In last six months, a lot of small and mid sized software development businesses that have one-two mid sized projects in the work discovered that they only really need two people for all but the biggest software projects.
You need one senior software architect who does design aspects of the software, and one really skilled senior coder that is good enough to clean up inputs into AI and sanity check output of AI (usually with a help of a different AI model, as just like in real world, best method of sanity checking one person is to ask another person).
All the supporting coders, developers, designers, etc which is what fresh graduates are hired to do? They were support staff for those two before. And AI now does support for those two. Faster and with less mistakes overall once work flow for those two is properly built up. This is in fact the main current bottleneck for AI adoption. It's no longer model development. It's upper level people learning how to have AI handle their support tasks instead of support staff. The commands are different, as is methodology, as you no longer need to learn which of your support staff are good at what tasks, and what kind of words are needed to get them to do those tasks well enough. Instead you need to learn which models have what strengths, and what kind of inputs are needed to do those tasks well.
As often noted, AI doesn't replace "people". AI replaces "people doing support for those actually doing the critical work". People who still do things that are really important have more work than ever.
But all their support staff? That's going away once work flow for how to get same/better outcomes is worked out.
Notably this isn't just software development. We see this in everything from university research work (where feeding inputs into a model rather than having an undergrad sort through them is rapidly becoming a norm), to McKinsey and co consulting giants where there's no more junior consultants, it's all senior consultants having AI support them in terms of data collation etc.
If your job is fundamentally about providing some kind of a data generation support for someone above you, get to that level above you, be in terms of coding, collating, computing, inputting, or similar task. Or you have a very high chance of your job not existing soon. Because that is what current AI already does better than overwhelming majority of currently employed people who do those tasks. Current bottleneck is no longer in developing the models, but in the upper level people learning how to order AI to do the same tasks they have to order support staff to do right now. So everyone doing those lower level data related jobs are on a timer. When upper level people learn how to order AI assistants to do the support tasks, these jobs are going away.