Engineering journal
The Next Chapter of AI Is Systems Engineering
A reflection on agent interoperability, on-device inference, and how reliability can turn growing AI capability into useful, dependable systems.
Capability grows through connection
The most important AI news this year has not been a single model release. It has been the growing attention paid to what happens around the model: how agents identify themselves, discover tools, exchange context, request authority, and coordinate work.
In February, the US National Institute of Standards and Technology announced an AI Agent Standards Initiative (opens in a new tab) focused on interoperability, security, identity, and open protocols. The Linux Foundation has also reported broad organizational adoption of the Agent2Agent protocol (opens in a new tab), alongside the Model Context Protocol rather than in place of it.
Those developments bring a familiar systems lesson into focus: autonomy becomes more useful when its interfaces are explicit. A capable model is one component in a larger design. Its behavior can become easier to trust when contracts around identity, permissions, data, failure, and ownership are visible and testable.
Intelligence is moving toward the edge
Another signal is the movement of generative AI toward local devices. In May, Google introduced an AI Edge Portal for benchmarking models across more than 120 Android device types (opens in a new tab). Its measurements—initialization time, prompt processing, token generation, and peak memory—are a useful reminder that intelligence eventually meets physical constraints.
This is where AI begins to resemble the embedded and connected systems that shaped my engineering perspective. Memory, power, latency, thermal limits, connectivity, and privacy are not deployment details; they determine the behavior people actually experience. The path from a physical interface through embedded software and a connected platform to applied intelligence is one system, not a sequence of independent handoffs.
The closer inference moves to the device, the more valuable it becomes to understand that entire path. An impressive model can become a dependable product when the surrounding interfaces preserve context, timing, and safe control.
Reliability makes capability useful
The final trend I take seriously is a sharper distinction between capability and reliability. A 2026 study, Towards a Science of AI Agent Reliability (opens in a new tab), found that gains in task accuracy translated into much smaller improvements in reliability. It argues for evaluating consistency, robustness, predictability, and safety—not just whether an agent succeeds once.
That distinction aligns with how I want to approach applied intelligence. Evidence should be attached to the behavior it supports. Actions should be bounded. Failure should be observable. Interfaces should make assumptions and responsibility visible. The aim is not to wrap a system in AI, but to carry intelligence through the system in a form that can be tested, understood, and maintained.
My reflection is that the next chapter of AI belongs to systems engineering. Growing capability gives us an opportunity to design the agreements between models, tools, devices, and people more deliberately—and to carry those agreements into useful, dependable implementation.