The software successes I see in the App Store take a two-pronged approach: launch with a polished 1.0, but then iterate quickly with similarly well-built .1s, .2s, etc. The upside of releasing multiple updates after a good 1.0 is out the door (on the App Store at least) is that your app shows up more frequently in the App Store Updates area so if people haven't used your app in awhile, they may be incensed to give it another try. A downside of releasing too often is that all the great reviews you got for the 1.0 will be gone once enough reviews for 1.1 are in. If all you did was drop a bugfix into 1.1, you may not receive as many glowing reviews and your overall app rating will drop. This really sucks, but it's a reality of the ecosystem.
I expected this article to just recover some old material, but it actually raises some interesting points.
In particular, working on games, I believe (like with films), I don't think "release early, release often" is a good idea.
It might work with massive multiplayer games, where users tend to keep re-exploring the same content, but in general with games once someone has played a game once they don't tend to want to play it again.
I tried "release early, release often" with an iPhone game I wrote, and it was a mistake. The biggest problem is that users get angry if you take away, or significantly change, content from users. That's reasonable, but it means once you have added a level it is hard to take it away, even if you have a new, better level to replace it with.
Indeed. That's why 'small increments' are key there: as you say, don't significantly change content.
Blizzard have always followed this philosophy with their game balance changes: they make very small changes each patch. They are so careful that often, the effect they are trying to achieve won't be hit until after several patches. See, for example, druid effectiveness in Player-versus-Player: it's steadily, but very slowly, decreased since The Burning Crusade, as a result of druids being (perceived?) too good at surviving.
I mostly kept my Android game's existing content the same. The only changes I made were to slightly decrease difficulty each patch, as some areas were definitely too hard (the first boss could take up to 10 tries to kill, no matter what you did :).
One thing that users really love though is the release of a large chunk of content in one patch - e.g. a brand new area to explore. I don't think this translates well to websites or webapps - no matter how much of a killer 'feature', users can be very cynical or underwhelmed.
Finally, in enterprise software development, you need to be very careful with rapid release cycles as corporations are very dependent on the stability of your product, even if it's internal.
Horses for courses. The bottom line is to use the best strategy for the job.
It worked great for Minecraft: many ideas from the mod community were siphoned back into the main game.
The iPhone does make this difficult, as a release usually takes one week to go through Apple's approval system. Therefore you can't make mistake today and fix it tomorrow.
This is what the Android market is good for. However the stores are different, so lessons learned on the Market may not suit your iPhone audience.
Another week, another false dichotomy article on HN.
If there's an inverse correlation between the speed of deployment and the quality of your software, you're doing it wrong.
The Continuous Delivery philosophy is borrowed from Lean Manufacturing, and a fundamental tenet of that movement is "Build Quality In." In fact Lean is nearly synonymous with TQM, or Total Quality Management. If you've separated the two, then you've missed the entire point.
This tenet is carried over into the software world by Martin Fowler in his book Continuous Delivery, as well as all of the major advocates of lean software processes, for instance David Anderson and the Poppendiecks in their highly influential series of books on Lean Software.
Essential to a successful continuous deployment process is unit testing, continuous integration and built-in testing processes all along the way to feature deployment. These should be focused on quality all along the way: that's the point of continuous deployment, to allow for focused feature development that is highly tested and can be released with higher confidence than the old method of multiple feature release, which are heavy, harder to test and have more integration concerns.
The point of fast releases should never be to allow for "shitty software," but rather to deliver features that users can enjoy as soon as possible, create a process where deployments become highly automated, low-risk events, and gather feedback early and often on the experience of users with finished polished features so that effective product evaluation is happening constantly and the company can adapt quicker to new data and new trends.
