AICODEAgentic Programming · 67723 · 2026-7

Week 3 ·lab

How busy were your agents?

In Lab 3 you ran several coding agents at once on Coin Dash. Every pair’s report recorded when each agent was working. Here is the whole class, put together. All findings are anonymous.

Most of the time, no agent was working

The lab asked each of you to keep at least two agents busy. We took every student's lab window and measured, second by second, how many of their agents were working at that moment.

How many agents were working at once

Share of all students' lab time, by the number of agents working on that student's laptop at that moment.

Show as a table

Agents rarely waited on you to answer a question. Most of the gap came between tasks: an agent finished a feature in a minute or two, and then sat idle while you reviewed the diff, merged, played the game and decided what to ask for next. Many of you said exactly this in your reports: the bottleneck was the person, not the agents.

Agents per student

The same measurement, one number per student: on average, how many agents were working during your lab window? And what was the most you ever had working at the same moment?

Average agents working

Students, by agent working time divided by the length of their lab window.

Show as a table

Most agents working at once

Students, by the largest number of agents working at the same moment.

Show as a table

The whole class at once

Labs ran on different days and started at different times. Here we line every student's lab window up as if everyone had started at the same moment, and add up the agents working across the whole class.

Agents working across the whole class

Minutes from the start of each student's lab window. Hover for the numbers.

Agents working, summed over all studentsStudents whose lab window is still open

What you built

Features per team

Teams, by the number of different features they built.

Show as a table

The most popular features

Number of teams that built each one.

Branches and commits

Branches per team

Teams, by the number of feature branches in their history.

Show as a table

Commits per team

Teams, by the number of commits in their history.

Show as a table

Record holders

Here we list the team that outperformed everybody else, and its numbers next to the class median and mean, to give you a sense of what is possible. Well done on going to the extreme here!

“Working” means an agent was busy with your request. An agent that had finished and was waiting for your next message counts as not working.