> ## Documentation Index
> Fetch the complete documentation index at: https://docs.juicebox.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Welcome Kit: Automate your sourcing with Agents
This guide is part of a welcome kit series for Juicebox users. Check out the [previous guides on running a search](https://docs.juicebox.ai/welcome-kit-run-your-first-search) and [reaching out to candidates](https://docs.juicebox.ai/welcome-kit-reach-out-to-candidates) if you haven't used natural-language search, filters, criteria, shortlists, or sequences yet.
In the previous welcome guides, we showed you how to create a project, search for candidates using filters and criteria, shortlist top candidates, and create outreach sequences. If you need to review those topics, check out those guides first.
In this guide, we’ll set up AI Agents to source autonomously for us. You’ll learn how to:
* Set up an Agent for always-on, automated sourcing
* Calibrate the Agent to deliver high-quality leads
* Decide how you want it to deliver, approve, and reach out to leads
## What are agents and how can they help?
There’s a lot of talk in the market about “agents,” so let’s demystify what that means. AI agents are software capable of gaining context, making decisions, and taking multi-step actions to achieve a goal, with limited human oversight.
Juicebox Agents are essentially an extension of your recruiting team. Think of them like a world-class sourcing partner who is always working, proactively building your pipeline. Like a new member of your team, they need training and feedback to do their jobs well and understand the way you think about roles.
Agents don’t replace your human thought and judgment, so it’s important to know what to delegate to Agents vs. where it’s best for you to focus. Our most successful customers have a division of labor that looks like this:
**Agents:**
* Manage **top-of-funnel sourcing**
* Provide **proactively, deeply researched sourcing support**
* Work on **evergreen, always-open positions** where sourcing is continuous
* Work on **high-volume roles** where there’s a large search pool and consistent job titles and backgrounds across companies
* Use the ATS integration to rediscover **past ATS talent**, such as runner-up candidates from previous searches (available on Business plan only)
**Recruiters:**
* Focus on **candidate relationships and advising the business**
* Own the **strategy and decision making** for each role
* Work on **more strategic or specialized roles** with narrower search pools or more specialized candidates
Like a junior sourcing partner, an Agent will learn from your guidance and continuously search for relevant candidates to surface for you. But also like a junior sourcing partner, it needs your feedback to get better over time. We’ll show you how that works.
**Want more help?** Learn more about [Juicebox Agents](https://docs.juicebox.ai/juicebox-agents).
## Set up an agent
Running a search with a Juicebox Agent starts almost exactly like any other search. You still select **Create new project** for a new role and give it a title, like `Backend Engineer`.
*Just like other projects, use one project per role, so each Agent works on one role at a time.*
Up until now, you’ve probably been using the **Search** interface to describe the candidate. But now let’s switch the toggle over to **Agent**.
Similar to any search, you’ll describe the ideal candidate in natural language. Help the agent focus on the essentials by telling it the key skills or requirements for someone to excel in the role.
The Agent will work with any information you give it, even a quick natural-language blurb, but the best practice for faster, more accurate results is to give some context upfront to help the Agent learn. We recommend providing:
* The **job description,** pasted in or linked from your ATS
* Any intake notes you have, or have Juicebox record and store the **intake meeting** to train the agent (see below)
*Agents can also search the web, so you can tell them something like* `Find people currently at a Series B company that raised in the last 12 months` *or* `Find SWEs at competitor companies who recently had layoffs`
## Use intake notes to start an Agent (optional)
Juicebox can join your hiring manager intake meetings to record and transcribe, and provide live insights for you and your hiring manager to use as you collaborate. These notes can also give Agents context, so they start with the full picture of the role.
To run an intake through Juicebox, look for **Intake** under your project in the left-nav. You can write your own notes or use a template. Then have Juicebox start recording the meeting, in-person or over Zoom.
After the meeting, Juicebox stores the notes in a transcript along with an AI summary of the meeting. From here, just select **Start agent** and you can kick off an Agent directly from these meeting notes, with full context.
After this, the process follows the same steps as if you started it from a natural language search or a job description.
