Chapter 21 - The Multi-Agent Essay Feedback System
Submit, wait, revise: a system that scores, explains, and emails with zero ongoing effort
The Friday afternoon problem, and a different architecture
This chapter does not propose a faster version of the same broken feedback cycle - submit, wait days, receive comments. It proposes a different pipeline entirely: submit, wait minutes, revise.
| Stage | What happens |
| 1. Submission | Student submits an essay through a simple form |
| 2–4. Specialist agents (parallel) | Separate AI agents evaluate structure, language accuracy, and content/argument quality |
| 5. Delivery | Results are scored, explained in plain language, and emailed automatically - zero ongoing teacher effort after setup |
Full automation is a pedagogical choice, not a concession to convenience. Because the system runs without teacher effort, students can write more often without increasing teacher workload - and frequency of writing, not feedback quality alone, drives improvement.
Before you build: a setup checklist
- Confirm access to an AI platform that supports multi-step agent pipelines (see Appendix N ↗ for current setup options)
- Draft the grading rubric in full before building any automation - the system is only as good as the rubric behind it
- Pilot with a single class and a single essay type before rolling out across a full course
- Plan the revision cycle: how many automated passes before a human teacher reviews the result
Additional activities for Chapter 21
IndividualSetup project
Activity 21.A - Rubric-First Build. Before any technical setup, write the full grading rubric you want the system to apply, then check it against the five-agent structure to confirm each criterion has a clear agent responsible for it.
GroupPilotProject
Activity 21.B - Single-Class Pilot. Run the system with one class on one essay type for two weeks, comparing student revision rates and teacher time spent against the previous manual feedback cycle.
Further resources