Table of Contents
📋 The Notes Are Everywhere. The Assessment Is Missing.
In 2021 I wrote about borrowing three kinds of inpatient records to manage projects. It’s still the most-read post on my blog. Five years later, the problem it set out to solve has flipped, so I wrote a sequel.
🏥 Quick recap: three medical records, three project notes
Admission note → project plan
Progress note → work log
Discharge summary → project retrospective
The pain point back then: when we succeeded, we didn’t know how; when we failed, we didn’t know why.
🩺 What I didn’t spell out in 2021: SOAP
Progress notes have a fixed structure. S and O are records. A, the Assessment, is the thinking. What we lacked was never what we did, but the judgment, in the moment, of why.
🔄 Five years later, the problem has flipped
2021: no one wrote the notes
2026: too many notes, no one reads them, no one thinks
AI generates transcripts, logs, and weekly reports in one click. Hospitals have seen this before: copy-and-paste in electronic medical records. Notes every day, and an Assessment copied forward from admission to discharge.
Completeness of the record no longer means completeness of the thinking.
✍️ Splitting SOAP
S and O: let AI compile them
A: must be a human, or at least confirmed by one
P: humans decide, AI executes
Each week, add just two or three lines: What do this week’s events mean? Which assumption got overturned? Should anything change?
If you can’t write them, that’s a signal in itself.
🔁 The discharge summary becomes the next project’s admission orders
A retrospective shouldn’t just be “for reference.” Write the lessons back into the system as templates, skills, and defaults, and the learning cycle finally closes on its own.
One condition: what goes back must be judgment, not a play-by-play.
AI can write your progress notes for you. It cannot write your Assessment.
The purpose of a record is not to prove the work was done, but to make the next round better.
#CQI365 #QualityImprovement #ProjectManagement #SOAP #AIAgents
Revisiting three kinds of inpatient records and project notes, five years later — when AI starts writing your progress notes.
Previously
In 2021 I wrote an essay that borrowed three types of inpatient records to manage projects:
| Inpatient record | Project note | Purpose |
|---|---|---|
| Admission note | Project plan note | Think through goals, resources, and risks before starting |
| Progress note | Work log | Record changes and adjustments during execution |
| Discharge summary | Project retrospective | Sum up after closing, as a reference for next time |
The most important line in that essay was:
We used to make grand plans, but without work logs, when we succeeded we didn’t know how, and when we failed we didn’t know why.
Five years later, that essay is still the most-read post on this site. Looking back, the analogy holds, but it was only half-written. One key structure went unnamed. And five years on, the problem the essay set out to solve has flipped completely.
1. What I Didn’t Spell Out: SOAP
In 2021 I wrote that the value of a progress note lies in “presenting the patient’s treatment status through logical reasoning.” That’s true, but too vague. Every clinician knows that reasoning has a fixed structure, called SOAP:
| Progress note | Project work log | |
|---|---|---|
| S Subjective | What the patient says, where it hurts | Team feedback, stakeholder input |
| O Objective | Vital signs, lab data, imaging | Progress, numbers, actual output |
| A Assessment | What do these data mean? Is the diagnosis right? Where is the illness heading? | What do these developments mean? Is the direction right? Where are the risks? |
| P Plan | Next tests and treatments | What to adjust next |
S and O are records. A is thinking.
Whether a progress note is good comes down almost entirely to A: whether the data were turned into a judgment, and whether it explains why treatment was kept or changed. No matter how complete S and O are, without A you just have a list of data.
Go back to the 2021 line, “when we succeeded we didn’t know how.” What was really missing wasn’t records but the Assessment: the judgment, made in the moment, of why we did it. All that remained was what we did.
2. Five Years Later, the Problem Has Flipped
The 2021 problem was: no one wrote work logs.
The 2026 problem is: too many records, no one reads them, and no one thinks.
Once AI entered our workflows, records became very cheap. Meetings turn into transcripts and summaries automatically, agents leave complete execution logs, and weekly reports can be generated in one click. Work logs are no longer scarce; they are starting to flood.
Hospitals have seen this before: copy-and-paste in electronic medical records.
