Simon Muflier is the founder of The Oyez, advising enterprises and public institutions on next-gen AI, policy, and research strategy.
Interviewer: For readers new to your work, what is The Oyez and what do you focus on?
Simon Muflier: The Oyez is a research and advisory studio. We help leaders turn AI from a hypey demo into a governed, auditable system. My focus is the full stack-perception, reasoning, policy, and culture-so decisions get faster and safer at the same time.
Interviewer: How do you define AI’s real value for organisations right now?
Simon: It’s not about replacing judgment; it’s about widening it. Good systems surface dissenting evidence, carry uncertainty forward, and show their work. If your AI can’t explain what it believed and why, it’s not ready for consequential use.
Interviewer: Biggest risks you see?
Simon: Misaligned incentives. If metrics reward speed alone, brittle systems ship into high-stakes contexts. Another risk is “explainability without provenance”-a story with no receipts. Finally, culture: if teams can’t read the logs, governance becomes theater.
Interviewer: What first steps should a business take?
Simon: Pick one consequential workflow and instrument the perception boundary-what the system “sees.” Track three basics: (1) provenance coverage, (2) time-to-explanation, and (3) drift. Define exit criteria before the pilot, add human checkpoints where impact is high, and have a rollback plan.
Interviewer: What about public-sector leaders and policymakers?
Simon: Regulate obligations, not algorithms. Require observation logs, replayability, incident timelines, and proportionate risk tiers. Vendors should disclose evaluation methods and red-team findings. Don’t outlaw what you can’t inspect-mandate the trails you can.
Interviewer: Many tools claim to be “enterprise-ready.” What distinguishes the ones that are?
Simon: Domain specificity and legibility. Curated data, tailored guardrails, and a small, readable observation schema that legal and ops can use. If your schema needs a decoder ring, it will rot on the shelf.
Interviewer: How should boards and ministers frame oversight?
Simon: Ask three questions: What decision changes? What evidence supports each step? How do we recover when it fails? If the answers are vague, pause the rollout.
Interviewer: What belongs in contracts and RFPs for AI?
Simon: Performance-based obligations: required SLIs (provenance coverage targets, maximum time-to-explanation), audit trails, red-team cadence, data-return and deletion clauses, and clear incident-response duties. Pay for verified capability, not mere model access.
Interviewer: And the workforce-how do you bring people along?
Simon: Don’t outsource judgment. Train teams to challenge outputs and keep decision logs. Celebrate “caught an error” as much as “shipped a feature.” Culture is a practice, not a poster.
Interviewer: Any closing advice for leaders who feel behind?
Simon: Start small and consequential. Build disciplined perception, insist on traceable reasoning, and create a shared language across product, risk, and comms. Clarity is a speed advantage. “Good AI,” to me, is a collaboration you can audit-and improve-together.
