From the theoretical framework of AI agents in orthopaedics to the 60-skill ORTHO-X OrthoFlow™ system for surgeon-governed clinical decision support.
Only months ago, a leading orthopaedic journal set out a clear vision for the next phase of AI in orthopaedics: a shift from narrow, isolated tools toward agentic systems capable of supporting complex, multistep workflows across the patient journey.
ORTHO-X OrthoFlow™ is that vision, implemented — sixty composable, surgeon-governed skills.
In late 2025, Knee Surgery, Sports Traumatology, Arthroscopy (KSSTA) published a review by Oettl, Hirschmann, Samuelsson and the ESSKA Artificial Intelligence Working Group — “Artificial intelligence agents in orthopaedics: concepts, capabilities and the road ahead“ (DOI 10.1002/ksa.70109). It is one of the clearest statements yet of where orthopaedic AI is heading. Its central argument: the era of narrow, single-task models — read this X-ray, predict that one outcome — is giving way to AI agents that coordinate the whole patient journey, from first referral through diagnosis, scheduling, the operating room, and rehabilitation. The authors describe a “society of mind”: many specialised agents working in concert, under human direction, augmenting rather than replacing the surgeon.
It is an excellent map. But a map is not a road.
A review article — by design, and by the authors’ own candid framing — describes a destination and names the obstacles. It does not deploy. And the obstacles the KSSTA authors name are precisely the ones that have stalled orthopaedic AI for a decade: most of the electronic health record is unstructured and messy; agents need validation and explainability, not just accuracy; evaluation has to include cost-effectiveness, not only task completion; liability must be traceable to a named human; consent has to cover agents that learn from data; bias and access-equity must be watched; and regulators are still writing the rules.
Orthopaedic care is not a decision. It is a journey.
A patient arrives through a referral, an emergency, a second-opinion request, a fracture clinic, a sports injury, an arthroplasty consultation, or a postoperative complication. From there the case travels across imaging, classification, treatment planning, implant selection, surgical preparation, the operating room, ward care, discharge, rehabilitation, home monitoring, billing, registry reporting, outcomes, and sometimes post-market evidence. Most AI tools touch one fragment of that journey. The operational question is not “which model is best?” but: what would it take to coordinate the entire journey with safe, bounded, auditable assistance — while the surgeon stays responsible? The answer is not one large chatbot. It is a structured agentic workflow layer.
This is where ORTHO-X OrthoFlow™ begins
ORTHO-X OrthoFlow™ is the implementation of that roadmap: sixty composable skills spanning the entire orthopaedic patient journey — referral and intake, imaging and classification, differential reasoning, treatment and implant planning, the operating room, discharge, care at home, outcomes and registry, billing, quality, education, research, and a dedicated governance layer — coordinated as exactly the “society of mind” the paper describes, with a master orchestrator sequencing the right specialised skill at each step. It is built on open standards: the SKILL.md agent-skill format and the Model Context Protocol (MCP) — so it runs today in a modern agent environment and air-gapped on a hospital’s own hardware.
ORTHO-X OrthoFlow™: Selection of OrthoSkills
in Claude Cowork User Interface
Limited open version available at
https://github.com/MAIVAN-ai/OrthoSkills
© MAIVAN.ai Medical AI Validation Network, 2026 coordinator@maivan.ai
What makes ORTHO-X OrthoFlow™ more than a demonstration is that the paper’s hard problems are not footnotes — each has an answer built in. The roughly 80%-unstructured-EHR problem is met by a normalisation front-end before any reasoning happens. Explainability and cost-effectiveness — the two axes the KSSTA authors say current evaluation omits — are wired directly into a certification harness that decides, per skill and per risk tier, whether a model may act autonomously, must remain human-in-the-loop, or is not certified at all. Liability, consent (including consent for an agent to learn), bias and access-equity, audit traceability, and regulatory escalation are not aspirations bolted on later; they are a governance domain of their own. The field described the guardrails. ORTHO-X OrthoFlow™ ships them.
And it holds the line the paper draws most firmly: agentic behaviour delivered through composable, governed skills — not an opaque autonomous system. Every clinical output is a draft a surgeon reviews and signs. ORTHO-X OrthoFlow™ informs and plans; it never executes surgical action.
