The meta-layer · the steward
Everything else on this site is the output — a QX-250 digital thread carried from a customer RFP to a flight-verified design. The AI Lifecycle Harness is the engine: the application that reads the RFP, frames the work as an agentic Work-Breakdown Structure, orchestrates the Siemens Xcelerator tools across the whole lifecycle, and measures itself as it goes. This weekend it framed and drove the build — with a human still prompting each step. It is built to steward that loop going forward.
The loop it stewards
The QX-250 program did not start with a drone — it started with a customer need, written as a request for proposal modeled on real U.S. Air Force solicitations. Everything downstream is an answer to it. The harness's job is to carry that need, unbroken, all the way to a verified aircraft:
How it knows what to build
A harness can only steward what it can represent. The RFP is decomposed into a structured program of work — an agentic WBS — in which deliverables must meet criteria that satisfy accomplishments that deliver to milestones. That structure is what turns "build a drone" into a list of concrete, checkable units of work an agent can execute and a human can govern — the same Program-Planning / IPP&E backbone the managed program already uses, pointed at the agent's own work.
A concrete artifact — a requirement set, a schematic, an assembled model, a verification result.
Machine-checkable conditions the deliverable must meet — the line between "asserted" and "proven".
Deliverables roll up into accomplishments, which deliver to program milestones and the gates a human approves.
This weekend — the honest baseline
Here is the distinction that matters. Every human touch this weekend was direction, not operation: no human ever clicked a button, dragged a node, or typed a value in NX, Capital, Teamcenter, Simcenter Amesim or MADe. The AI operated every tool itself — by API, MCP bridge and browser automation. The Time-to-Solution Dashboard counts 26 human touches (22 prompts, 2 approvals), and every one was a human saying what to do next — never "here, let me do that part."
What has to become true
The harness is only as good as the tools it can drive and the honesty of its own measurement. Two views on this site are its instruments:
Can each tool be driven headless, MCP-native, agent-legible and orchestratable? The readiness the harness depends on — and what must change to get there.
Open scorecard →The measurement: human touches, iterations, rework, seam cost, verification coverage — the baseline the harness has to beat, build over build.
Open dashboard →