AI, Bubbles, and the Future of Developer Workflows

The early buzz that AI would replace all developers has quieted, and in its place a new wave of skepticism has emerged. Some argue that generative AI is simply a bubble ready to burst. That may be true, since hype cycles always peak and corrections always follow. Even if AI experiences a dramatic contraction, history shows that the long term impact will remain.
Anyone who lived through the dot com bubble understands this pattern well. The crash was severe, but it also laid the foundation for the modern Internet, cloud computing, and nearly every consumer product we use today. A possible AI bubble burst would likely follow the same path. Expectations would correct, but the structural changes would continue to shape how we work.
One of the most significant of these changes is the rise of agentic workflows. These are systems where AI not only answers questions but also executes multi step tasks autonomously. When used well, these workflows speed up work that once required research, custom code, or tedious manual processes.
A small but meaningful example is this very blog.
For years my blog lived on a Fly.io hosted site with its content stored in a SQLite volume. Over time I wrote less, used the site less, and eventually discovered that the entire thing had stopped functioning. Once I revived my portfolio site and wanted the blog content back, I faced a challenge. How do I migrate data from a dead platform into my CMS of choice, which in this case is Sanity, especially when the original database was not directly accessible?
This is where Claude Code came in. Claude Code is Anthropic's agentic coding environment, and with only a few conversational prompts it handled the recovery and migration from start to finish. It completed work that would have taken hours of manual effort.
Here is how it went.
- I asked:
"I have data in a Fly.io SQLite volume. Can I pull it down and review it locally"
Claude suggested steps for manual export, asked clarifying questions, located the volume, and downloaded it. - Next:
"Can you export the data into CSV files for easier review"
It exported everything perfectly. - Then:
"Can you convert the Post.csv data to match my Sanity schema"
Claude produced fully translated JSON for each post. - Finally:
"Can you write me a script using the Sanity CLI to import this"
It generated a runnable import script.
When errors appeared, I pasted them back and Claude resolved them one by one.
Within one hour all of my old Markdown based posts were recovered, downloaded, transformed, and imported into my new Sanity CMS with almost no manual coding.
Not bad.
Whether or not AI is in a bubble, workflows like this are not going anywhere. They are too useful, too practical, and too transformative for everyday development tasks. Developers are not being replaced, but the way we build software is changing quickly, and agentic assistance is becoming standard in the modern toolkit.

