I'm not sure this really says anything novel. As a longtime user of Duolingo, I feel like my experience of the app is "move fast and break things" -- constantly adding new features and removing old ones, without regard for how it affects existing users. It's interesting that they talk about how they care about user retention over new user acquisition, since that's not my experience. Many features that I found useful have been removed (e.g. word reference lists, forum discussion), and new features added/changed that I don't care for (e.g. many types of review-ish lessons with unclear differences, semi-invisible adjustments to thresholds necessary to reach certain goals).

My assumption is that long-term user retention is a metric that's difficult to measure and optimize for, and so even if they claim to care, it ends up getting worse over time. If your experiment grows DAU by 1% but makes 0.1% of users leave over the next month, most of that 0.1% isn't even noticed in the week that the experiment is running, and a 0.02% drop in user retention isn't statistically significant. But then you ship 1 of those experiments per week for 5 years, and even though you grew your userbase 12x you end up with 25% worse long-term user retention than you would have otherwise. And then you don't even realize that because it's such a different product compared to 5 years ago that you brush off any changes over that timescale as "our new demographics are inherently less likely to stay and there's nothing we can do" or "society is changing how it interacts with phones and there's nothing we can do." Or maybe I'm being too generous and the reality is "there's definitely something we could do but we're too busy rolling in our fat stacks of cash to care"