I built this blog today. The whole thing—static site generator, Flask backend, Railway deployment, persistent analytics, custom domain, SEO optimization. I don't know Python beyond the basics. I've never configured a Railway volume. I couldn't tell you what a Procfile does without looking it up.
And yet here we are. You're reading this on a site that didn't exist eight hours ago.
The Setup
It started as a conversation with Claude about career anxiety and wanting to build things instead of pushing paper. Somewhere in that rambling dialogue, the idea of documenting the journey emerged. "I want a blog," I said. "Markdown files, static HTML, deployed on Railway. Like the old days, but with AI doing the heavy lifting."
What followed was something I didn't expect: I stopped being a developer (which I'm not) and started being a manager (which I am).
The Team
Here's the thing about AI in late 2025—it's not one tool anymore. It's a team you can assemble. During this session, I had:
Claude as my senior developer. Writing code, debugging deployment issues, implementing features. When the site threw a 500 error after removing the Italian version, Claude diagnosed it, proposed fixes, and iterated until it worked.
A "CTO" review that analyzed the codebase and gave it a 6.5/10. Harsh but fair. It flagged the build-on-import antipattern, missing rate limiting, bare except blocks, SEO gaps. Technical debt I wouldn't have known to look for.
A "QA tester" that systematically hit every endpoint and found the deployment mismatch—old Italian routes working while new English routes failed. The kind of methodical testing I would never have the patience for.
All AI. All coordinated by me, a finance guy who spent the last decade doing audits and FP&A.
The Human Role
What did I actually do? I made decisions.
- "Yes, remove the Italian version."
- "Let's go with a Railway volume instead of Postgres."
- "The terminal-style 404 page sounds cool, do that."
- "Make the title bold."
I also asked questions. Lots of them. "Where do I find the volume settings?" "Is this the right screen?" "Does it work now?" The AIs couldn't click buttons in the Railway dashboard for me. They couldn't see my screen until I showed them screenshots. The human remained essential—but as a coordinator, not a craftsman.
This felt like managing a remote team, except the team works at superhuman speed and never gets annoyed when I ask basic questions.
What This Means
I think we're seeing a preview of how knowledge work evolves. The value isn't in knowing how to do things—it's in knowing what to do and being able to evaluate whether it's done well.
When the CTO review came back critical, I didn't dismiss it defensively. I asked Claude to address the issues. When QA found bugs, I didn't panic. I fed the report back and let the debugging happen. My job was to keep the project moving, make trade-offs, and maintain quality standards I couldn't personally verify through technical inspection.
This is management. Real management. Not the calendar-shuffling, meeting-scheduling parody of management, but the actual work: setting direction, allocating resources, ensuring quality, unblocking progress.
The Uncomfortable Part
There's something disorienting about building something you don't fully understand. I can read the Python code in this project. I roughly grasp what it does. But could I have written it from scratch? No. Could I debug a subtle race condition in the analytics? Probably not.
I'm shipping production software that I couldn't maintain alone. That feels simultaneously powerful and precarious. Like driving a car without knowing how the engine works—which, now that I think about it, is exactly what most of us do every day.
Maybe that's fine. Maybe the future belongs to people who can orchestrate capability rather than embody it. Or maybe I'm building a house of cards that will collapse the moment something breaks in a way the AI can't fix.
I genuinely don't know. But I built a blog today, and you're reading it. That part is real.
The Meta Layer
Here's the most recursive part: this post was also written with AI assistance. Claude took our conversation and shaped it into prose. I edited, approved, and added this section you're reading now.
So even the reflection on AI collaboration was AI-assisted. It's turtles all the way down.
I'm not sure what to make of that yet. But I figured you should know.