LaunchDarkly

I use LaunchDarkly for feature flags to turn on and off features that we deploy into production. I have also used LaunchDarkly to manage AI in production. We're using it for AI config, which are basically the prompts. So we are using it for flags and we also use it for the prompts, to store the prompts.

LaunchDarkly has helped me solve the challenge of iterating quickly on prompts and instructions. Having the instructions and prompts in LaunchDarkly makes it fast to iterate and just like flags, I can go into a UI and make the change, rather than having to deploy code, because the CI workflows tend to be a lot slower. We ended up wanting to have the instructions as code anyway, so we had to work on a specific pipeline to be able to still store that from our side as code and also publish it to LaunchDarkly. The main advantage is the same as the flag's flexibility—basically being able to change it quickly without having to actually wait for a PR to be merged and something to be deployed.

LaunchDarkly works well and it's nice to have the control of a feature flag as a whole. It's very handy. The only thing that I would like is a more structured way to visualize the flags, especially when you end up having too many flags and there is some hierarchy. The UI and the way that it is presented is very flat. When you end up having structures and dependencies on flags, there is no dashboard that would give you exactly what you want, especially when you're working on a particular project. Because the way that we have in our organization is that we have multiple products, multiple features and multiple teams that are working on features. You end up having a lot of different flags that are presented in a flat manner, and you may have a main one that when it is turned on, the child ones are evaluated based on it, and you may have many depths on that relationship. If there were a dashboard or a certain view that would make that structure apparent and very obvious, it would be useful.

Also, because you have different values depending on whichever environment that you're interested in, you have to actually dig into what are the relevant flags and then go to a particular environment that you are focused on. If there would be some sort of dashboard that would give you all of that as a whole, not individually by flag, but as a whole structure is constructed and your interest checked on, it would be handy from the flag side.

I would like to see all of the flags and their relationships and I would like to see a tree type view. As a developer, I may be working on a particular piece from a particular team. If I'm able to just group them the right way based on the hierarchy of the flag and just select the particular feature that I'm working on and only see that and see all of the hierarchy, and also see all of the values based on the environment that I am interested in at a glance, would be very handy and that would save a lot of time.

I have been using LaunchDarkly for maybe six months overall. I have been working in my current field for seventeen years.

There was one time that there was an outage, and we had to act on it because the default that we were getting actually ended up breaking some other feature.

I am not aware of any performance issues, but I am aware of one time that there was an outage and we had to act on it to fix it.

We have a thread in Slack where we are raising issues directly to LaunchDarkly support. I'm not really involved in that, but I do know that this thread exists.

I have used an in-house solution previously. Comparing to that, LaunchDarkly is very feature rich, but it's basically because we were doing it ourselves. We are also using LaunchDarkly for AI configuration and that also seems intuitive, but it also ends up suffering from the same symptom of having everything be very flat. We have been having issues with certain variations of AI config being too large. There is a certain limit that LaunchDarkly imposes on how much you're able to post as a variation for an AI config. We have been having a lot of issues with that recently. We did a migration from the in-house solution to LaunchDarkly.

I didn't actually do the platform part, but from my point of view, I was just moving existing flags to be in LaunchDarkly flags. I didn't do any work regarding that, but because it was already taken care of, I was just using the flags.

There's a platform team that is taking care of the maintenance side, so it doesn't require maintenance on my end.

I wouldn't be in the position to actually compare metrics, but LaunchDarkly seems to be pretty performant. The client in the browser gets the events fairly quickly.

I give LaunchDarkly a rating of eight out of ten overall for everything.

Regarding my use cases, I am cautious about breaking customers in production, so I generally put LaunchDarkly flags on pretty much any feature that is going to be rolled out to production to slowly roll things out and ensure nothing is broken. Additionally, it is valuable for developing a new feature and then launching it all at once when it is ready to go, so I do both of those.

I am not using LaunchDarkly to manage any AI in production, as other teams may be doing that. At the moment, my use of AI is limited to coding assistance, while other teams are integrating AI into the product.

What I appreciate most about LaunchDarkly is that it is very easy to use. I estimate I have around 30 flags, though I have not counted them recently. Finding the flags is straightforward; I search for the flag I need and then proceed with the setup for whatever I need to do. One useful aspect for my workflow is that whenever I set a flag or change something, I ask my coding LLM to confirm that everything is working as it should, whether it is to catch a typo or ensure I did not add the wrong merchant ID.

What I dislike about LaunchDarkly is the inability to quickly find all the flags I have currently running and their statuses. I want to know if they are rolled out partially, but I have to check externally by asking my coding agent about all my LaunchDarkly flags and their statuses. Recently, I forgot about a flag I rolled out to a couple of merchants, and I noticed I was not alone; several people had flags running for months with only one or two merchants enabled. A feature to alert me if a flag is underutilized or needs attention after a week of being rolled out would be beneficial.

Overall, I have been working with LaunchDarkly for a couple of years, having used it at Instacart and currently at Thanks, but it was not utilized at Thomson Reuters. I have a couple of years of experience with it.

I have not encountered any scalability issues with LaunchDarkly. I work for a mid-sized company with hundreds or thousands of users, and I have not run into scalability issues here. While my experience at Instacart might have been different, I was not using it as in-depth as I do now. I currently use LaunchDarkly on a daily basis.

I have not integrated any other tools with LaunchDarkly; it stands by itself in my usage.

Other than the server that checks the status whenever I request it, which saves me from going directly to LaunchDarkly, I do not have additional tools integrated.

I would give LaunchDarkly a rating of nine out of ten because I never give out tens, as that implies no room for improvement, and everybody can improve. Overall, LaunchDarkly is good; it is easy to use, and I can see what is happening relatively quickly.