Cartoons by Hasan Ahmad, DO
Visual takes on clinical informatics, AI governance, and health system strategy, originally shared on LinkedIn. 42 cartoons, newest first.
-
Pilots Are Not a Strategy
Pilots on their own aren't a strategy. Connecting them is.
-
Minutes Saved Is a Leading Indicator, Not the Outcome
Minutes saved is the number everyone reaches for, and it’s also the one that tells you the least about whether ambient AI is actually helping.
-
Chat Is an Interrupt. Govern It Like One.
You can override a pop-up without guilt. It's a lot harder to ignore a colleague.
-
Review the Stack, Not Just the Rule
It isn't bad alerts anymore, it's the stack.
-
Efficiency That Cannot Be Banked Will Be Spent as More Work
AI can hand a clinician back part of their day. The question nobody seems to be asking is whether the health system will let them keep it.
-
Brilliant AI + Clumsy EHR = Still a Bad Day
The clinician experiences the integration, not the algorithm.
-
Polished Drafts Can Hide Real Uncertainty
It's the draft that's politely wrong and weirdly confident.
-
Saying No Is Part of Good AI Stewardship
Turning things off isn't failure. It's stewardship.
-
AI Strategy Is a Signature Policy with Better Slides
So I've come to believe the signature question is the strategy.
-
Patients Are Choosing Their Own Labs. Primary Care Needs an On-Ramp.
Patients are starting to decide what they want measured.
-
Autonomy Without Ownership Is Just Speed
If an agent can close a loop the treating clinician never saw, you didn't automate care. You automated absence.
-
Capture Everything ≠ Communicate the Point
If the HPI takes longer to read than the visit took, we didn't automate documentation. We just automated a new kind of noise.
-
Leading Indicators First, Dollars Second
One for what clinicians feel by week two. One for what finance needs by month twelve.
-
The Best CDS Is Invisible Until It Matters
Make the correct thing the easy thing to do and build it into the workflow instead of adding a bolt-on alert.
-
Time Is Recovered. Thinking Must Be Protected.
But if the tool takes away the step where the thinking used to happen, the thinking doesn’t just relocate on its own.
-
If They Still Alt-Tab, It Isn’t Integrated
If your clinicians still alt-tab to reach your “integrated” tool, you didn’t buy integration. You bought a portal with a better logo.
-
Don’t Buy a Megaphone for a Dirty Chart
Here’s the thing I keep coming back to: in an EHR full of AI, the problem list is the prompt.
-
Agentic AI Without Defined Supervision Is Just Distributed Risk
If your organization is piloting agents, the supervision model deserves as much design time as the workflow itself.
-
Capability Without Accountability Is Just Automation with Liability
If we don’t get clear human accountability sorted out, these tools are just going to turn into fancier versions of “the computer said so.”
-
The Most Dangerous AI Failure Is the One That Never Looks Dramatic
Hallucinations make the headlines. Drift is what shows up in the outcomes.
-
The Tech Is Ready. Our Operating Models Mostly Aren’t.
Epic is solving the “build it” problem while quietly handing us the “keep it running” problem.
-
When AI Acts, Someone Must Own the Outcome
When the agent actually does something, who owns the outcome?
-
Who Ensures the Clinical Judgment Stays Visible?
When the AI becomes the main author of the note, whose clinical story is it anymore?
-
Hours Returned Mean Nothing If Decisions Don’t Improve
It’s the easiest number in the room. It’s also the least useful.
-
Patients Already Brought AI into the Room. Design for It.
Pretending it isn’t happening doesn’t make it go away.
-
Reconciling the Patient’s AI
Med rec taught us that naming the task is the first step to actually designing for it.
-
I Connected My Health Record to ChatGPT Health This Week
But access and understanding are not the same thing.
-
AI Technical Debt: Today’s “Helpful” Features… Tomorrow’s Headaches
It’ll accumulate one “helpful” feature at a time.
-
It Was About Time
A project has to earn a physician’s time, not consume more of it.
-
Nobody Ever Puts in a Ticket to Delete Something
If your intake process only has a form for additions, that’s not governance. That’s accumulation.
-
AI Wasn’t the Headline—It Was the Infrastructure
The best technology eventually becomes invisible.
-
Clinical AI Doesn’t Break. It Just Stops Being True.
Treat your models like antibiograms.
-
Will You Tie Your Fees to Those Numbers?
If your AI product works as well as your slide deck says, put it in the contract.
-
That Gap Has a Name: Shadow AI
Slow governance doesn't stop AI adoption. It just pushes it underground.
-
What If an AI Use Case Couldn’t Get Approved Unless It Fit on One Slide?
So the rule I'm toying with: if it can't fit on one slide, it isn't ready.
-
AI Governance Is Not Just About What We Launch… It’s Also About What We Stop
“What should we launch?” is the easy question. “What should we retire?” is the one that separates real governance from theater.
-
Shadow AI Is a Signal, Not a Scandal
But governance that only says “no” will always lose to a workflow that is painful enough.
-
AI Is Moving Faster Than Our Governance
It is that our governance model is still catching up.
-
Shared Risk Builds Real Trust
If an AI vendor will not share any of the risk, that tells you something.
-
From CDS to AI
Because an AI tool can be accurate and still fail.
-
Instant Orders Workflow Live at Parkview
In healthcare, automation only adds value when it is paired with clinical governance, workflow validation, and clear accountability.
-
Clinical AI Governance
Start with the workflow. Validate with the specialty. Govern at the enterprise.
