AI Evals: What the Checkmarks Actually Prove
Your team ran evals. Every test passed. The results look like the QA matrix you've signed off on a hundred times. They're not. Here's what AI evals actually catch, and where they go silent.
Yoram Friedman, MD
(102)Physician and enterprise product leader. Author of the Agentic AI for Product Leaders series. Writes on AI, healthcare, and what changes when software starts acting. 15+ years at SAP, Walmart, and regulated health tech.
You Can't Measure What You Didn't Design
After launch, trust builds naturally and supervision erodes naturally. If the product wasn't designed to hold oversight stable, the agent ends up at an autonomy level nobody authorized. You can't measure what you didn't design.
Change Management for Agentic AI - Designing the Human Side of the Transition
Most change management programs treat agentic AI as an adoption problem. The diagnosis is wrong. Change management is being treated as adoption when the real work is supervision. That is a different skill, and most teams are not designing for it.
Silent Degradation: What a Deployed Clinical AI Looks Like at Month Eighteen
Your agent shipped. Your team moved on. Silent degradation is what happens next, and nobody is watching for it. The most-deployed clinical AI in American hospitals ran at half its advertised accuracy for six years before anyone audited it. Three disciplines before the next ship.
What the Next Generation of Physicians Won't Remember
I don't remember much from school. My memory starts the day I entered clinical training. Medicine is learned by burning encounters into memory. AI is very good at removing that weight. The weight was the learning.
Your Agent Worked. Your Users Bypassed It.
Your AI worked in the lab. Your users bypassed it in the real world. The gap isn’t the model, it’s the environment. Without the right incentives, workflows, and accountability, even the best-designed AI gets reduced to output generation. The system around it determines everything.
Who Owns the Patient Relationship Now?
The physician was the gatekeeper. The hospital was the hub. AI changed both. Now Amazon, Google, and OpenAI are racing to own what comes next. The patient relationship is the prize, and the bidding has started.
The Parallel Health System
Patients aren't waiting for the healthcare system to catch up. They have wearables, direct-access labs, referral-free MRIs, and AI interpreting all of it. The parallel system is already running."
Patients Are Not Waiting for Permission
She arrives with a plan her AI already helped her build. The physician now has two choices: become a trusted continuum who adds what AI cannot, or become a friction point blocking a plan she already made. Only one of those sustains the relationship.
The Catch-22 of Business AI
Enterprise AI has a structural catch-22: context lives where you cannot run agents, and compute lives where context does not exist. Move the data and you lose the meaning. That gap is why most deployments produce outputs that are technically impressive and operationally thin.
The Patients Healthcare AI Was Not Built For, Are Now Its Most Frequent Users
1 in 3 Americans uses AI for health advice. The heaviest users are uninsured, low-income adults who cannot access a doctor. The AI they are relying on was built on data from patients who look nothing like them. That is not an equity talking point. It is a product failure.
Prompt Engineering Is a Temporary Skill
A developer, an accountant, a graphic designer, a film director, a composer, and a product manager all use the exact same interface to communicate with AI: a text box. That has never been true of mature technology. It will not stay true for this one.