Teach the engineering,
not the syntax.
The process is the curriculum now.
Your students already use agents, and the assignment that asks them to write a function by hand is measuring something the industry stopped paying for. What remains teachable — and what employers are desperate for — is the discipline of supervising work you did not personally type.
ASE as lab equipment: every decision in a student’s run is visible, timestamped, and reviewable.
Assessment broke before the curriculum did.
Detection is a losing arms race. Instrumentation isn’t.
You can’t tell who wrote the submission.
Plagiarism detection was built for copied text, not for generated work. Every honest attempt to police it costs faculty time and produces false accusations.
Graduates who can prompt but can’t supervise.
Students learn to get output and not to verify it. Employers report the same gap: adoption is universal, and the judgment to review what comes back is missing.
Ban the tools or ignore them.
Both choices leave the student unprepared for the job they’re about to take, and neither survives contact with a laptop and a wifi connection.
Make the invisible part of the work visible — and gradable.
If the process is instrumented, you can teach it, assess it, and stop guessing about authorship.
Definitions, claims, verdicts, gates.
The student’s work is the definition they wrote, the scope they set, the claims they accepted, and the failures they caught — all captured as artifacts you can mark.
The run record shows the work.
Instead of asking whether AI was used, the record shows exactly how it was used, supervised, and verified. That’s the skill you meant to assess anyway.
Industry harnesses, governed.
Students run the same agent CLIs they’ll use professionally, inside the governance structure serious teams put around them.
Self-hosted, your models.
ASE runs on institutional infrastructure with your own model access — including fully local models when student data or licensing constrains what may leave campus.
What a governed assignment looks like.
One lab, four artifacts, all of them markable.
Students define the problem
Before any agent runs, the student produces a versioned problem and architecture definition — the part that used to be invisible.
They configure the constellation
Roles, scopes, models, and gates. Deciding what an agent may not do is itself the exercise.
The run happens under supervision
Claims are made and refuted, gates hold or open, and the student rules on what escalates.
You grade the record, not the guess
The evidence bundle shows the decisions, the failures, the remediation, and the judgment applied at each step.
Instrumentation that outlasts the assignment.
These artifacts serve grading, accreditation review, and the student’s own portfolio equally well.
- Per-student run records — chain-hashed evidence of how each submission was produced and supervised.
- Review artifacts — the claims a student accepted, the ones they rejected, and what that cost.
- Reusable course packs — definitions and constellation templates you author once and reissue each term.
- Portfolio evidence — graduates leave with demonstrable supervised-agentic work, not just repositories.
- Self-hosted deployment — institutional infrastructure, institutional model policy, no student data leaving campus.
This one rarely shows up alone.
Bring a course. We’ll bring the lab equipment.
Walk through a governed assignment end to end, from student definition to graded evidence bundle.
Elevate Your Vibe with ASE Precision.