✨ For Visitors ✨

Overview

This page is for those who said "tell me more" after hearing about our Agentic Studio course in Spring 2026. The rest of the site is what we provided to students during the semester (and we didn't always do a good job of keeping it up to date).

In a nutshell, we:

• have a hypothesis that teaching early stages of agentic coding amounts to teaching core principles of software engineering and design
• wondered where our expert blind-spots were around students' abilities and interest in using these principles when developing applications with agents
• wanted to explore these ideas with students who had had some, but not too much, programming experience

So we ran a research project under the structure of a group independent-study course.

You can hear the hour-long summary of the course and its design in the Oxide and Friends podcast episode that Shriram and Kathi did in May 2026.

Further down, we provide info about our research design, a summary of our observations and findings, as well as a more detailed schedule of the topics and activities (along with relevant handouts), and where we'll be going from here.

Research Design

As part of the coding projects in the course, students were required to keep reflection journals about their experiences: what feelings were invoked at different points, what technical content they drew on, what they felt they needed to know, and how they split up the work between themselves and Claude. They also submitted their chat transcripts alongside their assignments (which came in via GitHub).

Students also wrote various documents to separately capture what they were learning or what advice they would give to other students about working with agents. These were given as assignments.

Our research questions include:

• How do Claude’s capabilities and behaviors interact with students’ experience (technical and affective) of the program-development process?
• In what ways do students utilize software engineering practices, and when do they do so relative to what Claude was doing?
• How do student reactions to working with Claude and software engineering practices change across the semester?

Findings

We are actively analyzing the data in Summer 2026. So far, we have found:

• Most students developed their own personal boundaries for when to use agents and for how much
• A majority of students were concerned about their lack of understanding of their code bases
• Most students were concerned about the loss of learning if we put this in intro courses
• Students found significant value in the peer-review and crit activities
• Several students wished the course had spent more time on advanced agentic skills (like orchestration and developing Claude skills), though we had warned upfront that we wouldn't be covering those. Others felt the course hadn't taught as much as it could have (though they noted the likely differences between 2nd and 4th semester students in this regard)
• Students by-and-large adopted testing and design practices, including the use of model-view-controller, but they did not pick up on externalization as much as we might have liked
• Students appreciated how the variety in the projects let us expose them to many areas of CS that are covered in more advanced courses

Did we uncover blind spots? Yes, though they were different spots for different faculty (we won't name names!)

• Students didn't necessarily understand how websites worked
• Even in a small class centered around group feedback, some students were still reluctant to talk
• The prior experience differences were more noticeable than we expected in discussions and submitted work
• The later assignments took students longer than we anticipated, particularly as they ran into systems issues and token/plan limits (they had $20/month Claude plans by design); this in turn curtailed some of what we asked them to do.

We are still extracting blind spots from the student journals (as part of data analysis).

Looking Ahead

In Fall 2026, we will be expanding the experiment downward by weaving agents into an experimental section of an intro CS course with programming novices. We still intend to teach students how to program manually, but we will include some agentic projects so that we can also motivate design and testing with more nuance that our intro courses usually allow. Part of the challenge here will be training teaching assistants in the code and design review skills that such a course will demand.

Shriram will also expand the experiment upward by re-designing his programming languages course, bringing in many of the themes (types, constraints, testing) from this one.

Our project designs focused on implementation and testing, not front-end UI design. We realized how much our students would benefit from some attention to this area.

Detailed Class Schedule