Assessment After the Final Product

Topic

Assessment After the Final Product: Making student thinking visible in an AI-rich classroom

This professional development proposal helps teachers redesign assessment around visible thinking rather than final-product policing. In an AI-rich environment, a polished essay, slide deck, solution, or summary is weaker evidence of learning than it used to be. The answer is not endless surveillance or AI-detection theater. The answer is to gather better evidence: drafts, transcripts, annotations, live questioning, student explanations, process artifacts, and moments where students must show judgment.

Audience / Use Case

Primary audience:

• Teachers worried that AI has made take-home work unreliable.
• School leaders designing assessment guidelines.
• Department teams revising common assignments.
• Instructional coaches looking for concrete, non-punitive alternatives to AI detection.

Best use cases:

• A 90-minute PD on assessment redesign.
• A department workshop where teachers convert one vulnerable final-product assignment into a visible-thinking assessment.
• A faculty conversation on academic integrity that separates “teaching AI literacy” from “gathering valid evidence of learning.”

The session should reassure beginners while still challenging enthusiasts: the goal is not to grade everything students do. The goal is to choose better evidence at key moments.

Framework / Core Claims

1. Final products are weaker evidence in an AI-rich classroom

Take-home essays and polished products are not dead, but they are no longer sufficient as the default proof of learning. Steven Mintz argues that work that matters will increasingly need to be “work witnessed”: handwritten prompts, attendance-linked tasks, student-led sessions, discussion, presentations, and live questioning (content/articles/stevenmintz-substack-com-ai-killed-the-take-home-essay-covid.md).

This does not mean abandoning writing or projects. It means changing what counts as evidence. The final product becomes one artifact among several, not the whole case.

2. Visible thinking beats policing final products

AI detection tools and punitive audits can create a climate of suspicion while still failing to answer the instructional question: what did the student understand, decide, revise, and transfer? Visible thinking asks students to make their process legible enough for feedback and evaluation.

Visible thinking may include:

• Pre-AI notes, predictions, outlines, or attempts.
• Annotated AI transcripts.
• Draft comparisons.
• Claim-evidence maps.
• Live oral defense or conference questions.
• Reflection on what AI contributed, what the student rejected, and why.
• In-class checkpoints where students apply the same concept without AI.

3. Chat transcripts can be annotated evidence, not surveillance logs

Mike Kentz’s “grading the chats” frame argues that student AI transcripts can show prompt writing, decomposition, response analysis, iteration, and refinement (content/articles/mikekentz-substack-com-how-grading-the-chats-makes-learning.md). The best version of this practice is not “turn over everything you typed so I can catch you.” It is “select and annotate the parts of your process that show your thinking.”

The transcript becomes a text students read critically. Students can highlight where they:

• Framed the problem.
• Made an initial attempt.
• Asked AI for targeted help.
• Challenged or corrected AI.
• Checked a source or claim.
• Revised the product based on judgment.
• Rejected an AI suggestion.

A key implementation recommendation from the source material: teachers should first grade or annotate their own AI interactions. This helps teachers understand what mature evidence looks like and prevents unrealistic expectations.

4. Productive friction is an assessment design requirement

One of the strongest lines from the Kentz source material is: “AI wasn’t replacing student work. It was replacing student struggle.” That sentence should anchor the session. The issue is not whether students used a tool; the issue is whether the tool removed the productive friction needed for learning.

Productive friction can be designed through:

• First attempts before AI.
• Required because/but/so reasoning constraints from The Writing Revolution.
• Long-division-style foundations before calculator-style leverage.
• “Vibe coding” principles: foundation before leverage, director not passenger, attempt before augmenting, catch the machine.
• Reflection prompts that ask students to identify what was hard, what AI made easier, and what they still had to decide.

5. AI literacy and assessment integrity are separate problems

This is a high-leverage distinction. Teaching students to use AI well is an instructional goal. Determining whether an assessment produced valid evidence is an assessment-design goal. They overlap, but conflating them creates confusion.

For example:

• A lesson may allow AI because the objective is to critique AI output.
• Another assessment may restrict AI because the objective is unaided fluency.
• A third may require AI use and assess the student’s judgment in managing it.

The PD should help teachers write assignment-specific AI use statements rather than universal slogans.

Proposed Activities

Activity 1: Vulnerability audit of a familiar assignment

Teachers bring one current assignment and identify where final-product evidence is weak.

Prompts:

• What does the final product currently prove?
• What could AI produce or heavily assist?
• Which learning objective must be witnessed more directly?
• Where would a short checkpoint reveal student understanding?
• What should students stop doing so teachers are not grading more total work?

Deliverable: a revised evidence map showing which artifacts will count and why.

Activity 2: Transcript studio

Teachers examine a short student-AI transcript, ideally fictional or anonymized. They annotate it as if it were a reading passage.

Look for evidence of:

• Task framing.
• Decomposition.
• Prompt revision.
• AI response analysis.
• Pushback.
• Verification.
• Student decision-making.

Then teachers discuss what is grade-worthy and what should remain formative. The point is not to grade longer; it is to grade smarter.

Activity 3: Model your own AI use

Before asking students to submit annotated chats, teachers complete a miniature AI task themselves and annotate their own transcript.

Possible task:

• Ask AI to help revise an unclear assignment prompt.
• Ask it to generate misconceptions for an upcoming unit.
• Ask it to critique a rubric.

