Uploaded September 2026 | Updated November 2026, 2 hours ago
In the previous videos, I built a new QTI-based quiz authoring system with AI and got the Common Cartridge export working in both Canvas and Sakai.
Now comes the less glamorous part:
Code review.
I send the Quiz 1 branch through CodeRabbit and work through the findings one by one with Cursor and ChatGPT. Some are easy fixes. Some turn into architectural discussions. Some reveal bugs I simply had not tested yet.
We dig into things like POST/redirect form handling, session state across multiple tabs, missing quiz references, validation failures, UTF-8 case folding for human-entered answers, QTI CDATA handling, bad imported data, duplicate answers, and protecting critical legacy code paths.
One of the most interesting problems is what to do when a lesson references a quiz that has been deleted. Canvas silently removes the broken reference. Sakai keeps it visible to the instructor as a missing item while hiding it from students. After thinking it through, I decide the Sakai behavior is better — and we implement it.
There is also a good example of where AI shines: doing a broad, methodical sweep through a code path and asking, “Can any of these inputs blow us up?” Humans tend to think depth-first. AI can patiently do the British Museum version and inspect everything.
By the end, the branch is cleaner, safer, and more believable — but I’m still not pretending that one review means the code is perfect. The next step is to keep adding features and keep exercising the code until confidence grows naturally through use.
This episode is about the part of AI-assisted programming that matters most in production:
AI can help you write code incredibly fast. You still have to earn trust in it.
In the previous videos, I built a new QTI-based quiz authoring system with AI and got the Common Cartridge export working in both Canvas and Sakai.
Now comes the less glamorous part:
Code review.
I send the Quiz 1 branch through CodeRabbit and work through the findings one by one with Cursor and ChatGPT. Some are easy fixes. Some turn into architectural discussions. Some reveal bugs I simply had not tested yet.
We dig into things like POST/redirect form handling, session state across multiple tabs, missing quiz references, validation failures, UTF-8 case folding for human-entered answers, QTI CDATA handling, bad imported data, duplicate answers, and protecting critical legacy code paths.
One of the most interesting problems is what to do when a lesson references a quiz that has been deleted. Canvas silently removes the broken reference. Sakai keeps it visible to the instructor as a missing item while hiding it from students. After thinking it through, I decide the Sakai behavior is better — and we implement it.
There is also a good example of where AI shines: doing a broad, methodical sweep through a code path and asking, “Can any of these inputs blow us up?” Humans tend to think depth-first. AI can patiently do the British Museum version and inspect everything.
By the end, the branch is cleaner, safer, and more believable — but I’m still not pretending that one review means the code is perfect. The next step is to keep adding features and keep exercising the code until confidence grows naturally through use.
This episode is about the part of AI-assisted programming that matters most in production:
AI can help you write code incredibly fast. You still have to earn trust in it.










