Prepare
Gather the relevant context, normalize the inputs, complete safe work, and surface uncertainty before it becomes someone else’s problem.
Healthcare AI operations
Make AI useful where the real work happens.
I design AI-enabled operating systems around healthcare workflows, starting with the administrative and financial work that slows behavioral health practices and networks: clinical notes, eligibility, claim preparation, and follow-up. Software handles agreed preparation and checks, people keep consequential judgment, and every outcome leaves evidence your team can inspect.
22 to 17.5 minutesCharting time across all encounters at One to One Health
$1M+Claims volume across submission, remittance, and payment in six months
Human-ledPeople keep control of high-impact actions
$1M ARRTheraMatch and TPN Match reached $1 million in ARR and more than 1 million lives under care in less than 12 months.
Ask what needs to change
Do not ask where to add AI. Ask which work the organization should be able to do reliably.
A model matters, but it is rarely the operating system. Lasting value comes from the context, rules, interfaces, approvals, feedback, and ownership around it.
Decide
Put the recommendation, evidence, policy, consequence, and single human decision in one place.
Carry forward
Execute the next allowed action, preserve the decision history, measure the outcome, and improve the workflow without losing control.
What we design
Build around how people will use AI.
A credible AI workflow includes what happens when the model is right, uncertain, unavailable, or outside its authority.
Workflow and control
- Event, context, action, and human-decision mapping
- Deterministic rules and model boundaries
- Role-based approval and escalation paths
- Audit, feedback, and safe retry behavior
Evidence and production
- Outcome, quality, latency, and cost measures
- Evaluation sets and failure-mode testing
- Data, privacy, integration, and vendor choices
- Pilot design and production-readiness plan
How it starts
Choose one operational burden worth changing.
Map what people do today
Follow the inputs, exceptions, decisions, handoffs, risk, time, and downstream cost before proposing automation.
Define safe system authority
Separate deterministic work, model-assisted work, required human judgment, and actions the system must not take.
Run a measurable pilot
Test the full operating loop with real users and explicit success, failure, trust, and adoption measures.
Applied perspective
From clinical AI to what I’m building now.
At One to One Health, I developed the initial AI infrastructure for provider SOAP notes. Charting time fell from 22 to 17.5 minutes across all encounters while maintaining or improving both clinician and patient satisfaction. Claims Native applies bounded automation and explicit human gates to small-practice revenue work. CraftingCopy.ai connects generation to a governed brand operating model.
Each example starts with a different workflow, but the discipline is the same: make the outcome visible, preserve the right human authority, and test the system where it can fail.
Common questions
What practical buyers ask first.
What is an AI operating system in healthcare?
It is a governed workflow in which software gathers context, performs bounded work, presents clear evidence, asks people only for consequential judgment, and records what happened. Aim for dependable capacity to handle specific tasks.
Where should a healthcare company start with AI?
Start with a costly, repeated workflow whose inputs, allowed actions, human decisions, and measurable outcome can be described. Avoid beginning with a model or a broad AI mandate.
Can AI remain useful without acting autonomously?
Yes. Many valuable workflows automate preparation, checking, routing, summarization, and follow-up while reserving clinical, financial, compliance, or release decisions for the appropriate person.
Start with the workflow
Bring one task your team keeps repeating.
If you already know the workflow, my agent workflow pilot builds one agent around sources your team trusts and measures time saved against the effort of reviewing its output. If you are still choosing the workflow, start with the two-week assessment.
Related reading
Related reading on AI in healthcare operations.
Guides for practices and essays for founders and operators, gathered in one place.