Pre-Visit Chart Prep: EHR AI vs Custom Agent vs Manual

By Jude Lee · · Comparison

Medical assistant and office manager reviewing the day's patient schedule on a computer at a small clinic

The agent wave is real — and it’s arriving at large health systems first

In late 2025, Fierce Healthcare reported that Epic is expanding into an agent platform with Cosmos-powered predictions and deeper workflow automation, and Portal ERP reported that Oracle Health expanded its Clinical AI Agent with automated coding, dictation and chart review. Both are meaningful signals about where the category is going.

They are also, for most readers of this site, news about somebody else’s EHR. Independent primary care, specialty, dental and behavioral-health practices mostly run athenahealth, eClinicalWorks, Tebra, SimplePractice, Open Dental, Dentrix or similar. Enterprise agent features get demoed long before an equivalent shows up in your release notes — and when it does, it may be limited to modules you didn’t buy.

So the practical question isn’t “is Epic building agents.” It’s: for the specific job of getting a chart ready before a visit, what should a ten-person practice actually do this quarter?

What “chart prep” actually contains

Before comparing tools, write down the job. In most practices it’s a bundle of unglamorous lookups a medical assistant, hygienist or front-desk lead does the afternoon before, or at 7:40am:

Roughly half that list is administrative and lives in your practice management system, not the clinical chart. That split matters enormously for which option fits.

Four options, compared honestly

Option 1: your EHR vendor’s AI. Biggest advantage by far: the data is already there, and you’re presumably already under a business associate agreement with that vendor. No new integration, no new PHI pathway, no new security review. Disadvantages: you get their opinion of what chart prep means, you often can’t tune it, and it usually can’t see anything outside the EHR — your fax inbox, your standalone imaging software, your separate payment platform. Evaluate it like any other module; we walk through the questions in our guide to evaluating your EHR vendor’s AI agents.

Option 2: a point tool. This is the document-AI, fax-ingestion and pre-charting category: vendors that ingest inbound records, extract structured data and drop a summary or discrete values into the chart. Good ones are genuinely strong at OCR and classification — a job general-purpose models still fumble on bad fax scans. Shortlist them on three questions: does the vendor already have a documented, live integration with your specific ambulatory or dental system (several do, which removes the hardest part of the work); does it write structured data back into the chart or just attach a PDF; and does pricing scale per page, per seat or per provider? The cost is another vendor, another BAA and another integration to maintain.

Option 3: a custom agent connected through MCP. The Model Context Protocol is an open standard for giving an AI assistant governed access to specific tools and data. In practice you build a small MCP server exposing a handful of read-only functions over your systems — get_todays_schedule, get_results_since_last_visit, get_eligibility_status, get_outstanding_forms — then define a skill: a packaged, reusable instruction set telling the assistant exactly how to assemble a prep sheet, in the same order, every time. Step zero is unglamorous and non-negotiable: confirm a usable read API actually exists for your system, and find out what it costs. Many small-practice platforms — Dentrix, eClinicalWorks and SimplePractice at common tiers among them — either offer no general read API or gate it behind a partner/developer program with application review, certification and recurring fees. If that door is closed or priced out of reach, options 1, 2 and 4 are your real menu. If it’s open, the mechanics are covered in our Claude + EHR via MCP build guide.

Option 4: keep it manual, but on a checklist. Under-rated. A one-page standardized prep checklist, done by the same person at the same time daily, fixes a surprising share of the pain and costs nothing but discipline. If your prep is inconsistent because nobody defined it, automation will just make the inconsistency faster.

EHR-native or point tool
Fastest to switch on. No new PHI pathway if it’s your existing EHR vendor. Subscription-shaped cost — though EHR-native AI is often bundled into a suite or priced per provider at rates the vendor won’t publish, so you may not be able to isolate what the feature costs you. Fixed scope, limited tuning, blind to systems outside its own walls, and you wait on their roadmap.
Custom MCP agent + skill
Spans systems the EHR can’t see; you control exactly what data leaves and what the output looks like. But: real build and maintenance cost, you own the security review, API access may not exist, and you need a BAA with your model provider before any PHI moves.

Where the agent actually breaks

The American Medical Association’s piece on guardrails for fast-moving AI makes the point most operational leaders reach independently: deployment speed is not the constraint, governance is.

A chart-prep agent’s job is to make sure nobody walks into the room uninformed. It is not to decide what the information means.

PHI guardrails you can’t skip

Any path except pure-manual moves PHI somewhere. Non-negotiables, per HIPAA:

Modeling the value without making up numbers

Don’t accept anyone’s ROI headline, including ours. Build your own from inputs you can measure this month:

__ min
Prep time per chart today (time it for one week)
Your measurement
__ visits/day
Visits requiring prep
Your schedule
$__/hr
Loaded hourly cost of the person doing prep
Your payroll
__ min
Human review and correction time per drafted chart
Your pilot correction log
__ /month
Visits rescheduled or delayed for missing info
Your front-desk log

Monthly labor exposure today = (prep minutes ÷ 60) × visits per day × working days × loaded hourly rate. Model the automated case with the same formula but swap prep minutes for review minutes — the human review step the pilot below requires is a permanent line item, not a temporary one, and leaving it out is the most common way these models flatter themselves. Then add the second, usually larger bucket: visits that ran long, got rescheduled or triggered a same-day scramble because a result or auth wasn’t in the chart, multiplied by your average visit revenue. Subtract software cost plus the internal hours to build and babysit it.

Often the labor savings alone don’t justify a custom build, but the recovered visits and the reallocation of a good MA from clerical lookups to rooming and recall work do. Decide which you’re actually buying. Our framework for choosing between agents, RPA and plain rules applies: if the prep steps are deterministic, rules-based automation is cheaper and more reliable than an agent.

A 30-day pilot that won’t blow up

  1. Write the checklist first

    Document exactly what a perfect prep sheet contains, in order. If two staff members disagree, resolve it now. This becomes your skill definition later — and it’s valuable even if you automate nothing.

  2. Pick one visit type

    Established-patient follow-ups, or hygiene recalls. One provider. Not new patients or complex consults.

  3. Run assisted, not autonomous

    The agent drafts; a human reviews and posts. Log every correction and the minutes it took for two weeks. That log is your real accuracy data and your review-cost input.

  4. Decide on evidence

    If corrections are rare and confined to one category, fix that category and expand. If they’re scattered and unpredictable, the data access is probably the problem, not the model.

  5. Only then expand scope

    Add visit types, then providers, then the second system. Keep human review on anything clinical indefinitely.

As of 2026, enterprise EHR vendors are shipping increasingly capable chart-review agents, and that pressure will push similar features down into the systems independent offices actually run. Good reason to define your chart-prep standard now — so when the feature arrives, you can tell within a week whether it beats what you already do.

Not sure where to start?

Get a free automation audit: we map your scheduling, intake, insurance, billing, and patient communication and show you what's worth automating — before you spend a dollar.

Get a free automation audit

Related guides