## Initial Agent setup
Once you enter your natural-language description, things will start to look different from a regular search.
Before finding candidates, the Agent will use any tools you've connected, like your ATS, notetaker apps, and drive/docs, to research the role. We call this Contextual Intelligence.
Then it will start to map the talent market for your role. As it goes, it analyzes the full candidate pool and tests different variations of the search criteria. This makes sure you're getting a strong pipeline, not just a few good profiles.
It will adjust filters and criteria as it goes, or even ask you for clarifications. You'll see the Agent's reasoning as it goes, so you can always give it feedback or additional input by chatting with it. You can also see and change the filters or criteria manually from the right-hand nav.
For example, it might see that the titles are too specific and broaden the search, and propose a new approach:
*If the number of potential leads is high during this initial search, don’t worry. That’s a good thing! This is not the number of candidates it will send you. It’s the size of the pool that it’s going to evaluate against your criteria.*
**Want more help?** Learn more about [setting up Agents](https://docs.juicebox.ai/juicebox-agents#setting-up-an-agent).
## Calibrate the Agent (one-time setup)
The first time you set up an Agent, you’ll need to calibrate it so that it learns your patterns of judgment and consistently sends you good candidates. Think of this like training a new teammate.
The Agent will give you three profiles to evaluate. Select **Review leads** to review each of them like you would any candidate, not looking for perfection but for match potential.
* If you would take a call with this candidate, select **Looks Good** to approve them.
* If you wouldn’t take a call with this candidate, select **Not a Fit** to reject them.
Always provide feedback when Juicebox asks **What’s wrong with this profile?** Be specific with your feedback, similar to how you’d train a new sourcer on your team to learn what you’re looking for. For example:
* `Right skills but wrong company stage. We're looking for hypergrowth experience, not enterprise.`
* `Job hopping pattern, three roles under 2 years, doesn't match our retention bar.`
* `Avoid candidates who worked at Company X.`
You can also provide “yes, but…” feedback when you approve candidates, such as:
* `Strong technical background but I'd prefer customer-facing experience`
* `Background is in internal L&D but I'd prefer someone who worked in Customer Success`
* `5 YoE is a maybe, I'd look for candidates with more experience`
The Agent is learning from all this feedback and getting better at pattern-matching, so *more feedback is always better.*
Keep approving or rejecting candidates that the Agent provides until it's done calibrating. The Agent will ask you to keep calibrating until you're consistently approving candidates (roughly two out of every three in a batch).
You'll know it's done when it moves on to outreach setup. **If you haven't gotten to the settings questions yet, the agent isn't done calibrating and will not run.**
The Agent keeps learning as you accept or reject more candidates over time. You won't need to do a full calibration again unless you significantly change the parameters of the search.
**Want more help?** Learn more about [setting up and calibrating Agents](https://docs.juicebox.ai/juicebox-agents#setting-up-an-agent).
## Decide on settings
After calibration, choose your outreach approach first. Then set your daily lead target and approval preference.
**Outreach:** Choose one of these options and select **Confirm**:
* **Build me a sequence** asks your Agent to draft an AI-generated sequence using the context from your conversation.
* **Use an existing sequence** lets you select outreach you have already prepared for the role.
* **I'll do this later, shortlist for now** collects candidates in your shortlist without sending outreach. You can add them to a sequence later.
**Review your sequence:** If you choose **Build me a sequence**, the Agent opens a **Sequence** pane beside your conversation. Share real outreach examples and selling points, then ask the Agent to refine the draft. You can also edit the title, sender, message, and timing directly in the canvas. Select **Preview** on each step to check its personalized copy.
You continue using the sourcing Agent chat here. [Composer chat](/sequences#generate-with-ai) is the separate side panel you use when creating a sequence from **Sequences**.
Changes save automatically. Check the save status, use **Close side pane** to return to the full conversation, and select **Open sequence** on the sequence's chat card whenever you want to reopen it. If prompted, choose **Connect Gmail** or **Connect Outlook** in chat to connect the mailbox that will send your outreach.
