My AI Agent Isn't Allowed to Decide Anything

Описание и применение

Система Shot‑Delivery Guardian использует ИИ для анализа статуса производства фильмов и рекомендует, какой элемент можно отложить, чтобы избежать просрочки, соблюдая ограничения по утверждению клиента, приоритету режиссера и зависимостям.

Задача

When a film or TV show pipeline stalls, a human must quickly decide which work to delay to avoid missing deadlines, but this manual process is slow and error‑prone.

Результат

In live tests, the system matched the fixed rule’s recommendation 60% of the time, identified backlog causes in 44 seconds on average, and never violated any of the safety constraints (client approvals, director flags, or downstream dependencies).

Рабочий процесс

• 1. Monitor pipeline stages for backlog and deadline status.
• 2. Query three monitoring tools: overall backlog, work item metadata, and detailed stage timing.
• 3. Apply the fixed rule to filter out client‑approved, director‑flagged, or dependent items.
• 4. Rank remaining items by slack hours and select the one with the most spare time.
• 5. Output the recommendation as a note on the operator’s dashboard.
• 6. Operator reviews and confirms the delay decision.

Доказательства

I built a system that watches the work as it moves through each step, notices when a deadline is about to be missed, figures out why, and says which piece of work is safe to push back.

Источник

The AI never got an important decision wrong, because I never let it make one.

Источник

def score_bump_candidates(shots):
eligible = [
s for s in shots
if not s["client_approved"]
and not s["director_flagged"]
and not has_downstream_dependents(s, shots)
]
return sorted(eligible, key=lambda s: slack_hours(s), reverse=True)

Источник

During one real test, the system hit a real problem reaching one of its three tools halfway through. It said so, plainly, in its own words.

Источник

3 of 5 real tests matched what the fixed rule alone would've picked. About 44 seconds on average to look into a problem and respond, timed against the real live system. Zero broken rules across every single test.

Источник