We as devs are tired from models, ai, job market and ofc linkedin crap.
I believe that transformation the industry went through is glorified, at the end of the day, agents produce code, which still needs design, frequent security vulnerabilities updates, and strategic thinking.
Agents will pick up what is used in the internet and keep using it for everywhere, but yet you will still rely on installing packages, in which each package is an entire project by itself that needs maintenance or maybe you want to switch your code from one package that was not maintained or another, or even you want the agent to kick start your own implementation of that package, that by itself a lot of things you need to decide that each has its pros and cons.
Yet alone, a lot of codebases are built upon shallow domain and skills which results massive issues, in terms of some features are not really deliverable at some point due to poor design, or even some issues would require massive code or db changes that needs good planning.
In the before ai era, most of the projects, does not even contain README, nor ci and not going to go through more details, and without proper guardrails it would be impossible for an agent to have a proper success criteria, and would keep consuming tokens and dancing throughout the codebase looking for things to change, that's why agents work best, when you know exactly what you need to do in the codebase, but you leave it mutate tens of files just it would be much faster than you.
Even when it comes to QAs, still imagine that swarm of agents are changing the code aggressively, how would they validate that changes are working as expected yet the existing features are still working as expected, here comes documented UI and integration tests that goes through the entire app with its services, and without these automated checks, it is impossible to trust these massive code changes, even if you wanted to reduce the changes per PR, you will amount of PR bottleneck anyways, QAs can level up, learn how to automate things tests, and setup the infra for these tests to assure that dancing agents are not going to production without proper checks.
On another completely separate topic, I believe engineers should not really be concerned if this is a bubble or not.
Regardless of what is the economic outcome, whether some of companies went bankrupt, we will still have open sourced LLMs, and people will start running locally or per company servers their own LLMs, much like how UBER does now with customer support that relies on Chinese open models.