As I sat down to drinks the other night with an old coworker, the topic of AI and ridiculous KPIs around it came up. I've heard and read some similar things, at least in bigger tech organizations where a developer has to increase commits by X% or pull requests by Y% and that AI should make that goal easily achievable since it can solve all problems! In my opinion, this is yet another case of consulting firms convincing the C Suite that the new flavor of the week will change the world and how they work in it and boy are they raking in the money selling that story.
Similar to the "Agile" revolution before this and even further back, I've sat through and listened to those consultants sell each idea to wide-eyed execs that want a silver bullet to their current problems. Each one is branded with the key to unlocking your team's productivity but let's be honest, writing the code to solve a problem has never been the real bottleneck, especially in larger organizations.
Let's think back to your last feature delivery. The business/product people wanted something that I'm sure they thought was very simple. How hard can it be to add a label on a page that shows if I'm on track for retirement for example? Often times, what seems like a simple request like this will evolve into a complex feature because there are hidden requirements to think about. One might say, "Feed their request to Claude and see what pops out", and I'm sure Claude will give you some approximation of code that would do something somewhat related to the request. The bigger problem is context and that context is lacking for both AI and the team until they dig deeper to figure out what it really means to say "someone is on track for retirement". These discussions usually result in finding the data that needs to be used, following data governance guidelines, finding or writing the APIs needed to access that data, discovering security concerns, etc. A lot of that will be ignored by any AI or productivity framework that has been put into place and all of this is assuming that the development team gets a concrete feature request at the start. What I just described is a pretty ideal world where the development team is spending 80-90% of their time refining the feature request to make sure they're building the right thing because its unusual for everyone to understand a request the same way that the requestor meant.
In the end, productivity increases come from clear communication, context, and understanding. Without these, the same problems will appear time and time again, no matter if you have the "perfect" Agile team or the "perfect" AI to implement your line of work. Often times it is hard enough to get the right audience of people in the same room, but even when that's been accomplished, the content that's being delivered is high level at best and doesn't adequately describe the goal or objective. On the other hand, we need to stop blindly trusting what some consultant says or what we see as the next "silver bullet for productivity" and take some time to understand what the tool or framework is set out to do and how to use it best. I use AI every day in my development, mostly as a search and discovery power house and to "rubber duck" my code and its fantastic at doing that in most cases. I wouldn't ever trust it to fully understand my thought process that went into the prompt asking it to create something for me though, especially without checking the work afterwards, but there seems to be a decent proportion of people both technical and non-technical that implicitly trust the AI knows what it is doing which is a frightening prospect.
Some of the success stories that you'll hear is primarily in smaller organizations and garage bands. I fully believe that introducing AI into a small organization can increase productivity because the overhead of everything else is so much smaller. There are less hoops and red tape to go through so when you consider the proportion of work from earlier, that likely drops for 80-90% in discovery and prep to 10-30%, so of course if you can cut time from developing code then go for it. For the majority of us though, we don't exist in those new, small organizations. We need to contend with "legacy" code, processes, and procedures that have been put in place which means we have to spend more of our productive time with others and communicating. If we really want to increase productivity, we need to find ways to make our time with others extremely productive such as bringing a set agenda, getting the right people in the room, and making decisions (!!!) so that those that need to build the thing can have a clear picture of what needs to be built.
Put simply, productivity doesn't come from whatever framework or tooling is popular this week/month/year. Productivity comes from building trust within your team and organization, allowing them to set themselves up for success with the right people, and letting them communicate clearly without worry of blowback. We can only be productive when the goals and objectives are laid out clearly and everyone understands the work that needs to be done to accomplish them. AI, Agile, etc. aren't going to solve that problem for you, put your people first and let them do the work you've hired them to do instead of trying to grab at any metric that has a number attached to it.