Report
Beyond the System of Record
By Zoe Goldman
How AI could shift healthcare compliance from documenting the work to supporting it while it happens.
In a survey of 1,095 nurses at a large U.S. health system, nurses estimated that 43% of their shift was spent documenting and that 45% of that documentation was duplicative or unnecessary.
Walk into almost any medication room or staff office and you're likely to see some version of the same sign:
“But did you document it?”
It is funny because everyone recognizes the joke — and revealing because it captures how thoroughly documenting the work has become conflated with doing the work. EHRs have made the clinical record more legible, searchable and auditable. But their fundamental role is to preserve an authoritative record of what happened, not to serve as the real-time operating system for everything that needs to happen next.
The pressure on staff is therefore two-sided: document more thoroughly because the record may later become evidence, but do not allow documentation to take time away from the work the record is meant to represent.
But creating a comprehensive electronic record does not guarantee that every required event is consistently captured in the right place or carried through the right workflow. In 2025, the HHS Office of Inspector General found that nursing homes failed to report 43% of falls involving major injury and hospitalization in the resident assessments used to calculate federal quality measures.
The events happened. Residents went to the hospital. Claims existed. Yet the information still did not reliably make it through the required workflow.
That is not simply a documentation problem. It is an execution problem, and it is where AI’s larger opportunity begins.
Report
Beyond the System of Record
By Zoe Goldman
How AI could shift healthcare compliance from documenting the work to supporting it while it happens.
In a survey of 1,095 nurses at a large U.S. health system, nurses estimated that 43% of their shift was spent documenting and that 45% of that documentation was duplicative or unnecessary.
Walk into almost any medication room or staff office and you're likely to see some version of the same sign:
“But did you document it?”
It is funny because everyone recognizes the joke — and revealing because it captures how thoroughly documenting the work has become conflated with doing the work. EHRs have made the clinical record more legible, searchable and auditable. But their fundamental role is to preserve an authoritative record of what happened, not to serve as the real-time operating system for everything that needs to happen next.
The pressure on staff is therefore two-sided: document more thoroughly because the record may later become evidence, but do not allow documentation to take time away from the work the record is meant to represent.
But creating a comprehensive electronic record does not guarantee that every required event is consistently captured in the right place or carried through the right workflow. In 2025, the HHS Office of Inspector General found that nursing homes failed to report 43% of falls involving major injury and hospitalization in the resident assessments used to calculate federal quality measures.
The events happened. Residents went to the hospital. Claims existed. Yet the information still did not reliably make it through the required workflow.
That is not simply a documentation problem. It is an execution problem, and it is where AI’s larger opportunity begins.
Beyond the System of Record
By Zoe Goldman
How AI could shift healthcare compliance from documenting the work to supporting it while it happens.
In a survey of 1,095 nurses at a large U.S. health system, nurses estimated that 43% of their shift was spent documenting and that 45% of that documentation was duplicative or unnecessary.
Walk into almost any medication room or staff office and you're likely to see some version of the same sign:
“But did you document it?”
It is funny because everyone recognizes the joke — and revealing because it captures how thoroughly documenting the work has become conflated with doing the work. EHRs have made the clinical record more legible, searchable and auditable. But their fundamental role is to preserve an authoritative record of what happened, not to serve as the real-time operating system for everything that needs to happen next.
The pressure on staff is therefore two-sided: document more thoroughly because the record may later become evidence, but do not allow documentation to take time away from the work the record is meant to represent.
But creating a comprehensive electronic record does not guarantee that every required event is consistently captured in the right place or carried through the right workflow. In 2025, the HHS Office of Inspector General found that nursing homes failed to report 43% of falls involving major injury and hospitalization in the resident assessments used to calculate federal quality measures.
The events happened. Residents went to the hospital. Claims existed. Yet the information still did not reliably make it through the required workflow.
That is not simply a documentation problem. It is an execution problem, and it is where AI’s larger opportunity begins.
AI is Moving Upstream
One of the first major applications of generative AI in healthcare has been helping clinicians create the record more efficiently.
