Ask anyone who has lived through a traditional healthcare technology implementation what they remember, and you'll hear the same story: months of requirements gathering, a workflow specification thick enough to stop a door, a go-live date that slipped twice, and a system that, on day one, already didn't quite match how the clinic actually worked.

‍

That model made sense for the technology it was built around. An IVR tree or a scripted phone workflow is rigid by nature. Every branch, every prompt, every transfer rule had to be designed upfront, because the system could only do exactly what it was told. So implementation became an exercise in trying to predict every path a patient might take, before a single patient had taken one. Get it wrong, and you were back in the change-order queue.

‍

AI agents don't work that way. And implementing them the old way wastes everything that makes them valuable.

From scripting the path to defining the destination

The fundamental shift is this: traditional systems were built around paths. An AI workforce is built around outcomes.

‍

When the Access Agent answers a call, we haven't scripted the conversation. We've defined the goal (get this patient rescheduled, get this refill request captured and routed, get this caller to the right person with full context) along with the guardrails and escalation rules that keep every interaction safe. The agent finds its own way to the outcome, the way a skilled team member would, adapting to how each patient actually talks rather than forcing the patient through a phone tree.

‍

Implementation has to mirror that. Instead of specifying every branch upfront, we spend our time on the things that actually determine success: understanding how the clinic operates today, who handles what, where the operational exceptions live, and what "done" looks like for each workflow. The structure isn't gone. There are specific milestones and clear goals at every stage. But the structure is built around outcomes, not scripts. Fixed milestones. Flexible paths. That's the whole philosophy in four words.

‍

This is the same conclusion the broader industry is reaching. McKinsey's research on agentic AI deployments puts it plainly: onboarding agents is more like hiring a new employee than deploying software. Harvard Business Review now recommends literally creating an onboarding plan for AI agents, complete with a job description, a training period, and ongoing feedback. We agree. We've just been operating that way long enough to know what it looks like in a live clinic.

Live in weeks, not quarters

The second shift is speed, and it's not speed for its own sake.

‍

A traditional implementation took three to six months because the system had to be complete before it could be useful. An AI workforce inverts that. A deliberately focused first release, covering the highest-frequency, lowest-risk workflows handled end to end, can be live and creating value for patients in three to four weeks. We've taken customers from kickoff to go-live in 24 days. 

‍

Why start narrow? Because the fastest way to learn a clinic's real call patterns, the unwritten rules your staff carry in their heads, the exceptions no requirements document ever captures, is to get on the phones. Every early call teaches us something a discovery workshop never could. We earn the harder calls by nailing the simple ones first.

Implementation never really ends, and that's the point

Here's the comparison we use internally, and it's the one that changes how customers think about the whole engagement:

‍

If you hired a great new patient access coordinator, you wouldn't expect them to handle every call type on day one. You'd get them productive on the core work in their first few weeks. Then you'd keep going: adding skills, refining how they handle edge cases, reviewing the tough calls together, giving feedback, expanding their responsibilities as they earn them.

‍

That is exactly how we run the Access Agent. Go-live isn't the finish line; it's the end of week one of a very good hire's career. Every week after, we review real calls and real metrics with our customers (job completion, transfer drivers, where patients got stuck) and we make changes: a refined prompt here, a new workflow turned on there, an escalation threshold tuned for the patient population actually calling. Then we measure whether it worked, and we go again.

‍

That weekly loop is the product as much as the agent is. It's why implementation and ongoing success aren't two phases separated by a handoff at Health Note. They're one continuous motion, run by one team, against one set of goals. The clinics that see the best results aren't the ones with the most elaborate launch plans. They're the ones in the loop with us, watching the agent get better every week the way a strong new hire does.

What this means if you're evaluating AI for patient access

Three questions worth asking any vendor, including us:

‍

How long until the first version is live and handling real patient calls? If the answer is measured in quarters, you're buying the old model with new branding.

‍

What happens in the eight weeks after go-live? If the answer is "support tickets," you're buying software. If the answer is a structured cadence of measurement, refinement, and expanding scope, you're onboarding a workforce.

‍

And when our workflows don't match the spec, what breaks? Because they won't match; no clinic's reality matches its documentation. The right answer is: nothing breaks, because the system was built to pursue your outcomes, not your org chart circa the discovery phase.

‍

Every patient reaches care. That's the outcome we hold ourselves to, and it's why we'd rather have an agent live in three weeks, learning from real patients and getting measurably better every week, than a perfect specification that's six months from answering its first call.

‍

Health Note's Access Agent answers every call, books visits, and routes safely. Live with customers in weeks. See how implementation works: Request a demo.

‍

Jake Sirus is Head of Customer Operations at Health Note, where he helps healthcare organizations turn AI from a promising technology into a practical, measurable workforce. With 15 years across Epic enterprise implementations, virtual care, and operational transformation, including multimillion-dollar deployments at some of the country's largest health systems, Jake has a rare vantage point on what separates AI that demos well from AI that actually delivers.At Health Note, his work centers on designing scalable implementation models, translating clinic workflows into outcome-driven AI processes, and building feedback loops that make systems better every week, all while giving healthcare teams more capacity to focus on patients.

‍

‍

‍