As electronic records spread, progress notes appeared every day and grew longer every day. Look closely, though, and the Assessment barely changes from day to day, copied forward from admission all the way to discharge. The records are impeccably complete, but the thinking is absent. A common type of accreditation finding is exactly this: plenty of documentation, yet no visible clinical reasoning behind the decisions.
AI amplifies the problem. Copy-and-paste at least required an action; now even that is gone. A record that looks complete, fluent, and correctly formatted can be generated at any time.
Completeness of the record no longer means completeness of the thinking.
The same thing is happening on the wards: generative AI is starting to help draft admission notes, weekly summaries, and discharge summaries. The draft can be handed to AI, but whoever signs it is still responsible for the judgments inside.
3. Splitting SOAP: AI Writes S and O, Humans Own A and P
Since records have become cheap, rather than resist it, divide the labor:
| SOAP | Content | Who writes it in 2026 |
|---|---|---|
| S Subjective | What happened, who said what | AI can compile it: meeting transcripts, messages, feedback |
| O Objective | Progress, data, output | AI can pull it automatically: task status, numbers, files |
| A Assessment | What does this mean? Is the direction right? | Must be a human — or at least confirmed by one |
| P Plan | What to change next | Humans decide, AI executes |
With this split, people actually write less: no more typing up S and O by hand. But what remains carries more weight: every work log entry needs one human-written line of Assessment.
In practice it can be simple. Once AI has compiled the week’s S and O, you only add two or three lines:
- What do this week’s events say about where the project is heading?
- Which assumption in the original plan did this week’s data overturn or confirm?
- So, should anything change next week?
If you can’t write them, that is a signal in itself: either the project really isn’t moving, or you aren’t actually managing it — you’re just watching AI manage it.
AI can write your progress notes for you. It cannot write your Assessment.
4. The Discharge Summary Becomes the Next Project’s Admission Orders
In 2021 I wrote that the discharge summary “serves as a reference for the next visit,” and the project retrospective “serves as a reference for future work.”
The word “reference” now looks too passive. In reality, most retrospectives sink into a folder once written, and few people go back to them when the next project starts.
The AI era offers a new possibility: a retrospective doesn’t have to be a note for people to read; it can be written straight back into the system.
- Processes learned on this project become templates or skills the next one can use directly
- Pitfalls hit this time go into the standing instructions an agent reads before starting work
- Corrected time estimates and risks become the defaults for the next plan
For the first time, the three records truly form a loop: admission note → progress notes → discharge summary → written back as the next project’s admission orders → admission note…
The cover image of the 2021 essay was a diagram of a “learning cycle.” Five years later, that cycle can finally close on its own: lessons no longer depend on someone remembering to look them up. They appear by themselves when the next project begins.
There is one condition, though: what gets written back must be Assessment, not a play-by-play. Stuff a whole log back into the system, and the next project will suffer from “too many records, no one reads them” from day one.
5. Settling a Five-Year-Old Debt
In 2022 I left a comment under the original essay: I was bringing project-management ideas into Scrivener using these three types of notes, and would share how it went.
I never did.
The reason was simple: the workflow didn’t fit. The Scrivener approach didn’t last, and my three types of project notes moved back to Obsidian.
That episode is itself a discharge summary: the diagnosis was “tool incompatible with workflow,” and the treatment was returning to the original system. The only regret is that the summary came five years late.
It also echoes the point above: however good the plan, if no judgment is recorded during execution, all you have afterward is “and then it stopped.”
Closing
Taken together:
In 2021 I said: projects need admission notes, progress notes, and discharge summaries, because we are too good at planning and too bad at recording.
In 2026 I would add: records are no longer scarce; the Assessment is. AI can write S and O, but humans must own A and P. And the discharge summary shouldn’t be just a reference; it should be written back into the system to become the starting point of the next project.
Hospitals require a daily progress note not to prove that a doctor saw the patient, but to force the doctor to stop once a day and ask: what is actually going on with this patient right now?
Projects are the same. The purpose of a record is not to prove the work was done, but to make the next round better.