Why this matters

Most AI writing quietly assumes abundant cloud compute, stable connectivity, and a single centralized system boundary.

I am interested in a different class of problem: what happens when intelligence has to run closer to the device, under tighter hardware budgets, or inside workflows where autonomous components must coordinate under real constraints.

This is where edge machine learning and agentic AI become part of the same conversation.

On the edge, the questions are about latency, thermals, memory, power, offline behavior, deployment surface, and inference economics. In agentic systems, the questions shift toward context transfer, coordination, observability, recovery, and software architecture between semi-autonomous components.

Put differently: one line is constrained by hardware, the other by distributed cognition. Both are engineering problems.

Reading path

Start with the system boundary

Then move to edge computing directly

What I want this section to become

  • Edge ML as production engineering, not just hobby hardware.
  • Inference architecture under constraint, including cost and deployment tradeoffs.
  • Agentic AI beyond demos, especially around coordination, monitoring, and failure handling.
  • The overlap between the two, where local inference, orchestration, and autonomy start to blur.

This is one of the directions where the blog will keep growing.