What it feels like for a surgeon
Surgeons do not need another inbox. They need better case context, better preparation, better documentation, better follow-up, and fewer preventable workflow failures. ORTHO-X OrthoFlow™ is built around the practical questions that define an ordinary orthopaedic day: Is the referral complete? Is the imaging adequate? Is there a red flag? What is the likely classification? Which pathway is being considered? What must be checked before theatre, and documented during it? Is the patient ready for discharge? What needs monitoring at home? What outcome data should be captured? What needs sign-off? — and, perhaps the most valuable question an agentic layer can keep asking on your behalf, what is missing?
Why this matters, depending on where you sit
Hospital CEOs and CFOs — Orthopaedics is a high-volume, high-cost, high-complexity service line, and fragmentation is where the money and the quality leak: delays, cancellations, incomplete documentation, coding leakage, avoidable readmissions, thin outcome capture. ORTHO-X OrthoFlow™ is the coordination layer that turns a department into an Agentic Smart Ortho Clinic — structured intake, pathway-aware triage, perioperative support, discharge and care-at-home continuity, QA and benchmarking, billing completeness, registry and post-market readiness, governance by design. It adds intelligence across your systems; it does not replace them. And your case data stays a sovereign asset, not a leak.
CTOs and CIOs — the next phase of healthcare AI is won by safe orchestration, not isolated model calls: internal-only data access, controlled tools, role-based permissions, audit logs, human sign-off. Because ORTHO-X OrthoFlow™ speaks SKILL.md and MCP, the systems you already run — EHR, PACS, registry, billing, analytics, and engagement/CRM layers such as Salesforce — connect as tools, not as a rip-and-replace. Hospitals do not need a black-box chatbot; they need an auditable agentic layer that sits safely beside existing infrastructure. That makes ORTHO-X OrthoFlow™ a platform to build on — and an open invitation to deployment and integration partners, including the Salesforce/MCP ecosystem.
Clinical and quality administrators — consent, traceability, bias monitoring, and regulatory flagging are first-class capabilities here, because in a clinical setting governance is the product, not the paperwork.
The validation path
ORTHO-X OrthoFlow™ is, today, an educational and decision-support reference — and the honest next step is validation, before deployment at scale. The approach is deliberately staged: first synthetic and de-identified retrospective cases; then prospective silent-mode testing alongside the team; then supervised use under mandatory surgeon sign-off; and only then can a specific skill, in a specific pathway and context, move toward certified local use through the harness.
We are seeking selected clinical, hospital, academic, and technology partners to validate ORTHO-X OrthoFlow™ in bounded pathways — for example a hip-fracture workflow, orthopaedic smart second opinions, an arthroplasty pathway, postoperative wound and recovery monitoring, registry and post-market evidence capture, hospital QA and M&M review, or structured teaching and assessment.
An honest word, and a new category
ORTHO-X OrthoFlow™ is decision-support and educational tooling — not a medical device, not medical advice — and like the literature it builds on, it is at an architecture-and-certification stage: the direction is endorsed by the field’s leaders; the clinical results are earned, per deployment, through staged, evidence-gated pilots. We think that honesty is the point. The hospitals and partners worth working with want the guardrails as much as the capability.
What this adds up to is a new category — Agentic Orthopaedics: not autonomous surgery, not unsupervised medical advice, not another chatbot, but a structured, auditable, human-led agentic layer for better orthopaedic care, better hospital workflows, better education, and better evidence.
The KSSTA authors closed by calling engagement with these technologies “no longer a niche interest but a professional necessity.” We agree — and we have turned the concept into something you can install, test, and pilot.
A limited open version is available at https://github.com/MAIVAN-ai/OrthoSkills-plugin.
ORTHO-X OrthoFlow™ lives as a working, installable package in our private partner repository, provisioned to hospitals and partners under evaluation.
🩺 Educational / decision-support reference only — not medical advice. Surgeon sign-off required for every output.
Talk to us
Pilots · Deployment & integration partnerships (e.g. Salesforce Health Cloud/ MCP) · ORTHO-X Data Commons Cooperative membership
MAIVAN.ai PBC by Bluenaut Matching Services AG — Kilchberg/Zurich, Switzerland
Building the Agentic Orthopaedics layer for surgeon-governed clinical intelligence.