Your whole product rides on other people’s models. When the next model generation makes raw AI good enough that governance feels unnecessary — or the providers launch their own board products — what happens to you?

The full executive answer

I would separate the two threats, because they have opposite answers. If models get dramatically better, our costs fall and our product improves overnight — we are deliberately model-agnostic, running Anthropic as primary and OpenAI as secondary through a provider abstraction with a model registry, so we swap engines without rearchitecting. But better models do not make governance unnecessary; they make it more necessary, because boards will trust AI with bigger decisions. Nobody argued that better pilots made flight recorders obsolete. The demand driver for our category is not model weakness — it is accountability: the EU AI Act, SR 11-7-style model-risk expectations, and audit committees asking "prove how you knew." Those pressures intensify as capability grows.

The scenario where a foundation-model provider ships a boardroom product themselves is the sharper risk, and I plan for it the same way I plan for Microsoft: they are structurally conflicted as both the model vendor and the auditor of their own model, which is exactly the independence problem SR 11-7 exists to prevent in banking. Our cross-provider consensus — two independent providers checked against each other, a third model adjudicating disagreements above a measured threshold — is a control a single-vendor product cannot honestly replicate, because you cannot independently challenge yourself. Independence is our permanent structural position, not a feature.

Current state versus roadmap, plainly: today the abstraction, dual-provider consensus, and per-generation cost telemetry are live and CI-tested in the codebase. On the roadmap is broadening the adjudication pool as new frontier providers mature, so the independence argument gets stronger over time, not weaker.