Waterfall → Iterative → Agentic: Our new handbook on AI Systems Engineering

Every generation of hardware has demanded a new way of working.

When hardware was mechanical, teams could define the full system up front and execute in order. When hardware went digital, with dozens of ECUs, hundreds of sensors, and millions of lines of code, that sequential approach broke down. Iterative methods replaced it: shorter feedback loops, earlier cross-domain integration, digital tools that let you change a design without starting over.

Now hardware is converging on autonomy. The number of cross-domain interfaces, regulatory frameworks, and interdependent artifacts has grown beyond what iterative methods alone can manage. The methodology needs to change again. The teams that move early will compound their advantage while the rest spend years catching up.

The problem is not speed. It is coverage.

A modern hardware program generates thousands of requirements across dozens of subsystems. Interface documents get revised, test procedures modified, compliance matrices extended. No human team can review all of it. They sample. They spot-check. They review what seems highest risk and hope the rest is fine.

Under schedule pressure, which is always, the sampling rate drops further. Inconsistencies slip through, not because engineers are careless, but because comprehensive review at scale is beyond human capacity.

The answer is not to review faster. It is to stop relying on periodic review altogether.

What changes

Chapter I of the AI Systems Engineering Handbook lays out the structural shift:

The methodology evolves. Sequential development fit mechanical hardware. Iterative methods fit digital hardware. Autonomous systems need continuous alignment, where AI monitors every requirement, interface, and test case on every change rather than at periodic reviews.

The role of the engineer evolves. The Systems Engineer becomes the Systems Architect. Execution time compresses. Architecture time expands. The engineer's job is not to do less. It is to spend all of their time on the decisions that shape the product.

The infrastructure requirements become clear. A system of record, a change management model, codified standards, and an ecosystem of agents operating within defined boundaries. Without these, AI amplifies inconsistency rather than reducing it.

Read Chapter I