Set a take-home lab assignment today, for example "write a program to check whether a number is a palindrome", and a good share of the submissions will have been written by ChatGPT. They will be correct, neatly indented and politely commented. And they will tell you almost nothing about who in the class can actually code.
Banning ChatGPT does not change that, because a take-home program cannot be watched. Trying to spot it from the code alone does not work either: a correct short program looks the same whoever wrote it.
What does work has two halves. Move the marks to where the student has to write the code themselves. Then use AI on your side of the desk, so that doing so does not cost you every evening of the semester.
Which lab work ChatGPT can do for a student
| Lab work | Can ChatGPT do it for the student? | What the mark still tells you |
|---|---|---|
| Take-home textbook program (palindrome, factorial, bubble sort) | Yes, in seconds | Very little |
| Practical record file | Yes. The programs and outputs can be generated and copied out by hand. | Neatness and effort, not ability |
| Take-home mini project | Mostly, with some editing | Some, if the student is asked about it afterwards |
| Timed in-lab test in full screen, copy-paste blocked | Much harder. Copy-paste is blocked and tab switches are recorded. A phone on the desk is still the invigilator's job. | Whether the student can write working code alone |
| Short viva on the student's own code | No | Whether they understand what they wrote |
The pattern is simple. The further the work is from the lab and the teacher, the less its mark means. What a browser can and cannot enforce during a test is covered in detail in stopping copy-paste and tab-switching without a proctoring tool.
Move the marks into the lab
Take-home work still has a place. It is practice, and practice with ChatGPT open beside you is not useless. It just should not carry many marks. One way to split a semester's internal lab marks:
| Part | Share of lab marks | Why |
|---|---|---|
| Short in-lab tests, every two or three weeks | 60% | Written by the student, under the same conditions for everyone |
| Viva on two of the student's own programs | 20% | Checks understanding, and costs a few minutes per student |
| Take-home assignments and record | 20% | Rewards practice without letting it decide the grade |
The objection comes straight away, and it is fair. Six in-lab tests a semester means setting six papers, and each one means finding questions, checking them and marking 60 programs per section. That is exactly the work AI should be doing.
See this in your own lab. Teemly sets coding tests from your question bank, marks every program against test cases, and explains the results in plain English. We set it up free.
Book a free demoUse AI to set the in-lab test
On our demo college, we opened the question library, pressed Build with AI and typed:
Teemly AI built a draft called "Python Loops and Functions Test", 30 minutes long, from the department's own question bank:
| # | Type | Question |
|---|---|---|
| 1 | Coding, easy | Write a function digit_sum(n) that returns the sum of the digits of a non-negative integer n. |
| 2 | MCQ | How many times does for i in range(5) run? |
| 3 | MCQ | Which keyword defines a function? |
| 4 | MCQ | What does a function return if it has no return statement? |
The draft exactly as Teemly AI built it.
We asked for two coding questions and got one. The note it left explains why:
This is the behaviour to look for in any AI tool for teachers. The bank was short, so it said so. It did not write a second question to fill the gap, because a question nobody has checked has no place in a test that counts. The fix takes a few minutes: add one more easy coding question to the bank yourself, then build again.
Once the paper is sent, the marking takes no time at all. Each program runs against hidden test cases, and the marks are the share of cases passed. Building a mock placement coding round with AI walks through that marking step with four real student answers.
Use AI to read the marks
Here is a question that matters more now than it did two years ago: which students do much better on work done outside the lab than inside it? If you enter take-home marks in Teemly as a paper test, you can ask exactly that.
Our demo class has no take-home marks, so we asked the nearest question we could: a written mid-term against the online lab tests, for a class of 45.
The answer began:
Teemly AI's answer on our demo college. The names are demo students.
It then gave both students' marks paper by paper, and said what the data could not show: the mid-term is one paper, and some students missed it.
The useful part is what it did not do. Asked to find students with a suspicious gap, it found none and said so. It did not invent a suspect to give a more interesting answer. When a real gap does show up, it is a reason to call the student in for a short viva, never a verdict. Teemly AI makes no integrity decisions. Faculty make every one.
Five questions to ask any AI tool in your department
AI tools for teachers are arriving faster than anyone can evaluate them. These five questions sort them quickly, and the answers are short.
| Ask | Why it matters | Teemly AI's answer |
|---|---|---|
| Are student names sent to the AI? | Student data should not leave for a model that does not need it | No. Students are replaced by labels like "Student 7" before anything is sent, and names are put back on your screen afterwards. |
| Can it state a number that is not in the marks? | An invented average in a report to the HOD is worse than no report | No. Every number in an answer is checked against the marks before you see it. |
| Does it mark students or decide cheating cases? | Those decisions belong to someone who can be asked to explain them | No. Programs are marked by test cases, and faculty make every integrity decision. |
| Can it send anything to students on its own? | A draft should stay a draft until a teacher has read it | No. Nothing reaches students until you review it and send it. |
| Does it write questions? | An unchecked question in a test that counts is a risk | No. It picks only from your own bank, and tells you where the bank is short. |
The short version
Students now have AI on their side of the desk. Faculty should have it on theirs. Let take-home work be practice, put the marks into short in-lab tests, and let AI make those tests cheap: one sentence to set, automatic marking, and plain-English answers about what the marks show.
To see an in-lab coding test from the student's side, the two-question sample test at teemly.in/try needs no sign-up. And to set one for your own department with Teemly AI, see Teemly's coding assessments.