AI agents can enhance the results of Language Learning Models (LLMs) by breaking down tasks into a series of steps. Key design patterns for these agentic workflows include reflection, planning, multi-agent collaboration, and two-step workflows using LLMs with other tools. These workflows can be applied across various fields such as marketing, sales, customer success, and revenue operations, improving tasks like email drafting, sales call analysis, go-to-market strategy development, ad campaign creation, lead qualification, customer support, financial forecasting, and customer feedback analysis.

Instead of giving an LLM one instruction and getting a single response (zero shot), you can break down your task into a series of steps and have the LLM complete each step, like an agent.

Non-agentic workflow- In the example above instead of just having asking to write an essay and AI writes it all out and finishes.

Agentic workflow - writer, researcher, editor, grammar, critic, etc.

Agentic workflows can significantly improve the results you get from LLMs. In the example below, an agentic workflow was able to outperform a more recent LLM model (GPT-4) by using an earlier model (GPT-3.5) with a workflow designed for agents.

There are four key design patterns for agentic workflows:

• Reflection: Prompting the LLM to review its own code and fix errors.
• Planning: Having the LLM plan out a complex task by breaking it down into steps.
• Multi-agent collaboration: Using multiple LLMs working together, each playing a different role. An example is having one LLM act as a coder and another act as a reviewer.
• Two-step workflows (Tools + LLM): Using LLMs in conjunction with other tools. For instance, an LLM could be used to write code, and then a separate tool could be used to test that code.

Examples of AI Agents for GTM

Reflection

• Marketing: Have an LLM review a drafted marketing email for tone, grammar, and CTA (call-to-action) effectiveness. Instruct the LLM to identify areas for improvement and suggest revisions to enhance engagement.