As part of this year’s Engineering Leadership Summit: Virtual Edition, we spoke to Edith Harbaugh, CEO and Co-Founder of LaunchDarkly. She discussed Progressive Delivery and the importance of Cycle Time. Below is an excerpt from the fireside chat portion of Edith’s session, edited for length and clarity.
Hillary Nussbaum, Content Marketing Manager, Code Climate: Welcome to the closing session of this year’s Engineering Leadership Summit, featuring a conversation with Edith Harbaugh, CEO and Co-Founder of LaunchDarkly. Edith and I will be talking about continuous delivery, specifically talking a little bit about what she wrote in the foreword of our upcoming book, The Engineering Leader’s Guide to Cycle Time. Let’s jump right in. Edith, do you mind introducing yourself? Tell us a little bit about what you do and how you got to where you are today.
Edith Harbaugh, CEO and Co-Founder of LaunchDarkly: Hey, I’m Edith Harbaugh, I’m CEO and Co-Founder of LaunchDarkly, a feature management platform. We have customers all over the world, like Atlassian, BMW, who rely on us to get the right features to the right customer at the right time. And we do this at massive scale — we serve trillions, trillions with a T, of features every day.
That’s great. Tell me a little bit about what software development and delivery were like when you first started, and how things got to where they are today, and then where LaunchDarkly fits into all that.
So I started back in the ’90s, and I always loved software because it was a way to bring stuff to life, it was a way to create something from scratch and get it out to the world. The way software was built back in the ’90s was very different than it is now, and that was because you had to physically ship software. So you would have a team of engineers at Microsoft, for example, who would work for three years, and they would release it to manufacturing and print all the bits on a bunch of disks, and then ship them out. And you would actually go to the store and you would buy a box of software, and then you would sit at your computer and you would feed all those disks into your computer, and that would create Microsoft Excel or Microsoft Word. This was pretty cool because then you could use Microsoft Excel, which I still love.
It also created a lot of habits around software — when it went to manufacturing, it had to be perfect. Somebody’s going to buy a 25-disk box of Microsoft Word and then install it on their computer, and it better be good because they’re not going to do that every day. They might not even update it every year. You might have service packs where somebody would go and get an update, but most people are fine if it works okay.
Microsoft had a very small bug in Excel where if you did some esoteric computation, they were off by some fraction, and that was actually a huge deal. So because software was so hard to get, there was a lot of pressure all through the release train to make it perfect. This has to be right, we only get one chance. We only get one chance to get this out to our customers and this has to be absolutely right, we don’t get a second chance.
How exactly do you feel that impacted things moving forward? Even as people started to change their delivery model, do you still see holdovers from that today?
The massive sweeping change that happened — and it’s still happening, to be honest — is from packaged software to software as a service in the cloud. I’m looking at my computer right now, it doesn’t even have a slot for disks, it just doesn’t work that way anymore. You get your software from the cloud, virtually. And what this has really changed is that it’s okay to move a little bit faster, because you can change it. The whole idea that we’re going to have Windows 3.7 Service Pack 6 is just gone now, now you can get a release whenever you want.
And how that’s changed software is it’s really freeing and liberating in a weird sort of way. Think of the poor engineers of the ’90s, when you only had a three year window to work and then ship. It got very political, it got very hard, because you’re like, “I have to make a guess about what people will want three years from now. If I’m wrong, I won’t have a chance to fix it.” So there’s just a lot of polishing and then an inability to adjust on the fly. If you were a year and a half into your release cycle and you realized you were completely wrong, it was just really hard to scrap that. So the biggest change has been with Agile and software in the cloud, that you have a lot more freedom. You could do a week long sprint and say, “Hey, let’s get this out to market. Let’s test this, let’s do this.”
And where do Continuous Delivery and Progressive Delivery fit into all that? Agile is a step past waterfall, but I would say that Continuous Delivery takes that even a step further. How do you see that all fitting together?