The results are promising. In one health-system evaluation, ambient AI reduced average time in notes from 6.2 to 5.3 minutes per appointment. The percentage of clinicians who said they could give patients their undivided attention rose from 57.9% to 93%.
That alone matters.
But writing the record faster may be only the beginning.
There is also an emerging tension as AI moves directly into the EHR.
A 2026 npj Digital Medicine perspective warned that generative AI can increasingly blend AI-generated and human-generated content inside the clinical record. The authors argue that healthcare will need better traceability: the ability to distinguish AI-generated text from human-written content, understand what information informed an AI recommendation and preserve the provenance of what ultimately appears in the record.
That is not an argument against AI in the EHR. AI can make documentation substantially easier.
It is an argument for being very clear about what different systems are responsible for.
When the EHR is the authoritative record used for clinical care, regulation, billing and potentially litigation, who observed something, what technology generated and what a human ultimately approved matter.
AI’s most interesting opportunity may therefore be partly upstream of the record.
The System of Record and the System of Action
AI can do more than translate a conversation into a note.
It can connect an observation to what needs to happen next: bringing together relevant context, identifying required follow-up, getting information to the appropriate person and keeping unfinished work visible.
That points to a distinction healthcare technology has largely blurred:
The EHR is the system of record. AI is enabling a system of action.
The EHR preserves the official account of what was assessed, decided, completed and documented. A system of action has a different job:
Observation → relevant context → owner → action → escalation → completion → record
Its purpose is not to replace the EHR or human judgment. It is to help ensure that the work the EHR will eventually document actually happens.
That distinction may become increasingly important as AI improves. The same technology that can understand fragmented information across a resident’s day can help determine what deserves attention, who needs to know and what remains unresolved—without requiring every intermediate thought, reminder or workflow step to become part of the permanent clinical record.
From Context to Action
Consider a caregiver who notices a meaningful change in a resident’s behavior.
Instead of relying on that observation to survive a verbal handoff or disconnected communication log, an intelligent operational layer could connect it with relevant history, surface the appropriate next step, route it to the right staff member and keep the follow-up visible across shifts. Once the appropriate action has occurred, the final human-reviewed documentation still belongs in the EHR.
In that model, the record becomes the receipt for a process the technology helped the team complete.
And there is reason to believe technology can influence the work itself, not simply its documentation. A systematic review of 35 computerized decision-support studies found improvements in 47% of measured care-process indicators and 40.7% of patient-care outcome indicators, including adherence to guidance, timely blood sampling, falls and pressure ulcers. The evidence was heterogeneous, but the larger point is important: technology can change what happens, not only what gets recorded afterward.
What This Means for the Industry
Healthcare organizations may need to stop expecting one system to simultaneously function as the legal record, communication network, task manager, policy interpreter and real-time operating system.
Those are different jobs.
And as AI becomes embedded throughout healthcare, separating those jobs may become more valuable, not less.
The important architectural question may no longer be simply:
How much AI can we put inside the EHR?
It may also be:
What should happen before information becomes part of the EHR?
A complementary operational layer could connect information, responsibility and follow-through while preserving a cleaner boundary around the authoritative record.
That would also change how organizations evaluate technology. Instead of measuring success primarily through completed fields or generated notes, leaders could ask:
Did the right person receive the information?
Did the required action happen on time?
Can we trace the path from observation to response?
Compliance teams could move from reconstructing failures after the fact toward identifying unfinished pathways while there is still time to intervene.
And frontline staff could receive compliance support as part of doing the work rather than as another documentation task added after it.
AI is Moving Upstream
One of the first major applications of generative AI in healthcare has been helping clinicians create the record more efficiently.
The results are promising. In one health-system evaluation, ambient AI reduced average time in notes from 6.2 to 5.3 minutes per appointment. The percentage of clinicians who said they could give patients their undivided attention rose from 57.9% to 93%.
That alone matters.
But writing the record faster may be only the beginning.
There is also an emerging tension as AI moves directly into the EHR.
A 2026 npj Digital Medicine perspective warned that generative AI can increasingly blend AI-generated and human-generated content inside the clinical record. The authors argue that healthcare will need better traceability: the ability to distinguish AI-generated text from human-written content, understand what information informed an AI recommendation and preserve the provenance of what ultimately appears in the record.
