An insight is hardening across the AI infrastructure community: Agent = Model + Harness. The model provides cognition — reasoning, language understanding, generation. The harness provides everything else: runtime, orchestration, tooling, memory, policy, sandboxing, verification, and observability.
Models are converging
GPT-4o, Claude, Gemini, Llama, DeepSeek — the frontier is increasingly interchangeable. The differentiation that mattered two years ago is eroding. What does not erode is the execution layer.
LangChain demonstrated this empirically: changing only the harness — not the model — improved a coding agent's score from 52.8 to 66.5 on Terminal-Bench. The harness, not the model, was the lever.
The $100 billion opportunity
If models are converging, the durable infrastructure opportunity is the harness: the execution layer every model runs through. Whoever owns that layer controls how raw intelligence becomes production work.
SIGNET is the procurement harness
SIGNET occupies this layer for procurement. It is model-agnostic, tool-agnostic, and organisation-agnostic. Any frontier model can serve as the reasoning component; SIGNET provides the runtime, the tool registry, the policy engine, and the audit trail that turn that reasoning into governed procurement actions.
That is the bet Concert is making: not on a model, but on the harness around it.