A search that returns nobody is information about the role
Sourcing starts with a search string that uses the role exactly as written. In most cases people run more searches rather than questioning the brief.
An experienced recruiter treats that empty result as part of the work: the job title is wrong in this market, the skill goes by a different name in this industry, and the seniority you want is not found in the region you are targeting.
The candidate list is not what makes the search worth doing. The discoveries are. If the role cannot be filled as written, you have to talk to the hiring manager. The point is that this conversation happens in week one instead of week eight.
The boundary is public sources and consent. If a person appears in a search they did not request, creating a candidate record from that result without their knowledge is a legal problem in many jurisdictions and a reputational issue everywhere.
How the agent is built in Agent Studio
The search strategy may adapt. The system prompt requires the agent to report every change and the reason for it. That explanation is the deliverable, not a side note.
The system prompt also sets the limits. The agent must use only public professional sources. It must not assess any individual. It must not keep personal data beyond the limits in your own policy. Retention is a policy decision, so the prompt refers to your policy rather than inventing one.
Prompt skills hold role vocabulary across sectors and your known talent markets, because one job has four different names.
The agent runs the search from an Agent Chat node, which uses an HTTP Request node and a Scrappey node.
The results go to Notion in two sections: a list of candidates, and what the search showed about the brief. Anyone who has done a difficult search reads that second section first.
The agent does not contact anyone. Whether to send a message is a person's decision. Treat the sourcing output as a list to consider, not as something to email.
Google Sheets records the search terms that led to certain results, so similar jobs are easier to find next time.
What this agent is built from
- System prompt: search strategy may adapt, and every change must be reported with its reason.
- System prompt: public professional sources only, no individual assessment, and retention per your policy.
- Prompt skills: role vocabulary across sectors and your known talent markets.
- Scrappey (Util): performs this search under the agent's direction, together with HTTP Request (Util).
- Agent Chat (Util): lets the agent adapt the search between rounds.
- Notion (Integration): receives candidates and, separately, what the search showed about the brief.
- When an ActionFlow is enough: If you fill the same role repeatedly, a saved search on a schedule is a workflow.
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.