That is not an argument against AI in the EHR. AI can make documentation substantially easier. It is an argument for being very clear about what different systems are responsible for.
When the EHR is the authoritative record used for clinical care, regulation, billing and potentially litigation, who observed something, what technology generated and what a human ultimately approved matter.
AI’s most interesting opportunity may therefore be partly upstream of the record.
The System of Record and the System of Action
AI can do more than translate a conversation into a note.
It can connect an observation to what needs to happen next: bringing together relevant context, identifying required follow-up, getting information to the appropriate person and keeping unfinished work visible.
That points to a distinction healthcare technology has largely blurred:
The EHR is the system of record. AI is enabling a system of action.
The EHR preserves the official account of what was assessed, decided, completed and documented. A system of action has a different job:
Observation → relevant context → owner → action → escalation → completion → record
Its purpose is not to replace the EHR or human judgment. It is to help ensure that the work the EHR will eventually document actually happens.
That distinction may become increasingly important as AI improves. The same technology that can understand fragmented information across a resident’s day can help determine what deserves attention, who needs to know and what remains unresolved—without requiring every intermediate thought, reminder or workflow step to become part of the permanent clinical record.
AI is Moving Upstream
One of the first major applications of generative AI in healthcare has been helping clinicians create the record more efficiently.
The results are promising. In one health-system evaluation, ambient AI reduced average time in notes from 6.2 to 5.3 minutes per appointment. The percentage of clinicians who said they could give patients their undivided attention rose from 57.9% to 93%.
That alone matters.
But writing the record faster may be only the beginning.
There is also an emerging tension as AI moves directly into the EHR.
A 2026 npj Digital Medicine perspective warned that generative AI can increasingly blend AI-generated and human-generated content inside the clinical record. The authors argue that healthcare will need better traceability: the ability to distinguish AI-generated text from human-written content, understand what information informed an AI recommendation and preserve the provenance of what ultimately appears in the record.
That is not an argument against AI in the EHR. AI can make documentation substantially easier. It is an argument for being very clear about what different systems are responsible for.
When the EHR is the authoritative record used for clinical care, regulation, billing and potentially litigation, who observed something, what technology generated and what a human ultimately approved matter.
AI’s most interesting opportunity may therefore be partly upstream of the record.
The System of Record and the System of Action
AI can do more than translate a conversation into a note.
It can connect an observation to what needs to happen next: bringing together relevant context, identifying required follow-up, getting information to the appropriate person and keeping unfinished work visible.
That points to a distinction healthcare technology has largely blurred:
The EHR is the system of record. AI is enabling a system of action.
The EHR preserves the official account of what was assessed, decided, completed and documented. A system of action has a different job:
Observation → relevant context → owner → action → escalation → completion → record
Its purpose is not to replace the EHR or human judgment. It is to help ensure that the work the EHR will eventually document actually happens.
That distinction may become increasingly important as AI improves. The same technology that can understand fragmented information across a resident’s day can help determine what deserves attention, who needs to know and what remains unresolved—without requiring every intermediate thought, reminder or workflow step to become part of the permanent clinical record.
From Context to Action
Consider a caregiver who notices a meaningful change in a resident’s behavior.
Instead of relying on that observation to survive a verbal handoff or disconnected communication log, an intelligent operational layer could connect it with relevant history, surface the appropriate next step, route it to the right staff member and keep the follow-up visible across shifts. Once the appropriate action has occurred, the final human-reviewed documentation still belongs in the EHR.
In that model, the record becomes the receipt for a process the technology helped the team complete.
And there is reason to believe technology can influence the work itself, not simply its documentation. A systematic review of 35 computerized decision-support studies found improvements in 47% of measured care-process indicators and 40.7% of patient-care outcome indicators, including adherence to guidance, timely blood sampling, falls and pressure ulcers. The evidence was heterogeneous, but the larger point is important: technology can change what happens, not only what gets recorded afterward.
What This Means for the Industry
Healthcare organizations may need to stop expecting one system to simultaneously function as the legal record, communication network, task manager, policy interpreter and real-time operating system.
