Framing the questions, the theory of change, the logframe, the indicators, and the proposal that wins the work. AI is strong at structure and weak at promises, so you supply the logic and the commitments.
AI drafts a theory of change from your programme description, builds a first logframe, proposes SMART indicators, and pressure-tests the narrative. You check every baseline, target and assumption, and mark anything drafted as needing validation.
Agree what the donor is really asking, and what a good answer would look like.
Describe how and why the programme should work, level by level.
Activities, outputs, outcomes, impact, with an indicator at each level.
Test each one for the SMART qualities; never invent a baseline.
Ask the AI to review it as a sceptical donor would, then fix the gaps.