A quick perusal of even the first chapter of any of the major volumes on Continuous Delivery should clear this up:
And there's a fantastic presentation by David Anderson on Kanban on InfoQ, which nicely illustrates why quality is one of the major reasons you should consider a rapid release process like Kanban:
I thought the continuous delivery book was written by Jez Humble and David Farley. Fowler provides a foreword for it http://www.amazon.com/gp/product/0321601912?tag=contindelive.... Other than that I totally agree with you comments - Lean Continuous and Agile are not orthogonal to Quality.
You're right, sorry. It's in the "Martin Fowler" Addison Wesley series of books, it's not written by Fowler. I've had the book on my shelf and read a good bit of it, and should have caught that!
"Martin Fowler" is on the cover in two different places, plus it looks just like the cover of Patterns of Enterprise Application Architecture, which he did write, so I've had that misconception in my brain for a while. Failure to test assumptions...
Offhand I can think of a few: ThoughtWorks uses it on most of their projects afaik. Github I believe does as well. They published an interesting blog on their deployment process a couple of months ago.
Where I work at LivingSocial the web application is deployed daily, sometimes dozens of times a day. I've recently taken over the Android project's infrastructure and we're in the process of implementing Continuous Delivery for that app as well.
David Anderson gives examples of implementing it at both Microsoft and Corbis, Bill Gates' other software company.
Some good points, but (there's always a "but" :-) as with most of these, let's call them modern processes (maybe not even that modern if you consider what the likes of Peter DeGrace were saying back in 1990), continuous delivery forces an engineering paradigm onto the (often unwilling) user.
What does the constant barrage of (for example) iTunes or Flash updates do to the user's perception of quality?
What is the impact of the Chrome release cycle on a corporate environment that needs to test each revision on a model office, in case there's an impact on one business-critical application or other?
Yeah this is a good point, and that's a balance I'm actually in the process of determining for my current project.
User expectations are different for mobile apps -- anecdotally it seems users not as annoyed by mobile app updates as they would be for Flash or even Word, because the process is a lot lighter and less painful. For web applications it's transparent to the end user.
But, yes for rich client apps it's a concern. I think the trick is to deliver features rather than bug fixes and shape user expectations towards getting something new from an update rather than getting nothing but fixes that should have been in place to begin with.
Expectations from software are lower than other products like cars, televisions. A car breaking down in the middle of a drive to work is far more "severe" an incident than Internet Explorer hanging and crashing. Further, software engineering as a field is very very young, with a far lower barrier to entry, resulting in products and processes which are much less evolved and mature compared to other industries. I think these are important factors why software tends to have lower quality bars than cars or physical products.
The Release Early/Often example describes a situation of releasing a product that does not have all the features of its competitors: "The Windows Operating System, in the initial version (1.0) did not support overlapping windows which was a breakthrough feature on the then contemporary Macintosh."
Identically, the Polished Release example also describes a situation of releasing a product that does not have all the features of its competitors: The first version of iOS "(then called as iPhone OS), did not even have MMS"
Seems like the author didn't pick a great example to demonstrate a Polished Release.
His example outside of software[1] is not the best either. Movies do have "versions", they are called cuts; an example of multiple published cuts for a movie is Blade Runner.
[1]> Every single release of a movie, whether it is The Jurassic Park and its sequels or The Matrix and its sequels or Steve Jobs’s own Toy Story and its sequels never had "versions".
Releasing well-polished version of products makes sense most of the time. With all due respect to movie directors, engineers, software developers and game designers you _mostly_ know how to do a movie, a car, an MP3 player, or yet another take on video editing program.
Sometimes, however, things are so new that the only way to get to the good result fast (or, to get to the good result at all) you need to release early, release often. Lotus 1-2-3 was such a huge improvement over paper sheets and such a new idea that it totally warranted "release early, release often" mantra.
I think release early, release often works well in some cases, especially for pre-release software. If you're working on the initial versions of something, that quick iterative process is key. Once you've released, however, you have an obligation to your users, so it becomes less of a valid mantra.
http://webcache.googleusercontent.com/search?q=cache:http://...