How to Build an AI Sales Agent Without Code (That Actually Closes Leads)

Your inbox has 47 unread messages. Twelve of them are leads. Three arrived yesterday.

By the time you reach them, two have already booked a call with a competitor.

A Harvard Business Review study found that companies contacting a lead within the first hour are seven times more likely to qualify it. Most teams take 24 hours or more.

The Short Life of Online Sales LeadsAre you confident that your company is effectively handling potential customers’ online queries? Think hard. Our research shows that most companies are not responding nearly fast enough. Companies in financial services, automobiles, education, software, health care, professional services, and many other industries have increasingly turned to the internet to generate sales leads. Indeed, corporate spending […]Harvard Business Review

This is the real sales problem for small teams: it's not about finding leads. It's about being fast enough to actually catch them when they show up.

An AI sales agent solves this at the root. Not by automating a flowchart. By doing the work itself, the way a human would, just without sleeping.

In this guide, you'll build a complete inbound sales agent on Rerun from scratch. No code. No API keys to manage. You'll see exactly what it does when a lead arrives, how to watch it work live, and what a qualified reply looks like on the other side.

In a hurry? Spin up your agent now and come back to follow along.

What an AI Sales Agent Actually Does

A sales agent is not a chatbot that answers FAQs. And it's not a Zapier workflow that sends a templated email when a form is submitted.

A real sales agent reads the lead, thinks about it, and decides what to do next.

Here's what that looks like in practice with Jules, Rerun's inbound sales agent:

1. A new lead hits your inbox (or form, or CRM)
2. Jules reads the message and scores the fit against your criteria
3. If the fit is strong, Jules sends a personalized reply and books a discovery call
4. If it's weak, Jules files it with a brief note and moves on
5. You see every step live on your dashboard, in plain language

No templates. No if/then branches to maintain. Jules decides.

Automation vs. Agent: Why the Distinction Matters

Before building, it's worth understanding what you're actually creating. "AI sales automation" gets used for very different things.

Automation reacts to triggers. An agent handles situations.

The difference matters when your leads don't fill out the form the way you expected, or when a reply comes in at 2am that needs a thoughtful response, not a canned acknowledgement. Speed and quality, at the same time, is exactly what an agent delivers.

What You'll Have at the End

By the end of this guide, Jules will:

• Monitor your inbox for new inbound leads
• Score each one against your ideal customer profile
• Send a warm, specific reply within minutes of arrival
• Book a discovery call for strong fits
• Log every action to your dashboard so you can review it

You'll set all of this up in Rerun. No code, no separate services to wire, no maintenance burden.

Step 1: Create Your Sales Agent

Open Rerun and create a new agent. Name it Jules. When prompted to pick a template, choose Sales.

The template comes pre-loaded with the skills Jules needs: reading and writing email, scoring inbound leads, and booking calls. You're not starting from zero, you're starting from a working baseline.

Templates are starting points. Jules's skills and brief are fully editable. The next two steps are where you make it yours.

Step 2: Connect Your Stack

Jules needs access to your tools. Head to the Connectors tab and connect:

• Gmail (or your email provider): so Jules can read inbound leads and send replies
• Calendly or Cal.com: so Jules can book discovery calls with a real link
• HubSpot or Notion (optional): to log qualified leads directly in your CRM or a database

Each connector is one click. Rerun handles the OAuth. No API keys to paste anywhere.

Step 3: Write Jules's Brief

This is the most important step. Jules's brief is the instruction set that tells it what a good lead looks like, how to respond, and when to book.