**Sourcing strategy:** When you are happy with a sequence the Agent built, confirm its request to use it for daily sourcing. The Agent then returns to **Last step, set your sourcing strategy**. If you picked an existing sequence or shortlist-only outreach, **Confirm** takes you straight to this step. Choose your daily volume and approval type together.
* **Volume:** Pick the number of net-new leads you have time to review each day. If in doubt, start with 25. Higher volumes may require additional Agent seats; the selector shows what is available.
* **Approval:** Choose **Manual** to review and approve each lead before outreach or shortlisting. Choose **Automatic** to let the Agent approve matching leads and carry out your selected outreach or shortlist action.
Start with **Manual** for the first few days, until you are approving about 75% of candidates. Review leads at least every three days: if they sit unreviewed for four days, the Agent auto-pauses.
Review the sequence and strategy settings, then select **Start sourcing**. Creating or editing the sequence alone does not start the Agent.
To change these choices later, open the Agent's **…** menu and select **Sourcing settings**. Changes apply to the running Agent; select **Done** to close the settings dialog.
## Refine over time
Once the agent is running, it sources on a schedule: around 9:00am, your local time, every day.
* If you set outreach and approval to automatic, it’s now running automatically. Sit back and relax.
* If you skipped the outreach step, you’ll need to go to your shortlist to manually send sequences.
* If you set approval type to **Manual**, you’ll need to review and approve candidates on a regular basis. Do this from within the Agent by selecting **Review Leads**.
*Depending on the search, the Agent may eventually run **out of leads**. That’s because there’s a finite number of relevant candidates for your search, and an Agent won’t compromise the quality bar after that.*
Every time you return to the Agent, you'll see what it learned and how it's adapting its search.
You can continue to chat with your Agent. This is especially useful if you have other apps connected to Juicebox. For example, you can ask it to:
* Deliver insights and recommendations on your pipeline
* Determine which criteria are limiting your qualified candidate pool
* Pull candidate lists from connected apps, like Google Sheets or your ATS
* Generate a talent market map into Notion
You may even want to create Agents for existing projects that started with manual searches. That’s as easy as creating a new search from the left sidebar. Just select **Create Agent** to add one to an existing project.
**Want more help?** Learn more about [reviewing leads](https://docs.juicebox.ai/juicebox-agents#reviewing-leads), [recalibrating Agents](https://docs.juicebox.ai/juicebox-agents#recalibrating-an-agent), or the [Knowledge Vault](https://docs.juicebox.ai/knowledge-vault).
## What's next?
Congratulations! By following this guide, you:
* Set up an Agent for always-on, automated sourcing
* Calibrated the Agent to deliver high-quality leads
* Decided how you want it to deliver, approve, and reach out to leads
* Resolved issues when the Agent pauses or runs out of leads
While this is the end of the Welcome Kit, there's more to explore in Juicebox. Check out these features next:
* Use [Juicebox for Slack](https://docs.juicebox.ai/juicebox-for-slack) to take action on candidates while you're on the go. It also lets you collaborate with hiring managers directly in Slack and give them visibility into the sourcing process.
* Use the [Juicebox MCP](https://docs.juicebox.ai/juicebox-mcp) to combine Juicebox data with other systems of record and get custom insights, directly in your AI tool of choice
* Use [Network Sourcing](https://docs.juicebox.ai/network-sourcing) to get warm leads directly from your teammates' networks
* Use [Talent Rediscovery](https://docs.juicebox.ai/talent-rediscovery) to source candidates you've already interviewed and re-engage them
* Use [Smart Reports](https://docs.juicebox.ai/smart-report) to generate shareable reports from your shortlists
Similar to any search, you’ll describe the ideal candidate in natural language. Help the agent focus on the essentials by telling it the key skills or requirements for someone to excel in the role.
