Not that I’m skeptical or nuthin’ …
The number being bandied about, and cited here last week, is that Amazon releases new software to production every 11.6 seconds.
It appears Amazon employs roughly 3,000 web developers, which means it’s arithmetic time. Let’s see: 86,400 seconds per day, 11.6 seconds per release, and that gets us to … carry the one … about 2.5 releases per day per web developer.
Does someone in earshot know enough about the situation to clarify exactly what each programmer is releasing 2.5 of to the wild every day at Amazon? Because never mind DevOps, and for that matter never mind some awfully fast coding. With developers this fast, who comes up with enough ideas to keep them busy, and how?
While the exact number seems suspicious enough to warrant skepticism, let’s give everyone in the DevOps public relations supply chain enough benefit of the doubt to accept that whether the actual number is 7,500, 750, or 75, Amazon releases a lot of new code to production every day.
When the new releases are going to a website or mobile app, as they are with Amazon, nobody has to worry about training or organizational change management. Amazon’s shoppers expect to figure out what they’re supposed to do next based on what they see on their computer or smartphone screens. They’re shopping, not engaged in mission-critical activities.
And what you see on Amazon, while it’s quite a bit more complicated than Google’s legendarily spare user interface, is still quite simple compared to what business users see when looking at a screen from one of the company’s enterprise applications.
Also: While there is, as has been pointed out in this space ad infinitum, not to mention ad nauseum, no such thing as an IT project (they’re all about business change or what’s the point? In case you haven’t been paying attention) when it comes to enhancing the company’s website, or mobile app, or, for that matter, software products like operating systems and office suites, for all intents and purposes the business change is done when the software hits the production servers.
That’s in contrast to any change you’d be likely to make to the company’s ERP, CRM, supply chain systems, or what-have-you. The whole point of changing or enhancing the company’s internal systems is to support a change in how we conduct business around here.
DevOps for internal applications is, that is, qualitatively different from DevOps for customer-facing applications.
Which brings up another thought: When it comes to customer-facing systems, faster is very often better. Even if it isn’t truly “time to market” in the sense of being able to sell “new and improved” before your competitors can, when interacting with Real Paying Customers, novelty has intrinsic value.
But when it comes to employee-facing applications, slip-streaming application changes mostly means confusing employees, disrupting an established business process with no clearly defined process change in view.
Internal releases aren’t just software changes except for those that do nothing but fix bugs or modify algorithms hidden from the view of the employees who see their results.
Internal releases are business change releases. Otherwise they’re just releasable builds — dots to be connected when the actual release is ready for deployment.
Releases that are business changes call for communication, retraining, sometimes changes in the metrics the company uses to gauge its effectiveness, and occasionally something even more extreme, like changes in reporting relationships.
There are situations when changes to business processes can be just as iterative and incremental as Agile makes software changes and enhancements.
But just as often, and especially when regulatory and compliance issues are involved, bundling business changes into controlled releases really is the better choice.
Business change like this isn’t necessarily better for being faster.
DevOps still might have a place here, but its place is more likely to be releasing a continuous stream of small changes to a sandbox environment than true continuous integration and delivery.
There’s little value in IT dramatically outpacing the natural pace of change in one or another part of the business. Unless, that is, the natural pace of business change is slower than it ought to be because everyone expects IT to be the bottleneck, and adjusts their velocity expectations accordingly.
Blowing up those expectations? It can do wonders for a business culture steeped in the perverse pleasures of slow.
My guess is that the Amazon figure just comes from their continuous “balancing” of prices. Which I would think is entirely automated. So yes there is a lot of site changes but they are not pushing features or bug fixes.
You might be right. Or, Amazon might have just had a lot of feature changes happen to stack up in that one month.
ManagementSpeak:Reduce your stress by submitting a new ManagementSpeak. It won’t reduce your workload, but it can be a lovely and subtle form of revenge for it.
Translation: It may not reduce your workload, but it will reduce mine.
🙂
I think this is really a question of how one defines “release”. Amazon does a lot of automated updates, testing out icons, layouts, colors, etc. I suspect these statistics are counting each of those things as a release. For example, when they came up with their one click purchasing, they tested many colors of the icon to see which one got the most clicks, and then set the color to be the most clicked option. Each tested color may have been counted as a release.
Notice now that Amazon’s buttons are now orange or gray (with shading), with the button they want you to click in orange, and other options in gray.
I posted the Amazon question to a local developer group, and got this response:
“That’s sometimes a matter of the scale of services they have. I worked there on a team in Seattle and we were one of the most continuously deployed in the company. We’d have maybe one production deployment every two days — sometimes less frequently when issues are found or if we enter high risk time periods.
“But there are hundreds, maybe thousands of services internally that each have their own cadence. I’d be interested to see the average of each service’s average cadence.”