Go to the Soul tab and set Jules's brief. Here's a starting point built around a typical B2B SaaS ICP:

{
  "role": "You are Jules, the inbound sales agent for [Your Company]. Your job is to qualify new leads, write personalized replies, and book discovery calls for strong fits. You represent the company professionally and warmly.",
  "ideal_customer": "B2B founders or operators at companies with 2 to 50 people, dealing with repetitive internal work or customer communication. They are already paying for at least one SaaS tool and are a decision maker or direct influence on buying decisions.",
  "qualification_criteria": {
    "strong_fit_signals": ["mentions a specific pain point", "clear use case described", "is a decision maker", "urgency expressed"],
    "weak_fit_signals": ["student or academic", "no budget signal", "agency doing spec work", "vague or no context given"],
    "score_threshold": 7,
    "scoring_note": "Score 1 to 10 based on fit signals. Write one sentence explaining your score."
  },
  "on_strong_fit": "Send a warm reply that references their specific pain point. Offer a 20-minute discovery call. Include the Calendly booking link.",
  "on_weak_fit": "Send a short, polite reply. Point them to the free trial or docs. Do not book a call.",
  "on_borderline": "Flag for human review in the To Handle tab. Include your score and reasoning.",
  "tone": "Warm, direct, no corporate fluff. Sound like a thoughtful human, not a CRM template."
}

Adapt ideal_customer and qualification_criteria to your actual ICP. The more specific you are here, the better Jules performs from day one.

Step 4: Set the Trigger

Jules needs to know when to run. Go to Scheduled Tasks and set up a real-time trigger on new emails received.

A few settings worth configuring:

• Filter: Only emails not from existing contacts (Jules handles new inbound, not ongoing threads)
• Frequency: Real-time gives the fastest response. Every 15 minutes works if you prefer batched runs.

Step 5: Watch Jules Work

This is the part that changes how sales feels.

When a new lead arrives, open your Rerun dashboard. You'll see Jules's status switch from idle to working. Click on Jules and open the Logs tab.

You'll see every action in real time, with timestamps and token usage:

• Reading the email
• Scoring the lead, with the score and one-sentence reasoning
• Drafting the reply
• Sending it
• Booking the call, if applicable

Nothing happens in a black box. If Jules makes a decision you'd have made differently, you can see exactly why it made the one it did. Then update the brief to correct it.

This is the core difference. Other tools automate. Rerun lets you watch the work actually happen. That visibility is what makes the agent improvable.

What a Qualified Reply Looks Like

Here's an example of what Jules sends to a strong-fit lead who wrote in about their manual follow-up problem:

"Hi Sarah, thanks for reaching out. Sounds like you're spending a lot of time manually following up with trial users who go quiet, which is exactly the kind of work Jules was built for. I'd love to show you how it handles that with your stack. Here's a 20-minute slot that should work: [booking link]. Looking forward to it."

Jules pulled "manually following up with trial users who go quiet" directly from her email. That's not a template. That's the agent reading the situation and responding to what was actually said.

Human-in-the-Loop: When Jules Asks for Help

Jules doesn't just run and hope. When it hits an edge case it's not confident about, it pauses and surfaces it for review.

If a lead scores a 6 out of 10 (borderline in the brief above), Jules flags it in the To Handle tab instead of making the call itself. You see Jules's score and its one-line reasoning. You decide: approve the reply, redirect it, or dismiss it.

You can also get notified on Slack. Approve from there, and Jules picks up exactly where it paused.

This is the human-in-the-loop pattern. Jules handles the 90% it's confident about. You handle the 10% that genuinely needs judgment. And each time you correct an edge case, Jules learns from it for next time.

How to Tune Jules Over Time

The first version of Jules won't be perfect. That's expected and fixable.

After the first week, review the Logs for cases where Jules sent a reply you wouldn't have sent, or missed a lead you'd have caught. Update the brief with what you learned. Two or three edits usually make a significant difference.

After the first month, look at your booking rate versus your response rate. If Jules is booking too many weak fits, tighten the qualification_criteria. If it's missing strong fits, lower the score threshold or add fit signals.

The brief is your lever. Every update makes Jules more accurate for your specific ICP, not just the generic template it started from.

Pricing

Jules runs inside a Rerun Space, your own private cloud machine. Agents run 24/7 with no run quotas and no per-execution fees.

Every plan starts with a free 3-hour trial. No credit card required.