- Before
- Teaches five sections, 152 students, and needs about 25 hours to give real feedback on one essay assignment, so Monday's essay comes back a week or two later.
- Now
- Grades six samples, sees where the agents lean, reviews the essays that need attention and releases the class the next school day.
A ninth-grade class set held at 80% agreement until one anchor was fixed
A district high school, the afternoon after an argument essay was due. An English 9 teacher's Period 3 class of 34 students wrote "Should the district move to a four-day school week?" from three sources. Here is how AI essay feedback reaches them, screen by screen, in the working solution.
A worked example on sample data. Names, IDs and figures on the screens and in this section come from the demo.
- 01The afternoon after the due date
What the agents did overnight, and one next step
The agents' briefingBy 5:46 p.m. the evening before, the Feedback agent had drafted 155 comments on 31 of Period 3's 34 essays. Three wait: one arrived late, one student had extended time, and page 2 of one handwritten essay was too faint to read. The Calibration agent agrees with the teacher on only 24 of 30 sample scores, so the briefing's next step is Calibrate first.
Fairness agent: changes on 3 essays. Writing process agent: one question on one student's essay.
- 02One click
Rubric on the left, the essay in the middle, comments on the right
English teacherOne student's essay, "Four Days Is Not Enough", sits in the center. The rubric rail shows the district's Grade 9–10 argument writing rubric with a proposed level for each of five criteria, 12 of 20 in all, which reaches the gradebook only when the teacher confirms it at release. Selecting Evidence lights up the three sentences its comment rests on; each comment quotes its sentence, names its rubric line and offers Keep, Edit, Drop and Ask.
Period 3 argument essay: 31 of 34 drafted, 0 released
- 03The student's own notes
A newcomer's Vietnamese notes beside the English
Fairness agentThe student arrived in February, is at English language development level 2 and outlined the essay in Vietnamese. The Fairness agent shows each of the five lines beside an English translation checked by the English learner specialist. Line 4 says parents work on Friday, so younger brothers and sisters stay home alone, and the Reasoning comment credits that thinking and asks the student to bring it into the essay.
- 04Three essays changed
Comments on the argument, not on dialect or grammar
Fairness agentDrafting never sees student names: 31 of 31 drafts were written name-blind. On the newcomer's essay four grammar comments became one pattern; on another, two comments that treated dialect features as reasoning problems were rewritten; a third student's Haitian Creole margin notes were translated. With names swapped, no score changed, and each change shows before and after with Accept and Restore.
- 05Calibrate first
Where the agent leans, before the teacher trusts it
Calibration agentThe teacher had graded six sample essays the day before. The Calibration agent scored them without seeing those scores and agreed exactly on 24 of 30 criterion scores, 80%, under the 85% gate that locks release. It rates Evidence one level higher than the teacher on 4 of 6, crediting cited sources even when nothing explains them, and suggests anchor wording drawn from the teacher's own scores.
Suggested anchor, level 2: "Evidence is listed or quoted without explaining how it supports the claim, however many sources."
- 06A few minutes later
One anchor change, and two sections re-scored
English teacherThe teacher applies the tightened anchor to their own sections. The agents re-score Periods 3 and 5: 9 Evidence scores in Period 3 and 7 in Period 5 go down one level, and 13 Period 3 Evidence comments are rewritten to ask for explanation. Agreement rises to 28 of 30 and the gate is met; the two remaining differences, on Organization and Conventions, are the teacher's scores and they stand.
Agreement rose from 80% to 93%.
- 07One question
Questions for the student, marked Not a finding
Writing process agent138 of one student's 184 words arrived in one paste at 11:41 p.m. on Sunday, and the essay cites a 31 percent figure that is not in Sources A–C. The agent marks it "Not a finding" and offers three questions to ask; nothing here reaches the student or the family. The teacher chooses to release the feedback and hold the score until they talk on Wednesday at 7:45 a.m.
- 08Before release
Six checks, and the three waiting essays drafted on the teacher's go-ahead
English teacherThe release check lists six gates. The late essay is accepted, the extended time is not counted as late, and the faint page 2 has been rescanned; the agents draft all three and the teacher keeps the five comments on each. Essays with open items are opened one by one, and a random spot check of five essays covers the rest of the batch.
- 09The release
34 essays released in one batch
English teacherThe teacher releases the class in one step and confirms the scores as the checker. The Release agent posts the comments to the learning management system, writes 33 scores to the gradebook with one held for the talk, and sends 4 family summaries. Of 170 drafted comments, 121 were kept as drafted, 38 edited and 11 dropped with a reason, and Recall stays open for 10 minutes.
- 10What the student sees
The teacher's comments, in English with Vietnamese beside them
Grade 9 studentIn the learning management system the student sees the teacher's feedback on "Four Days Is Not Enough", 12 of 20, each comment under the sentence it is about, with English and Vietnamese side by side. The family receives a short summary in Vietnamese, checked by the English learner specialist at 3:31 p.m. and sent with the release.