Teacher Removes Auto-Grading From His AI Tool After Student Incident

When efficiency breaks accountability

A mechatronics teacher returned from two consecutive field trips in 2024 to find roughly 450 ungraded assignments waiting. The volume pushed him to build his own AI grading assistant—one optimized for the messy reality of engineering coursework: design files, schematics, code samples, and photos of physical prototypes.

By April 2025, he had removed the tool's most automated feature: the ability to return AI-generated grades and comments to students before he reviewed them. The decision came after a student approached him to express gratitude for encouraging feedback on an assignment. The comment had motivated the student to revise and resubmit the work.

The problem: the teacher had never seen the comment. The AI had drafted, approved, and sent it without human review.

The friction point in automated grading

Steven Swanson, who teaches engineering and built the tool himself, said the feedback wasn't inaccurate. That made the situation harder to parse. After more than two decades in the classroom, he couldn't immediately articulate what felt wrong—only that something fundamental had shifted when a student received what appeared to be teacher judgment without the teacher actually making one.

The incident prompted Swanson to redesign his workflow. The current version presents a review dashboard where AI drafts every grade and comment against his rubric, but nothing reaches students until he edits, overrides, or approves each one. The system still saves time compared to manual grading, but now requires his eyes and judgment on every result.

Swanson teaches mechatronics, where efficiency and system optimization are core principles. He brought that mindset to his grading tool, tuning it to evaluate work against assignment instructions, handouts, and rubrics. The logical endpoint was auto-return—eliminating the last click between AI evaluation and student notification.

Removing that feature changed his understanding of what "human review" must mean. It can't be a glance at a score followed by an approve button. It requires checking student work against the rubric and owning the result that goes back to the student.

Policy and practice converging

New York City's public school guidance now states that AI must not replace educator decision-making. Other states are weighing requirements around human review and student data handling. Swanson's administrator raised two concerns when reviewing the tool: parents and students should know when AI assists with grading, and contested grades should be re-graded by hand.

Swanson's students know he built the tool. What they care about, he said, is whether feedback arrives quickly, rubrics are clear, and grades are fair. When the AI occasionally docks points for work it missed—usually due to cut-off screenshots or faint handwriting—students challenge the result. Swanson reviews the work and adjusts the grade.

He noted that students will challenge the AI more readily than they'll challenge him directly. A student will say "the AI got this wrong" but won't tell a teacher to their face that they made a mistake and owe points back. The AI draft creates space for students to speak up, while the final grade still passes through the teacher.

Why it matters

The distinction between AI that assists judgment and AI that replaces it will shape how schools adopt these tools at scale. Swanson's experience suggests the critical line isn't whether AI can generate useful feedback—it can—but whether students receive grades from a system or from an accountable human. As districts write policies on AI grading, the technical question of accuracy may matter less than the structural question of who owns the result.

Swanson advised schools to disclose specifically what AI assistance means: comments and grades drafted by AI and reviewed by the teacher. He also recommended addressing where student work goes, whether it's stored or used for training, and how secure the platform is.

When a student asks why they received a particular grade, Swanson wrote, the answer cannot be "because the system said so." It has to come from the teacher.

These details were first reported by EdSurge.