The center of gravity

This blog is now organized around AI Software Engineering.

That phrase is doing specific work. I do not mean “AI” as a synonym for prompting, nor “engineering” as a synonym for deployment scripts. I mean the full problem of building AI systems that remain legible, testable, cost-aware, and operationally reliable once they leave the notebook.

In practice, that breaks into four connected lines of work:

  • LLMOps as the successor to classical MLOps. Once prompts, tools, traces, evaluations, and agents become part of the system, the old abstractions stop being enough.
  • Systems design on GCP. Infrastructure choices shape latency, reliability, cost, and team velocity. I care about the engineering consequences of those choices.
  • Edge machine learning and inference. Raspberry Pi, Jetson, constrained hardware, and the question of what happens when the cloud is not the whole story.
  • Agentic AI as a software architecture problem. Coordination, observability, context handoffs, and failure modes between agents matter as much as model quality.

If you are specifically interested in hardware-constrained deployment and autonomous workflows, go next to Edge + Agentic AI. If you want the cloud-platform angle, browse the GCP-tagged posts.

Reading paths

LLMOps and production discipline

Architecture, inference, and hardware tradeoffs

Foundations that support the engineering work

What comes next

The direction from here is narrower and more deliberate:

  • More on LLMOps as an engineering discipline, not just a tooling stack.
  • More on GCP-oriented architecture for real production systems.
  • More on edge + agentic AI, especially where hardware constraints reshape system design.
  • More on AI Software Engineering as the umbrella that makes these threads cohere.

If you want the broadest orientation first, start with Start Here.