For the past few months, I've been building a sales BDR (Business Development Representative) agent. In larger software companies, the sales function is typically divided into multiple roles, with the BDR responsible for outreach, follow-up, and eventually booking meetings. These leads are then handed off to the account executives, who handle the actual sales process.

In the company I work for, we only have one sales person who handles both the BDR and account executive roles. This is why we decided to build a sales BDR agent to handle the entire process, from receiving leads from Hubspot or Salesforce to booking meetings.

We broke down the BDR agent's responsibilities into different tasks:

1. Lead Research: When the agent receives a new lead, it should research the prospect to understand their business and identify a potential value proposition.
2. Personalized Outreach: The agent should draft a personalized message to the prospect based on the research.
3. Follow-up: The agent should follow up with the prospect, as they often don't respond immediately.
4. Prospect Engagement: If the prospect responds, the agent should be able to answer their questions and log the information in the CRM.
5. Meeting Booking: Finally, the agent should be able to book a meeting with the prospect by checking the sales team's calendar availability.

However, we quickly realized that this process is quite complex, and even building an agent to autonomously manage the inbox and respond to emails is a challenging task. This led to a paradigm shift in my approach to building AI agents.

Initially, I had envisioned a "super agent" that could handle multiple tasks on its own. But as I built more agents, I realized that a more effective approach might be to have a team of specialized agents working together to deliver the desired results.

This approach has several benefits:

1. Easier to Update: Instead of changing the entire system, you can update a specific agent.
2. Compound Benefits: The fundamental capabilities, such as answering questions or scheduling meetings, can be reused across different job functions like sales, marketing, and support.
3. Reduced Building Costs: By reusing common capabilities, the cost of building new agents is significantly reduced.

In the case of our sales BDR agent, we implemented this team-based approach. The "manager agent" is responsible for categorizing and delegating tasks to different sub-agents, as well as completing key actions like sending and receiving emails, creating calendar events, and logging information in the CRM.

The sub-agents include:

1. Prospect Researcher: This agent has access to various data sources (e.g., Google, Apollo, LinkedIn) to understand the prospect and identify a value proposition.
2. Inbox Manager: This agent is trained on the company's data to respond to any questions the prospect might have.
3. Follow-up Agent: This agent has been trained on proven follow-up cadences.

When the manager agent receives a new lead from the CRM, it delegates tasks to the sub-agents, and they work together to complete the entire sales BDR process.

This approach has already delivered around 20% of the meetings to our sales team, and there are a few interesting observations I've made about adopting AI employees versus human employees:

1. Transparency: With AI employees, I can always trace back the steps that led to a decision, which allows me to make targeted adjustments to the agents' prompts and behaviors.
2. Communication Barriers: Unlike human employees, AI agents don't have communication barriers, as they can all share the same "memory" or knowledge base, making it easier to scale and compound their capabilities.

Building AI employees is not as difficult as it may seem. In the next section, I'll show you an example of how you can create an AI Reddit marketer in just 10 minutes.

1. Radical Transparency: With AI employees, it is much easier to understand the decision-making process and the steps that led to a certain outcome. This allows for better alignment and adjustments to the agent's behavior.

2. No Communication Barriers: Unlike human employees, AI agents can seamlessly share information and work together without any communication barriers. This makes it easier to scale and compound the capabilities of the AI employee team.

3. Easier to Compound and Scale: Adding more AI agents to the team is relatively straightforward, as they can quickly learn and adopt new capabilities. This allows the AI employee system to be easily scaled and expanded to handle more responsibilities.

4. Specialized Agents vs. Generalist Agents: The author found that building a team of specialized AI agents, each focused on a specific task, was more effective than trying to create a single, generalist AI agent. This modular approach makes it easier to update and maintain the system.

5. Leveraging Shared Capabilities: Many of the fundamental capabilities required for different job functions, such as answering questions or scheduling meetings, can be reused across multiple AI agents. This reduces the overall development cost and effort.

In summary, the author's experience with building and deploying AI employees in a real business setting highlighted the benefits of adopting this approach, including increased transparency, seamless collaboration, and the ability to easily scale and compound the capabilities of the AI employee team.

I often heard people saying that Reddit is a great place to market about your product or content, but I never get a chance to do so. I really want to create an AI Reddit marketer who basically can be woken up every time when someone creates a Reddit post that is about AI agents, and then I want to trigger this agent who can look at this Reddit content, drop a message that is relevant to the audience, and provide some value, then softly plug the product or service that I want to sell. In my case, it will be my AI YouTube channel.

To achieve this flow, I create a group of Agents:

1. Reddit Post Finder: This agent will be triggered like every hour and try to search for the top five most relevant Reddit posts that are relevant to the keywords or topics of the product/service I want to plug in.

2. Reddit Comment Writer: This is the agent that we train specifically to craft the Reddit post content and have knowledge about the way I drop messages as well as the product/service that we want to sell.

3. Reddit Command Poster: This agent is responsible to actually post the comment on Reddit.

I also give a special function called "wait" because Reddit API has a pretty strict rate limit, so if you just keep calling the API endpoint for post command, they will block you pretty soon.

To build this Reddit marketing team, I will use a multi-agent framework, and there are quite a few of them. Two most popular ones are AutoGPT from Microsoft and Crisp AI. While AutoGPT provides the most flexibility, it is a little bit more technical to get started, while Crisp AI gets really popular because it is very easy to set up.

I will give you a quick walkthrough about how I set up this team of agents in Crisp AI, and in the end, for people who are less technical, I also show you how can you set up this exact multi-agent system without much coding on a platform like Anthropic.

Let's get started!

Building AI employees is a powerful way to automate and scale various business functions. The key learnings from the speaker's experience include:

1. Breaking down complex tasks into specialized sub-agents can make the system more manageable and scalable, as opposed to a single powerful agent.
2. AI employees provide radical transparency, as their decision-making process can be easily traced and adjusted.
3. AI employees eliminate communication barriers, as they can seamlessly share information and compound their capabilities.
4. Tools like Anthropic's Claude and Hugging Face's Gradio make it relatively easy to set up multi-agent systems for various use cases, such as automating Reddit marketing.
5. Incorporating reflection and reasoning mechanisms into the agents, as demonstrated with the "reflect" tool, can help improve their decision-making and performance.

Overall, the speaker's insights highlight the potential of AI employees to transform how businesses operate, and the relative ease with which these systems can be developed and deployed.