The Agent will work with any information you give it, even a quick natural-language blurb, but the best practice for faster, more accurate results is to give some context upfront to help the Agent learn. We recommend providing:
* The **job description,** pasted in or linked from your ATS
* Any intake notes you have, or have Juicebox record and store the **intake meeting** to train the agent (see below)
After the meeting, Juicebox stores the notes in a transcript along with an AI summary of the meeting. From here, just select **Start agent** and you can kick off an Agent directly from these meeting notes, with full context.
After this, the process follows the same steps as if you started it from a natural language search or a job description.
## Initial Agent setup
Once you enter your natural-language description, things will start to look different from a regular search.
Before finding candidates, the Agent will use any tools you've connected, like your ATS, notetaker apps, and drive/docs, to research the role. We call this Contextual Intelligence.
Then it will start to map the talent market for your role. As it goes, it analyzes the full candidate pool and tests different variations of the search criteria. This makes sure you're getting a strong pipeline, not just a few good profiles.
It will adjust filters and criteria as it goes, or even ask you for clarifications. You'll see the Agent's reasoning as it goes, so you can always give it feedback or additional input by chatting with it. You can also see and change the filters or criteria manually from the right-hand nav.
For example, it might see that the titles are too specific and broaden the search, and propose a new approach:
* If you would take a call with this candidate, select **Looks Good** to approve them.
* If you wouldn’t take a call with this candidate, select **Not a Fit** to reject them.
Always provide feedback when Juicebox asks **What’s wrong with this profile?** Be specific with your feedback, similar to how you’d train a new sourcer on your team to learn what you’re looking for. For example:
* `Right skills but wrong company stage. We're looking for hypergrowth experience, not enterprise.`
* `Job hopping pattern, three roles under 2 years, doesn't match our retention bar.`
* `Avoid candidates who worked at Company X.`
You can also provide “yes, but…” feedback when you approve candidates, such as:
* `Strong technical background but I'd prefer customer-facing experience`
* `Background is in internal L&D but I'd prefer someone who worked in Customer Success`
* `5 YoE is a maybe, I'd look for candidates with more experience`
The Agent is learning from all this feedback and getting better at pattern-matching, so *more feedback is always better.*
**Review your sequence:** If you choose **Build me a sequence**, the Agent opens a **Sequence** pane beside your conversation. Share real outreach examples and selling points, then ask the Agent to refine the draft. You can also edit the title, sender, message, and timing directly in the canvas. Select **Preview** on each step to check its personalized copy.
You continue using the sourcing Agent chat here. [Composer chat](/sequences#generate-with-ai) is the separate side panel you use when creating a sequence from **Sequences**.
Changes save automatically. Check the save status, use **Close side pane** to return to the full conversation, and select **Open sequence** on the sequence's chat card whenever you want to reopen it. If prompted, choose **Connect Gmail** or **Connect Outlook** in chat to connect the mailbox that will send your outreach.
**Sourcing strategy:** When you are happy with a sequence the Agent built, confirm its request to use it for daily sourcing. The Agent then returns to **Last step, set your sourcing strategy**. If you picked an existing sequence or shortlist-only outreach, **Confirm** takes you straight to this step. Choose your daily volume and approval type together.
* **Volume:** Pick the number of net-new leads you have time to review each day. If in doubt, start with 25. Higher volumes may require additional Agent seats; the selector shows what is available.
* **Approval:** Choose **Manual** to review and approve each lead before outreach or shortlisting. Choose **Automatic** to let the Agent approve matching leads and carry out your selected outreach or shortlist action.
Review the sequence and strategy settings, then select **Start sourcing**. Creating or editing the sequence alone does not start the Agent.
To change these choices later, open the Agent's **…** menu and select **Sourcing settings**. Changes apply to the running Agent; select **Done** to close the settings dialog.
## Refine over time
Once the agent is running, it sources on a schedule: around 9:00am, your local time, every day.
* If you set outreach and approval to automatic, it’s now running automatically. Sit back and relax.
* If you skipped the outreach step, you’ll need to go to your shortlist to manually send sequences.
* If you set approval type to **Manual**, you’ll need to review and approve candidates on a regular basis. Do this from within the Agent by selecting **Review Leads**.
