The thinking stays open while the product takes shape.
Working positions on assignments, assessment, agents, and the learning record.

01Stop detecting AI. Start assigning it.
A working product thesis for turning ungoverned outside use into course-shaped work.
Detection starts after the work has already escaped the course. An AI Assignment begins earlier. The teacher defines the purpose, sources, help, challenge, and limits of the agent. The student works inside those terms. The result includes both the artifact and the learning process that produced it.
This is not a final slogan. It is a design test: does the assignment make better work possible, make delegation less useful, and give the teacher evidence worth reviewing?
02A grade should point back to the work.
Assessment becomes more trustworthy when every suggestion retains its exact source.
A rubric level without evidence is only an opinion with a number beside it. Atlas keeps the transcript passage, artifact revision, course source, rubric version, draft judgment, and teacher decision connected.
The system may help find and organize evidence. The teacher still decides whether that evidence supports the judgment and what becomes final.
03The agent should know the course it is in.
Useful AI behavior comes from the learning record, role, and teacher-authored terms.
A generic assistant can answer a question. A course-shaped agent can ask the right next question because it knows the approved material, the assignment boundary, the student's work, and the learning goal.
That context also creates responsibility. The role determines what the agent may see and do. The record shows what it used, what it changed, and what still needs a person to decide.
Bring the assignment that tests the thesis.
The most useful feedback comes from a real course where the current AI workflow is already failing teachers or students.
Start a product conversation