So Continuous Delivery was a step past agile in that it was this idea that instead of doing a release every year or every six months, that you could just release whenever you wanted. Jez Humble and David Farley really popularized this theory with a book called Continuous Delivery, where it’s like, okay, let’s make it so you can release any time you want. This was extremely popular and extremely scary at the same time. With big sites like Facebook, it just gave them this extreme competitive advantage. If you can ship 10 times an hour, or 50 times an hour, or 100 times an hour, and your competitors are shipping once a quarter, of course you’re going to have better software. You can fine tune things, you can dial them in, you can get them right. The downside to Continuous Delivery, and why we came up with Progressive Delivery, is that it’s terrifying. What I just said about how you can ship 10 times, 50 times an hour — if you’re a big bank or an insurance company, that’s absolutely horrifying. I’ve got customers who are depending on me, and I don’t want that. Or if you’re an airline, for example, you have regulations, you’ve got people in the field, you cannot have people just check in 50 times an hour and cross your fingers and hope that everything works out.
Even Facebook has moved away. Their mantra used to be, “move fast and break things,” now it’s “move fast with stable infrastructure.” How they got there was Progressive Delivery, which is basically having a framework which decouples the deploy, which is pushing out code, from the release, which is who actually gets it. And with Progressive Delivery, what you have is you have the ability to say, “Okay, I have this new amazing feature for looking at your bank balance, which makes it much clearer where the money is going.” And instead of pushing it out to everybody and crossing your fingers and being like, “Please let people not have negative balances,” let me deploy this code safely, and then turn it on, maybe, for my own internal QA team, my own internal performance testing team, and make sure I’m not overloading my systems, and let me roll this out maybe just for people in Michigan, because they might also have Canadian currency, because they’re so close to Canada. Let me just do some very methodical and thoughtful releases. So you get all the benefits of Continuous Delivery, but you have this progression of features.
What are the benefits to those more frequent releases? Obviously, as you’ve said, you can move more quickly, you don’t have a three year lag on what people want, but what are the other advantages?
You can move more quickly forward, you could also move more quickly back. Because if, from the very beginning, you say, “I am going to think carefully about how to segment this feature so that I can release it discreetly to some people,” that also means that you can roll it back very easily. So the way we used to build code was you used to have this massive, huge, monolithic code and you pushed it out, and then if something broke, you couldn’t fix it. You just had to take the entire release back, which was very, very painful. But if you say, “Okay, I have this new feature for bank balances and I’m just going to push this out alone, and if it’s not working and buggy, I can just turn it off.” That’s really awesome, that gives you, as a developer, a lot of stress relief.
Yeah. And to look at it again from the developer perspective, how can you empower developers to move quickly and to implement Continuous Delivery practices so it’s not just top down, someone telling everyone to move faster, but developers actually being in control and having the ability to move quickly?
I think it’s a restructuring of breaking down units of work into smaller and smaller and smaller units. If I want to move quickly, I need to break down this release that used to take six months into the smallest unit that I can do in a week? And then here’s a bitter secret I learned the hard way, some features don’t need to get built. Some features, if you get two weeks into it and you put it out to the world, and the world says, “You know, we don’t really care about that at all,” it’s much easier just to say, “Hey, we had these bright hopes for this feature, but we don’t need to release it. And if this was a six month project, we can save five and a half months of work.”
Making a small increment of progress and figuring out, “Hey, you know what? We really thought everybody wanted this, but they don’t.” Or, “Hey, we have this six month plan, and once we started going this way, we figured out that we really need to go a little more that way.” And if you are doing a whole six month release, you’re just basically drawing a line. If you’re doing it in two week sprints, you could be like, “Hmm,” and end up where you really want to be. Or even just say, “Hey, you know what? We don’t need to do this. This was a bad idea.” And it’s a lot easier, politically, to realize that two weeks in, than six months in.
Absolutely. It seems like you’d give the team a little bit more control over direction as well. And so given that a lot of teams have gone remote now, either permanently or temporarily — does the importance of Continuous Delivery change at all on a fully remote team? And then, are there particular considerations, is it harder, easier, how does it look?
