Start with upgrade cadence. Forty-three percent of respondents were already on Django 6.0 within months of it shipping [4], which means a large share of these shops can take a framework bump without standing up a project plan for it. That spare capacity is what the rest of the survey spends.
Now the arithmetic on the AI numbers, because the denominators move. Ten percent report using no AI tools regularly, so at most 90 percent use something [17]. Fifty-eight percent of that group is in a model every day, which puts daily use at roughly 52 percent of the whole sample [18]. The autonomous cohort is smaller than the coverage implies: 27 percent of AI users run multi-step tasks unsupervised [10], or about 24 percent of respondents [19].
Read the category boundaries before quoting any of it. Fifty-six percent of AI users say they use it purely for chat and advice [8], and 44 percent let it edit files or run commands when instructed [10]. Those two sum to exactly 100 [20]. But 59 percent also report having AI generate code that they then apply themselves [10], and 59 is larger than 44. Both can only hold if generate-and-paste is being counted inside chat. So "supervised" here means a human still moves the bytes into the repo. If you write a team AI policy against these percentages, that is the definition you inherit, and it says nothing about whether the human read the diff.
The sourcing deserves the same treatment. PyCharm co-runs the survey and publishes the highlights, and the same post offers 30 percent off the IDE with the proceeds donated to the Django Software Foundation [15]. The write-up's conclusion is that AI is changing the IDE faster than it is replacing it [16], supported by nearly half of AI users working through an IDE integration [9]. That is the finding most flattering to the publisher, which is a reason to take a second reading from your own team rather than a reason to discard it. For the number to transfer, your developers would have to have found the survey through the same channels and to read "IDE integration" to mean the same thing a respondent did.
This risk sits in the files that configure everything around the framework, not in the framework itself. Type hints are at 57 percent with no settled checker, which the survey itself calls a wide-open race [14]. The described direction of travel is agents moving from answering questions to editing files, running commands and completing larger tasks, with the boundaries between editors, terminals and automation blurring [22]. Every one of those changes lands in pyproject, the pre-commit config, the CI image and the merge rules. The core you are not touching is what gives you room to get those wrong and recover.
