Showing Your Work

AI-Assisted Programming for PhD Researchers

Day 3 afternoon is not a normal session. Everyone takes a turn at the front, shows what they built, and says what gave them trouble. No slide deck, no rehearsed talk.

How It Runs

Show the thing running, talk us through it. Plan for something like ten minutes each, but this is a guideline and not a timer: if a project sparks a discussion, we have the discussion. We take a break somewhere in the middle of the block.

Order is whatever works on the day. If you want to go first or last, say so.

What to Show

The thing itself. Run it. Show the output, and show the command or the steps that produce it. If your demo breaks live, that is fine, and it is often more interesting than a demo that works. Talk us through what it was doing before it broke.

How you worked with the agent. Where you wrote the spec, where the agent planned, where it implemented, where you checked it. Your spec and your commit history do most of this for you, so open them rather than describing them. Include the plain version of your AI-use disclosure: which tool, roughly how much of the code came from the agent, and where you went against what it suggested.

What went wrong. Where the agent got something wrong, where you had to step in, where you spent an hour on something you expected to take five minutes. This is the part everyone else learns from, so do not smooth it over.

Anything you would tell the others. If you found something that worked, or a trap worth avoiding, say it. One sentence is enough.

Nothing Here Is Graded

The course is pass/fail on attendance. Nothing you show this afternoon changes that, so there is no reason to polish anything or hide the parts that did not work. The half-finished project with an interesting failure in it is the more useful thing to bring.