Teachers then mark where they accepted, rejected, revised, or verified suggestions. This makes the hidden judgment visible and helps teachers avoid treating any transcript length as automatically good.

Activity 4: Visible-thinking assessment clinic

Teachers redesign one assignment using a simple structure:

1. Before: What must students do before AI can enter?
2. During: What process artifact will show thinking without becoming surveillance?
3. After: How will students explain what changed and why?
4. Witnessed: What short live or in-class check will verify transfer?
5. Stop doing: What old grading burden will be reduced or removed?

The “stop doing” step is essential for skeptic buy-in. If visible-thinking assessment only adds paperwork, it will fail.

Activity 5: Productive friction rewrite

Teachers revise one prompt to include a productive constraint. Examples:

• Before AI, write three possible claims and one piece of evidence for each.
• Use because/but/so to explain a relationship before asking AI for feedback.
• Ask AI for counterarguments only after drafting your own position.
• Require students to identify one AI suggestion they intentionally rejected.
• Add a post-session reflection: “Where did AI make this easier, and where did I still have to think?”

Specific Evidence Supporting the Framework

• Grading chats as show-your-work: Kentz argues that transcripts reveal prompt writing, decomposition, response analysis, iteration, and refinement. Teachers should first grade or annotate their own interactions before requiring this of students (content/articles/mikekentz-substack-com-how-grading-the-chats-makes-learning.md).
• Modeling effective AI use: Teachers can compare exemplar and non-exemplar transcripts, annotate them, co-create criteria, and design fictional transcripts around two or three key moves (content/articles/mikekentz-substack-com-how-to-model-effective-ai-use-in.md).
• Grade 12 comparative transcript pilot: A four-week pilot with 21 Grade 12 students from 14 countries found that 85.7% changed their AI approach, 47.6% became significantly more strategic, and 81% endorsed continuing the method. The limitations are important: small sample, self-report, short duration, task initially too simple, and rubric too fixed (content/articles/mikekentz-substack-com-what-happened-when-we-taught-ai-literacy.md).
• Take-home essay reliability problem: Mintz argues that take-home work has become less reliable as evidence and that important work will need to be witnessed through in-class writing, discussion, presentations, student-led sessions, and live questioning (content/articles/stevenmintz-substack-com-ai-killed-the-take-home-essay-covid.md).
• Student use is already widespread: Pew reports 64% of teens use chatbots, 54% use them for schoolwork help, and 59% think cheating with chatbots is regular (content/articles/pewresearch-org-how-teens-use-and-view-ai.md).
• Student voice warns against underground use: Students ask teachers to help them understand where AI fits because they are already using it constantly; criminalizing AI can push use underground (content/articles/fitzyhistory-substack-com-what-students-want-teachers-to-know.md).

Overlooked / High-Leverage Details

• Visible thinking beats policing final products. This should be the headline.
• Transcripts are texts to annotate, not surveillance logs. Have students select and explain meaningful excerpts.
• Start small. Transcript-based assessment can impose high cognitive burden on students and teachers.
• Use model transcripts and flexible rubrics before requiring real student transcripts. Students need examples of what mature AI use looks like.
• Productive friction is not anti-AI. It is the learning design that prevents AI from replacing the hard part.
• AI misuse can reveal bad incentives. If students outsource an assignment, the assignment may be rewarding performance over learning, speed over reflection, or compliance over meaning.
• Teach “AI can wait.” The pause before AI is an assessment design move.
• Build speed bumps. Pre-session purpose/boundaries and post-session reflection make use more intentional.
• Separate policy categories. “No AI,” “AI permitted with documentation,” “AI required and assessed,” and “AI used as object of critique” are different assignment types.

Skeptic / Implementation Cautions

• Workload is the vulnerability. Teachers need to know what they can stop grading. A transcript requirement added to an unchanged essay may be too much.
• Do not oversell transcripts. A transcript can still be performative, incomplete, or generated after the fact. It is evidence, not proof.
• Rules still matter. Visible-thinking practices should coexist with clear assignment boundaries.
• Avoid surveillance framing. If students feel every keystroke is being monitored, they may hide use rather than reflect on it.
• Equity must be concrete. If some students have better tools, paid access, quiet home environments, or stronger prior knowledge, visible-thinking requirements may amplify differences unless access and scaffolding are addressed.
• Small pilots are not settled science. The Grade 12 pilot is promising but limited. Frame transcript work as a practice to test and refine locally.
• Not every task needs AI documentation. Use process evidence where the learning stakes justify it.

Suggested Source / Wiki Anchors

• content/articles/mikekentz-substack-com-how-grading-the-chats-makes-learning.md
• content/articles/mikekentz-substack-com-how-to-model-effective-ai-use-in.md
• content/articles/mikekentz-substack-com-what-happened-when-we-taught-ai-literacy.md
• content/articles/stevenmintz-substack-com-ai-killed-the-take-home-essay-covid.md
• content/articles/pewresearch-org-how-teens-use-and-view-ai.md
• content/articles/fitzyhistory-substack-com-what-students-want-teachers-to-know.md
• content/articles/nickpotkalitsky-substack-com-thinking-with-ai-the-teacher-workshop.md

Possible Session Title Options

• “Show Your Thinking: Assessment Design After AI”
• “Beyond AI Detection: Visible Thinking in an AI-Rich Classroom”
• “Productive Friction: Keeping the Learning in AI-Assisted Work”