FROM THE AI FRONTIER
Seven questions to ask before launching an AI coworker
Choosing a model is one item on the list. Inspired by Philipp Schmid’s talk, we turn seven components into a practical brief for a scheduled assistant.
Before launching a work assistant, describe seven things: its model, environment, instructions, tools, access, action rules and trigger. This helps turn “have it make a report” into a specific assignment.
What the talk covers
At 1:00–1:45, Schmid compares preparing an AI coworker with onboarding a person. He later demonstrates a technical implementation using Gemini. For this review, we focus on the seven-part organisational framework and turn it into questions for a task brief.
Example: a morning management report
Imagine an assistant preparing a summary from a spreadsheet. Before connecting services, complete this brief.
- Model. What counts as an acceptable result on sample data: correct totals, readable text and the necessary detail?
- Environment. Where will the source spreadsheet and reports live? How will versions be identified?
- Instruction. Which measures belong in the report? Which period is the comparison against?
- Tools. Is reading a sheet and creating a document enough?
- Access. Which files does this task require? Who grants and revokes access?
- Rules. What happens when rows are missing or dates are wrong? Who reviews the draft?
- Trigger. When should the report run, in which time zone? What happens if there is no new data?
This is our illustrative task brief, not a ready-made technical configuration.
Before automation
Walk through the task once with the assistant. Record the details you had to explain: a column name, a date boundary or the output format. Add those clarifications to the brief.
Try completing the seven items for one report today. The blanks will show which decisions still need to be made before launch.
Sources
The starting point is a talk by Philipp Schmid of Google DeepMind. AI Engineer uploaded the video on November 6, 2026. This review was prepared on November 10 from its transcript.
- Philipp Schmid, Google DeepMind — оригинальное видео
- AI Engineer — страница выступления и расшифровка