7.7 billion tokens.
That's my Claude Code consumption in 2025—roughly 630 million tokens per month, putting me in the top 25 on the Claude Code usage leaderboard.
Here's what all that usage (plus tokens from other AI tools) translated into:
✅ One full-stack vertical AI app launched
✅ Two AI apps in the build
✅ Seven vertical AI agents created
✅ 10+ professional workflows experimented
✅ Deployments across Microsoft Azure and Google Cloud
✅ Hundreds of pages of documentation (notes, lessons learned, "what not to do")
After witnessing the scary pace of AI advancement in 2025, I see two things that could become even more valuable in 2026:
→ Domain-specific knowledge — for translating general LLM capabilities into real-world professional applications
→ The ability to build and deploy — putting together the right technology stack that fits into your organization's IT infrastructure, policies, and constraints
Do I worry about being replaced by AI, including the AI tools that I build myself? Not really. I'm more focused on offloading low-value tasks so I can spend more time on the work I actually enjoy—delivering urban infrastructure that shapes how people experience space.
And the more you work beyond the AI chatbox, the more you understand both the power and the limits of current LLMs. Replacing humans is not happening in 2026, at least not from what I'm seeing in real deployments.