To Agent or Not to Agent?
1. To Agent or Not to Agent?
Welcome back! So far, we’ve learned about the key components of AI Agents. That said, while agents can be really powerful, they're not always the right tool for the job. So when should you adopt agentic systems? And when should you not?2. A Tale of Two Customer Support Teams
To put this into context, imagine two different customer support teams. They both work at a retailer and want to augment their operations with AI. For customer support team A—80% of the support tickets are variations of: "How do I track my order?" "How do I return an item?" "How do I change my shipping address?" For customer support team B, 80% of tickets are complex issues like: "I was charged twice, but one order was cancelled, and I have store credit from a previous return that wasn't applied correctly."3. The Tale of Customer Support Team A
Taking a step back, 80% of the tickets the customer support team A faces have the following qualities in common: They require simple decision-making, they do not require accessing customer information and history, they all have discrete, predictable answers.4. The Tale of Customer Support Team B
On the other hand, the majority of customer support team B’s tickets have the following qualities in common: They require complex decision-making, they require accessing customer information and history, they require adaptive solutions.5. A Tale of Two Customer Support Teams
In this context, while both could benefit from AI augmentation, customer support team A does not need an AI agent and can opt for a simple chatbot that answers questions based on pre-trained knowledge—no tool use or action is required. Customer support team B, on the other hand, would benefit from an agentic solution that can access customer data, generate remediation strategies, implement them, and update customer support systems with them.6. When to Use AI Agents
In a nutshell, you should consider adopting agentic solutions when the problems you’re trying to solve require complex decision-making, rely heavily on unstructured data, have difficult to maintain rules, and require adaptive problem solving. Examples here include customer support systems like the ones customer support team B needed, coding assistants that can read code bases, provide updates, and implement them automatically, or deep research assistants that can take break down a research tasks into different steps, perform web search, access research sites and publicly available data, and synthesize results.7. The AI Agents Tooling Ecosystem
The next step after adopting AI agents is understanding the tooling ecosystem.8. The AI Agents Tooling Ecosystem
The AI Agents tooling space is dynamic and quickly changing. That said, you can see it operating on a spectrum that goes from: Off-the-shelf tooling that allows you to use an agentic system to tackle a specific problem—such as AI assisted coding, or deep research,9. The AI Agents Tooling Ecosystem
to low-code/no-code tooling that will enable you to build low agency maturity level workflows. Think of these as the next generation of workflow automation tools.10. The AI Agents Tooling Ecosystem
And finally, to AI agent frameworks that let you build agentic systems from scratch—which are most widely used by developers to build truly robust systems.11. The AI Agents Tooling Ecosystem
Each set of tools provides advantages and disadvantages, including ease of use, the type of use cases enabled, and customizability.12. Build vs Buy: A Framework
Like with most software, the choice between building and buying comes down to a few different factors. You should consider buying off-the-shelf tools if you're tackling a specific domain or use-case, there is already a mature, well-tested solution in the market, you want to minimize maintenance overhead.13. Build vs Buy: A Framework
You should consider buying low-code/no-code platforms if you need some customization but not complete control, your workflows are moderately complex but follow common patterns, you want business users to modify the agent without engineering help, you need to integrate with existing systems quickly.14. Build vs Buy: A Framework
And finally, you should build with agent frameworks from scratch if your use case involves proprietary systems, you're handling sensitive data, the agent is core to your competitive advantage, no existing solution meeting your specialized requirements, and you need complete control over the agent's behavior and evolution.15. Let's Practice!
But for now, let’s put our skills to use.Create Your Free Account
or
By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.