- CIOsurge
- Posts
- Three questions that expose whether an AI vendor belongs in your environment
Three questions that expose whether an AI vendor belongs in your environment
Plus: agent sprawl reaches a breaking point and why AI now demands infrastructure-level governance.
Zack Tembi
April 20, 2026
Three questions that expose whether an AI vendor belongs in your environment
Powered by Single Fin
Welcome to this week’s edition of CIOsurge!
This week:
Another amazing conversation with Lloyd Fobi. The key takeaway was a three-question framework that consistently separates credible AI vendors from well-funded noise, and it starts with who actually owns the data.
Salesforce, Databricks, and AWS all shipped agentic AI governance tools this week as enterprises running multiple agent pilots begin hitting a predictable wall, and the governance question is now as much a FinOps problem as a security one.
AI has followed the same adoption arc as cloud computing in roughly a quarter of the time, and most enterprises are still governing it like a software project, a category error with consequences that compound.
Let’s make this week a game-changer.
Stay sharp. Stay ahead.
💡 Guest Expert Insights: Lloyd Fobi
Most AI vendor evaluations go wrong before the demo ends.
I recently spoke with Lloyd Fobi, founder of Health Scale Partners and a longtime healthcare CIO and COO advisor.
He shared a framework he uses when evaluating AI vendors, three questions that consistently separate credible partners from well-funded noise.
The first is about data ownership. Do you maintain control of your data when using this product?
And can the vendor show their model performing against a diverse, independent dataset, not the one they curated to make the demo look clean?
Any vendor who cannot answer that directly has already told you something important.
The second goes deeper than a standard reference check. Not who is using this tool, but who like you is using it. For example, payer mix, patient population, system size, and volume all change how a model behaves in practice. A tool built on private payer data does not automatically transfer to a Medicare-heavy system. Forcing that specificity in the conversation tells you immediately whether the vendor understands your environment or is selling a category.
The third is the one that matters most.
Will they share risk with you?
If a vendor believes their technology will move your outcomes, they should be willing to tie compensation to results. The ones who walk away from that conversation have answered the question for you. The ones who lean in may have just found you a real partner.
Taken together, these three questions establish data integrity, ground performance claims in comparable environments, and put the vendor's conviction on the line. Any partner worth the relationship should be able to answer all three.
- Zack Tembi
Enterprises are running agents faster than they can govern them
Salesforce, Databricks, and AWS all released agentic AI governance features this week as enterprises piloting multiple agent products begin running into a predictable problem. The issue has a name: agent sprawl. Organizations running several agentic pilots in parallel are generating fragmented deployments, inconsistent permissions, and cost structures that are difficult to track. The launches are a direct response.
The hard truth is that governance was not part of the original agentic conversation. Speed was. Now that agents are executing multi-step workflows and touching sensitive data across systems, the controls that should have been built in from the start are being retrofitted after the fact.
For CIOs, architecture fit matters more than feature comparison. A governance layer that already lives inside a platform you run costs less to implement and delivers control faster than one that requires integration work before it adds any value. The other thing to lock down early is cost visibility. Any vendor positioning itself as your agentic control plane needs to show you a unified cost ledger across every agent, model, and tool in scope. If they cannot produce that picture, they are not actually governing the environment.
- Zack Tembi
Opinion: AI has crossed the infrastructure threshold. Most enterprise operating models have not.
AI has followed the same adoption arc as databases, networks, and cloud computing, but in roughly a quarter of the time.
It is now embedded in customer-facing processes, internal operations, compliance workflows, and competitive positioning simultaneously. Despite that, most enterprises still classify AI spending under software or R&D budgets, manage it through ad hoc working groups, and lack formal frameworks for model drift, vendor dependency, and data provenance.
The way most organizations are governing AI has not kept pace with how deeply it has embedded itself into operations. AI is no longer a tool that sits alongside your core systems. In many organizations, it already is a core system, and managing it otherwise is a category error with consequences that compound over time.
Moving AI investment into capital and infrastructure budgets, building governance functions with real operational accountability, and investing in the MLOps infrastructure that keeps models reliable are all part of the same shift.
The organizations that built genuine cloud-native capabilities ahead of the curve ended up with advantages that proved very hard to close. AI is following the same dynamic, and the window for deliberate positioning is narrowing.
- Zack Tembi
🗞️ At A Glance
💡 CIO Spotlights
Premier Health Appoints Margaret Lozovatsky, MD as system’s next Chief Digital Information Officer
Margaret Lozovatsky, MD joins Premier Health as Chief Digital Information Officer effective May 4, leading digital strategy, clinical informatics, and AI integration across the health system.
Former VP of Digital Health Innovations at the AMA and senior informatics leader at Novant Health, BJC HealthCare, and Cedars-Sinai, Lozovatsky brings 15+ years advancing clinical technology and national AI governance strategy.
COO Dr. Chad Whelan positions her as central to Premier Health's next chapter as the system deepens its academic affiliation with Wright State University and accelerates its clinical innovation agenda.
Autodesk appoints Mike Kelly as Chief Information Officer
Mike Kelly joins Autodesk as Chief Information Officer effective April 13, tasked with leading enterprise technology strategy, AI adoption, and the digital employee experience.
Former Operating Partner and first-ever CIO at Andreessen Horowitz and previous CIO at both Red Hat and McKesson, Kelly brings a strong track record across AI, cybersecurity, and data and analytics.
CEO Steve Blum positions the hire as a proactive move to build the execution foundation Autodesk needs to accelerate AI-driven outcomes as the company enters its next phase of growth.
Reply