Continuous and Progressive Delivery are even more important than ever. A lot of the things that you could do when you’re all in the same office, you just can’t do anymore. You can’t just yell out, “Hey, is everybody ready to ship right now?” You could try to do that on a Slack, but people might be getting coffee or taking care of a kid. So you have to have even more controls about what is happening when, what is being shipped out, and who has rights to see that. And you also may need to be able to react very quickly. I keep coming back to financial customers, because in the last six months, they have had, I think, 30 years of progress crushed into six months, in terms of people not being able to go into banks anymore. Suddenly everyone wants to check their balance at home, wire money at home, or transfer money. Before the answer was kind of, “Just go to your branch.” You don’t have that anymore, you have to be online. It’s not a choice, it’s not an either-or anymore. You absolutely have to have online support for a lot of functions.
The same thing is happening, sadly, to a lot of functions all over the world, where it used to be like, “Just get to your office fax machine.” You have to have a truly digital function for everything right now.
Yeah, it’s interesting. So it’s a change, not just for the developers, but a change for the end user, and it’s even more important that this innovation is happening, and it’s happening quickly.
Yeah. It’s happening extremely quickly. A lot of processes used to be half digital, half paper and now, you have to be fully digital.
Do you have any tips for teams that are working to coordinate while they’re distributed and working to implement Progressive Delivery?
Take it in small steps. I talked in the beginning about Microsoft transforming from a three year release to releasing every week or so, and I also said you need to do this very quickly, but you also need to do it right. So get comfortable with some low-risk features first. Don’t say, “Hey, we’re going to move from a three-month cycle to a daily cycle,” that’s terrifying. But if you’re at a three-month cycle, look at, “Hey, are there some low risk features that we could try to do in a week-long sprint? Is there something that we can carefully detach and figure out a basic rhythm before we get more into this? Let’s figure out the issues that we have, whether they be communication, tools, or process, on something where if it fails, of course failure’s never awesome, but it’s not hugely not awesome.”
I mentioned that you wrote the foreword to the Engineering Leader’s Guide to Cycle Time. Where does Cycle Time fit into all this? Why is it important when you’re trying to move to Continuous or Progressive Delivery?
Cycle Time is so important. The smaller the batches of work you do, the more impact each batch can have. If everything takes six months, you just don’t have many opportunities, you just don’t have many at bats. If you break it down so that you can say, “Okay, let’s try something new every month,” even that alone is a 6X improvement every month. If you break it down further, you just get more and more chances.
And the really seductive part of Continuous Delivery for a developer is that it’s really fun to see your stuff out in the world. If you’re an engineer, like I was in the ’90s, and you’re grinding away for months on a project, you don’t really get much of a feedback loop. You get your manager saying, “Hey, keep working,” but you don’t get any real validation from the customer. So if you can break down your Cycle Time so that you can say, “Every week, this is out in the hands of our customer, and they tweeted that they loved it, or they put in a support ticket that they had some issues but they liked it and wanted to work with us,” that’s really, really fun for a developer.
The worst thing for a developer is, number one, the worst thing is you ship something and it just totally breaks. The second worst is you work really hard on something and then nobody cares, and you’re like, “Well, I gave up seeing my kid’s practice, or I gave up this happy hour with a friend because I was trying to get this done, and nobody cares.” Which is just hard. So if you’re in this shorter Cycle Time as a developer, where you’re like, “Okay, we worked hard this week and we got it into the hands of the customer, and here’s the five tickets they already opened about how to improve it,” that actually feels really good. It feels like I am helping somebody.
Get articles like this in your inbox.
Trending from Code Climate
Engineering Leaders Share Thoughts on Leadership in Disrupted Times in a New Survey
For engineering teams, disruption to the business can have a significant impact on the ability to deliver and meet goals. These disruptions are often a result of reprioritization and budget changes on an organizational level, and are amplified during times of transition or economic instability.
Built In’s 2023 Best Places to Work — Why Code Climate Made the List
At Code Climate, we value collaboration and growth, and strive for greatness within our product and workplace. For us, this means fostering a supportive, challenging, people-first culture. Thanks to an emphasis on these values, we’ve earned spots on three of Built In’s 2023 Best Places to Work awards lists, including New York City Best Startups to Work For, New York City Best Places to Work, and U.S. Best Startups to Work For.
Turnkey Deployment Delivers Day-One Value for Yottaa
Learn how Yottaa gained immediate value from Code Climate Velocity right out of the box.