are trivial, and the process of delivering (deploying) those commits is only slightly more trivial. Thus, the time deltas between updates and their subsequent deliveries are approaching 0. Launches are going the way of the press release and the dodo.
GitHub is the most exciting name in software today. The project started humbly in 20081 and had its first TechCrunch post July 22, 2008.2 The next TechCrunch post about GitHub (and not merely mentioning them) was on July 24, 2010, when they reached one million hosted projects.3 To put that into perspective, Digg had 86 posts written about it on TechCrunch4 in the same time period. In fact, TechCrunch even wrote about Digg on July 22, 2008, in the post “Google in Final Negotiations to Acquire Digg for ‘Around $200 Million’,”5 the same day GitHub quietly launched Gists.
I use TechCrunch here as a short-hand for a “big launch” or a launch-scale event. When a young company wants press, the goal is usually a TechCrunch post. It is called the “TechCrunch of Initiation,” after all.
Without a single TechCrunch mention, GitHub sustained mind-boggling growth. That’s what solving a real problem looks like with a viral growth component built in. I still spend half of my day in GitHub, and the other half in LayerVault. They doubled in size twice,6 with nary launch. They did spend plenty of time doing healthy promotion of themselves and their ideas on their own blog.789
In July 2012, GitHub was valued north of $400 million, with cash flow to support at least 108 employees (deduced from pre-funding levels indicated in the “How to Build a GitHub” deck).10 That same month, Digg’s assets were sold to Betaworks for $500k.
While we reached out to the Verge to do a piece yesterday about our announcement of free accounts and forking, we did not concentrate on tracking down press for this launch. If anything, yesterday’s story was merely the final leg on a journey that we embarked upon nine months ago with a blog post, “Early Sketches from the Web’s Finest.” The post exploded and we learned something: Designers not only love seeing the finished products, but also seeing the process. This also coincided with a conversation we had with Kyle Neath where he pointed out, “There’s no way for me to show how awesome I am using LayerVault.” We set the goal of free accounts and worked backwards.
The next nine months we refined exactly what free accounts meant to LayerVault. We removed things11 and improved12 things13. We also slowly rolled out the necessities for what we deemed to be a complete free launch. Originally, a free launch had to include the following:
Bundling all of these together would have been dangerous. It’s possible that we could get the implementation of one of these wrong and that would hurt the whole release. And some things just don’t make sense to hold back, like incremental performance improvements. It was a good exercise in untangling, but we were able to separate all of these into separate releases. But didn’t we announce three of those yesterday? We did, but they had been done for weeks beforehand. (Also, Snape kills Dumbledore and Bruce Willis was dead the whole time.) We took a page from GitHub’s book and released all of this “dark,” sometimes months before we would announce it. (Releasing a product or feature “dark” is the process of shipping feature code to production but not exposing it.) Eliminating launches requires intense focus in two areas: shipping completed features dark and continuous deployment.
As of last week, we put the last building block in place to allow us to have seamless deployment across all of our products. The LayerVault application will now update silently after 15 minutes of inactivity. We did this by forking the Sparkle framework and improving upon it. You can grab that code in what we call AutoSparkle. We think this is the most aggressive form of updating desktop software the world has seen yet. Previously, the most aggressive was the “restart-to-update” seen in applications like Spotify and Google Chrome. There are things in the app that are dark, i.e. not visible to the user, that may not get turned on for a few months.
We are also content with artificially slowing growth while we tighten all of the screws. Just before 5 p.m. EST yesterday, the 2,000 free accounts we carved out for the day were used up. Each day, we release another 2,000 accounts at 12 p.m. EST. We are doing this as a hedge against any sort of overloading the system. It also has the intended side effect of making sure that most interested people are granted accounts first. The launch of free accounts was never intended to be an intense flash-in-the-pan, but a slow burn.
The next time you find yourself working for a company that’s laser-focused on a singular launch, it’s time to leave. The impending big launch communicates ineptitude, an inability to correctly assess the intricacies of a problem.
If you’re at the helm of a young company preparing for a launch, don’t. Roll out what you have today to the appropriate users. Get them to love what you’re building. A few developer-centric companies that are doing this amazingly well are Travis CI, CircleCI, and imgix. Hell, you can’t even signup for imgix. Resist the glitz and the glam of the press day. It’s short-lived and likely unproductive. It messes up your pipeline and provides only synthetic success. Focus on meaningful interactions with your users through channels you control.
Looking forward to showing you what we’ve got in store for next week.
—Kelly
http://en.wikipedia.org/wiki/GitHub “GitHub" ↩︎
http://techcrunch.com/2008/07/22/github-unites-version-control-with-the-pastie/ "GitHub Unites Version Control with Pastie" ↩︎
http://techcrunch.com/2010/07/24/github-one-million/ "GitHub Hits One Million Hosted Projects" ↩︎
http://techcrunch.com/tag/digg/page/2/ "TechCrunch: Digg" ↩︎
http://techcrunch.com/2008/07/22/google-in-final-negotiations-to-acquire-digg-for-around-200-million/ "Google in Final Negotiations to Acquire Digg for Around 200 Million" ↩︎
http://zachholman.com/talk/how-to-build-a-github/ "How to Build a GitHub" ↩︎
https://github.com/blog/13-hackers-agree-github-rocks "Hackers Agree: GitHub Rocks" ↩︎
https://github.com/blog/24-the-future-of-coding "The Future of Coding" ↩︎
https://github.com/blog/40-we-launched "We Launched" ↩︎
http://www.forbes.com/sites/tomiogeron/2012/07/09/cash-for-code-github-raises-100-million-from-andreessen-horowitz/ "Cash for Code: GitHub Raises 100 Million From Andreesen-Horowitz" ↩︎
http://layervault.tumblr.com/post/42295579315/simplifying-the-file-page "Simplifying the File Page" ↩︎
http://layervault.tumblr.com/post/42361566927/progressive-reduction "Progressive Reduction" ↩︎
http://layervault.tumblr.com/post/47793430523/a-better-faster-layervault "A Better, Faster LayerVault" ↩︎
The Launch is dead! What are you thoughts? I kind of agree, to a certain degree……
I bet launches go back even further. Launches are similar to how movies are still released today. You have a huge...
roccondil liked this Fantastic article!