A person walks onto your website...

Mixture of Agents (MOAT) - A Two Part Series is sponsored by Agent.ai - Discover, connect with and hire AI agents to do useful things.

February marks the 1 year anniversary of founding my own company and what started out as an experiment to bring GenAI to GTM is turning out to be a viable business solving real business challenges.

That being said the journey of figuring out how to solve real business challenges in GTM with GenAI has evolved significantly since I first went down this path.

Hypothesis #1 - Everyone will be a prompt engineer. We should all become experts at leveraging ChatGPT to solve business problems. FALSE. We want sales people to sell, not be prompt engineers. ChatGPT is quite useless in solving real GTM challenges as it does not have access to data and it's ability to build complex workflows to automate work is limited.

Hypothesis #2 - Agents will be key to solving real challenges in GTM. TRUE. Agents combine workflow, tools to access data, knowledge in your enterprise and prompting via LLMs.

For a quick refresher on Agents (here is a visual). I will stay away from the "if it is not autonomous" then it is not an Agent conversation for the purposes of this discussion.

Hypothesis #3 - I can solve most GTM use cases with one Agent Builder product. I thought this might be Clay.com. But over time what became obvious is that it is a very fragmented agent ecosystem and there will be no '1 ring to rule them all'. FALSE

Hypothesis #4 - We will orchestrate across agents from multiple vendors to solve real GTM use cases. TRUE (my current prediction). As anyone who makes predictions about anything AI, time will prove you wrong.

To show what Hypothesis #4 looks like in action, let's take a use case that is common for a few of my clients -

• Someone visits your website, but does not log in. They click around and then leave
• This could be worth someone to reach out to but how do you know who they are?
• There are products like Warmly.ai, RB2B.com that try and find with a degree of certainty which companies came to your website and with a lesser degree of certainty, who came to your website. That being said, if they can identify even 20% of your website traffic (this is US only), that is a decent chunk of leads.
• Now that we know who has come to your website, you want to get more information about them doing web research. You could enrich their information by integrating with a data provider.
• Now that you know a bit more about them, you want to find out if they fit your ICP (Ideal Customer Profile) and if they do, why.
• Lastly, pass on these warm leads to your sales team via Slack or generate personalized emails that you can use to put them in an email sequence.

As is true in all these use cases, I could do a lot of this in products like Warmly.ai or Rb2B.com. But invariably, as was the case with my customer, the way they determine the ICP and the kind of additional enrichment they want about the company and prospect, is not met with the out of the box functionality. So now we are in a multi-agentic world where I need to go to a lead enrichment agent and an ICP fit agent and a personalized outreach agent.

Will we get to world where either the AI SDR companies or the Warmly's of the world figure out personalization to a high degree of depth and in a way that is unique to my business, remains to be seen, but till then I have to orchestrate a series of agents, each really good at what they do.

The way I solved this for my client was via Warmly.ai / RB2B.com -> Clay.com -> Agent.ai Enrichment agent -> Clay.com -> Agent.ai ICP fit agent -> Clay.com -> Slack.

The Warmly.ai / RB2B.com to Clay.com integration is out of the box. It is a configuration. Clay.com calling Agent.ai agents is out of the box via HTTP API. The Clay.com to Slack integration again is out of the box. So all I had to do was build the Enrichment agent in Agent.ai (I could have used Clay credits but I did not want to). The ICP fit agent is a complicated flow as it checks for fit on 3 areas - LinkedIn Profile fit, LinkedIn Activity fit, Company Initiatives fit. These are then combined to generate an overall fit and reason and that is sent back to Clay.

Here is my Agent.ai Fit agent -

The flexibility I get with Agent.ai is that I can write very specific and tailored ICP fit agents. Plus I don't have to use Clay credits or struggle with Clay's UI when it comes to building complicated agents. Debugging is much easier in Agent.ai.

Why am I such a big fan of these multi-agentic systems - I can solve real business needs in GTM. Everyone can benefit from finding out the right warm leads from your website visitors and conducting outreach.

Secondly, I use each tool to do what it does best - Clay is great at orchestration, Agent.ai is great at Agent building and Warmly/RB2B are great at capturing website visitors. Together I get the best of all these products.

If you interested in seeing this in action, here is a link to a video - https://youtu.be/0m3fJnYUBHw

So my call to action for anyone who has got this far is to go sign up for Agent.ai and start building some agents and create your own multi-agent system.