---
title: Turn learning notes into a reviewable article with a content agent
description: >-
  Follow an actual first draft, six specific review comments and a revision.
  Includes source notes, a fact table and a reusable bilingual writing exercise.
canonical_url: 'https://onevium.com/blog/personal-agent-notes-to-reviewable-article'
language: en
published_at: '2026-11-08'
publisher: Onevium
---
# Turn learning notes into a reviewable article with a content agent

Follow an actual first draft, six specific review comments and a revision. Includes source notes, a fact table and a reusable bilingual writing exercise.

## The material was real, but the opening was still vague

Our first opening reported six sources, an inventory of sixty-two articles, and two selected directions. It concluded that a good agent workflow needs clear goals, sufficient context, and verifiable results.

The numbers were supported, yet the reader still could not see the problem this assistant would solve. Later instructions to prepare sources, propose topics, and check duplication were similarly abstract.

We revised that draft into the [personal research assistant article](https://onevium.com/blog/personal-agent-research-topic-cards). This companion explains the actual editing work: supply the evidence, produce a reviewable draft, give specific feedback, and check the revision. **Decisions that already happened and artifacts the reader can open are especially useful writing inputs.**

## Hand over a working packet

We used the previous task’s six source records, the 62-entry article inventory, and four topic cards. We did not introduce a fictional customer to stand in for the work.

source-notes.md records how far each source was read. topic-cards.md preserves the decisions to proceed or defer. editor-notes.md explains what actually completed and what did not.

For example, the editor read a complete automatic transcript but could not verify the video footage. Claude Code attempts produced no usable artifact; the material organization completed in Codex. That distinction determines whether a sentence can claim a successful product walkthrough.

[Download the working pack](https://onevium.com/downloads/workplace-editorial-lab.zip) and follow its writing-exercise instructions. It contains the actual first draft, review, and revision. English translations of Chinese artifacts are identified as editorial translations, not separate model runs.

## Make a short fact table before drafting

We organized the evidence into this table before writing the full article:

| Available evidence                   | Supported wording                       | Unsupported extension                                         |
| ------------------------------------ | --------------------------------------- | ------------------------------------------------------------- |
| 62 entries in the exported inventory | A snapshot of 62 registered articles    | Full-text deduplication completed; Google indexed 62 articles |
| Six source records                   | Six sources at different reading depths | Six videos watched in full                                    |
| Two proceed and two defer decisions  | Four editorial decisions                | Two articles completed; two topics have no value              |
| Saved draft, review and revision     | This editing process and its outputs    | Proven long-term productivity or success rate                 |

The table makes it easier to spot a summary that changes the underlying fact. It does not need to become a lengthy report.

Fact checking and revision took place within the same task and included editorial judgment. This was not a blinded comparison with independent reviewers.

## Ask for a small, reviewable first draft

This request packages the editing steps for another assignment. Replace the inputs and review the new output:

```text
Read the source notes, topic cards and actual work record.
Write for a workplace reader who wants to delegate one recurring task.
First list proposed factual claims and their input evidence.
Keep unverified features, results and numbers separate.
Then create an outline, an opening and one complete practical section.
That section must contain:
- the task and input;
- the decision or action taken;
- a substantive artifact excerpt;
- a check and a next step.
Distinguish the source's demonstration from our own work.
Do not call an unsuccessful product attempt a successful test.
Save first-draft.md and preserve the inputs.
```

An opening and one practical section are enough to expose a misunderstanding early. If the draft never shows what a topic card contains, adding more sections will not fix that omission.

## The six changes we requested

The full review is in outputs/editor-review\.md. Its six comments address specific gaps:

- Lead with the reader’s difficulty turning saved tutorials into a next action, rather than reporting input counts first.
- Replace abstract references to goals and context with the actual research topic selected and scheduled-workflow topic deferred.
- Add a copyable request with file inputs and a clear handoff format.
- Preserve the inventory’s scope: titles and descriptions, not full-text review.
- State the actual execution environment and provide the artifacts without adding unverified cloud or scheduling claims.
- Use the file list to locate materials while keeping the work itself as the main narrative.

These comments identify changes to evidence, examples, and information order. They give an editor more direction than asking for a more professional or deeper article.

## Compare the change at the same point in the draft

The revised opening starts with a recognizable problem: saved tutorials do not tell you which one to try, what it will produce, or whether you already have similar material. It then names the actual output: two selected directions and two deferred directions with reasons.

The middle of the article adds a real decision. “Check duplication” becomes a concrete case: a related scheduled-workflow article already exists, while the new source is only a description. Obtain a new mechanism and an actual task before deciding to update or create an article.

| First draft                               | What the revision adds                                                |
| ----------------------------------------- | --------------------------------------------------------------------- |
| Clear goals and sufficient context        | A reader brief, six evidence statuses and the actual inventory fields |
| Check whether it duplicates existing work | A specific related article, defer reason and missing material         |
| Make the next step clearer                | Write an opening from the notes and retain review and revision        |

The figure presents excerpts from the actual artifacts. The source text remains available in the working pack.

![Actual draft and revision compared: abstract goals and duplication checks become a reader problem, a concrete editorial decision and missing evidence to obtain.](https://onevium.com/assets/blog/workplace-editorial/writing-en.svg)

Typeset excerpts of actual editorial artifacts, not a chat screenshot or a model score.

## Check that a smoother revision still says the same true things

We reread the final text against the inputs, checking the inventory count, source IDs, decision states, and links to the referenced sources and related articles.

One automatic transcript read did not become all videos watched. Possible overlap identified from a directory did not become full-text deduplication. Two selected topics did not become two published articles. Each distinction has a corresponding file in the working pack.

Visual review asks similar questions: are the changes readable on a phone, does the caption identify an artifact excerpt, and is the key explanation also available as text? Attractive formatting cannot establish a factual claim.

The two articles share one working pack. Give its first draft and review to your assistant, then compare the revision with ours. The exercise has no single correct wording. Look for the reader’s problem, the actual decision, and the evidence boundary.

## Keep the parts worth reusing

For the next assignment, replace the sources, topic cards, and work record. Keep the sequence of fact table, small draft, specific feedback, revision, and rereading in a workflow file the assistant is explicitly asked to read.

After using it across different tasks, consider packaging the stable method as a Skill. The [Claude Code Skills documentation](https://code.claude.com/docs/en/skills) explains reusable instructions; our [Skill maintenance guide](https://onevium.com/blog/maintain-team-skills) addresses keeping those instructions useful. Saving the file does not establish future reliability.

This task left a first draft, concrete review comments, bilingual revised text, and a fact-check table. A later editor can see why the article changed and continue from the remaining work.

Source: [Teacher’s Tech — Build Your First AI Agent](https://www.youtube.com/watch?v=gHB4JFG9i3k)
