Consistency is the argument, and only if the candidate knows what is happening
First-stage screening interviews become inconsistent because different people run them and candidates are asked different questions. Interviews happen on different days, and the hiring team then scores conversations that cannot be compared.
A structured first stage does make the conversations comparable. A fixed script creates its own problem: when the candidate gives only a partial answer, the script continues, and the most informative thirty seconds of the conversation never happen.
When you notice a vague answer, you ask for a more specific one. You recognize when the answer is complete, then you move on.
None of this can stay secret. The candidate needs to know they are speaking to an automated interviewer. If you hide that it is automated, you are judging them on how well they talk to software.
How the agent is built in Agent Studio
The system prompt starts with a disclosure that the interviewer is automated and notes that a human option is available. If the candidate asks for a person, the agent hands over immediately.
Prompt skills hold the question set. For each question they state what a complete answer contains, so the agent can tell a complete answer from an incomplete one by that definition, not by counting words.
The system prompt names the topics it can cover and those it must avoid. Anything involving personal circumstances, health, family, or protected characteristics is entirely out of scope. That limit is a rule, not a judgment left to the model.
The agent does not assess anyone. It organizes the transcript by question, with follow-up questions included, and the hiring team scores later.
Durable sessions keep the conversation coherent for 20 minutes, rather than breaking it into a series of separate prompts.
Notion receives the organized transcript. Slack then notifies the hiring manager that it is ready.
If a candidate asks for a person, they get a person. Routing happens immediately, not after a retry.
What this agent is built from
- System prompt: disclose as automated at the start, offer a human alternative, and hand over on request.
- System prompt: topics that are out of scope entirely, stated as rules rather than judgment.
- Prompt skills: the question set and what a complete answer contains for each.
- Durable sessions: what keeps a twenty-minute conversation coherent.
- Notion (Integration): receives the transcript organized by question, with follow-ups visible.
- Boundary: no assessment, score, or recommendation is produced. The hiring team scores.
- When an ActionFlow is enough: If you do not need follow-up, a written questionnaire is a workflow that treats every candidate the same, and it is the better choice.
Frequently asked questions
Start building AI workflows
Create a free account, open a template or a blank canvas, and run your first ActionFlow.
Newsletter
Get product updates
New nodes, agents, and product notes. We send mail only when we have something worth opening.