Those are different jobs.
And as AI becomes embedded throughout healthcare, separating those jobs may become more valuable, not less.
The important architectural question may no longer be simply:
How much AI can we put inside the EHR?
It may also be:
What should happen before information becomes part of the EHR?
A complementary operational layer could connect information, responsibility and follow-through while preserving a cleaner boundary around the authoritative record.
That would also change how organizations evaluate technology. Instead of measuring success primarily through completed fields or generated notes, leaders could ask:
Did the right person receive the information?
Did the required action happen on time?
Can we trace the path from observation to response?
Compliance teams could move from reconstructing failures after the fact toward identifying unfinished pathways while there is still time to intervene.
And frontline staff could receive compliance support as part of doing the work rather than as another documentation task added after it.
From Context to Action
Consider a caregiver who notices a meaningful change in a resident’s behavior.
Instead of relying on that observation to survive a verbal handoff or disconnected communication log, an intelligent operational layer could connect it with relevant history, surface the appropriate next step, route it to the right staff member and keep the follow-up visible across shifts. Once the appropriate action has occurred, the final human-reviewed documentation still belongs in the EHR.
In that model, the record becomes the receipt for a process the technology helped the team complete.
And there is reason to believe technology can influence the work itself, not simply its documentation. A systematic review of 35 computerized decision-support studies found improvements in 47% of measured care-process indicators and 40.7% of patient-care outcome indicators, including adherence to guidance, timely blood sampling, falls and pressure ulcers. The evidence was heterogeneous, but the larger point is important: technology can change what happens, not only what gets recorded afterward.
What This Means for the Industry
Healthcare organizations may need to stop expecting one system to simultaneously function as the legal record, communication network, task manager, policy interpreter and real-time operating system.
Those are different jobs.
And as AI becomes embedded throughout healthcare, separating those jobs may become more valuable, not less.
The important architectural question may no longer be simply:
How much AI can we put inside the EHR?
It may also be:
What should happen before information becomes part of the EHR?
A complementary operational layer could connect information, responsibility and follow-through while preserving a cleaner boundary around the authoritative record.
That would also change how organizations evaluate technology. Instead of measuring success primarily through completed fields or generated notes, leaders could ask:
Did the right person receive the information?
Did the required action happen on time?
Can we trace the path from observation to response?
Compliance teams could move from reconstructing failures after the fact toward identifying unfinished pathways while there is still time to intervene.
And frontline staff could receive compliance support as part of doing the work rather than as another documentation task added after it.
From Proving Compliance to Supporting it
For decades, healthcare technology has become increasingly sophisticated at preserving evidence of what happened.
AI creates an opportunity to become equally sophisticated about what needs to happen next.
That is the larger transition:
For decades, healthcare used technology to prove compliance after the work. AI can help support compliance while the work is happening.
The future is not one with less documentation, less accountability or less human oversight.
It is one in which the systems responsible for action and the systems responsible for record work together without necessarily being the same thing.
The strongest systems will not simply create more convincing evidence that the right work happened.
They will help make the right work harder to miss.
For decades, healthcare technology has become increasingly sophisticated at preserving evidence of what happened.
AI creates an opportunity to become equally sophisticated about what needs to happen next.
That is the larger transition:
For decades, healthcare used technology to prove compliance after the work. AI can help support compliance while the work is happening.
The future is not one with less documentation, less accountability or less human oversight.
It is one in which the systems responsible for action and the systems responsible for record work together without necessarily being the same thing.
The strongest systems will not simply create more convincing evidence that the right work happened.
They will help make the right work harder to miss.
For decades, healthcare technology has become increasingly sophisticated at preserving evidence of what happened.
AI creates an opportunity to become equally sophisticated about what needs to happen next.
That is the larger transition:
For decades, healthcare used technology to prove compliance after the work. AI can help support compliance while the work is happening.
The future is not one with less documentation, less accountability or less human oversight.
It is one in which the systems responsible for action and the systems responsible for record work together without necessarily being the same thing.
The strongest systems will not simply create more convincing evidence that the right work happened.
They will help make the right work harder to miss.