Curriculum

AI UX and human oversight

AI UX and human oversight is 78 percent (71 companies), core, and central to the mid-course spine. The report wants oversight taught as a design decision, not a slogan: where to put human-in-the-loop, how to pick a level of autonomy, and ho

1. 07.01

   The four AI design patterns for a surface that

   You choose the AI interface pattern from your product's spread of user intents, instead of defaulting every AI feature to a chat box that most users cannot drive.
2. 07.02

   Levels of autonomy

   You set the level of autonomy for each AI action from its measured reliability and how reversible it is, and place the human in the loop where a mistake would be expensive, not where the demo felt shaky.
3. 07.03

   The four surfaces

   You design an AI feature as four separate surfaces, the input, the instruction, the output, and the feedback, each with its own craft, so a user can steer the system and catch its mistakes.
4. 07.04

   Stream the work

   You stream the work as typed events, commit to a run contract with a visible spend ceiling before the agent starts, and brief a returning user on what changed in thirty seconds.
5. 07.05

   The approval gate

   You place an approval gate where it actually catches the irreversible action, render the resolved payload the user is signing, and read the edit rate before confirm to prove the gate is real and not theater.
6. 07.06

   Claim-level citations

   You attach citations at the claim level, show confidence only where the user can act on it, and always leave a manual path, because the user's real job is to verify or route around the answer fast.
7. 07.07

   The empty box

   You design the empty box, the refusal, and the thumbs-down so each one tells the user what to do next and writes a row straight into your eval set.