Shaken, not Stirred: The 9 Questions You Need to Ask Yourself When Contemplating Building an AI Agent

The Agentic Era is Here: Should You Build an AI Agent?

Part 2 of my AI agents series moves from understanding the basics to practical application. Discover how the "JAMES BOND" approach can help you decide whether and how to build your first—or next—AI agent.

This isn’t about hype or FOMO; it's about making informed choices. By following my method, you'll create AI agents for the right reasons, not just to chase trends. Even if you’re not ready to implement AI agents today, this article offers actionable insights, including a custom GPT advisor built for this purpose. Join me on this practical journey into the world of AI agents and stay ahead of the curve.

AI agents are here! Well, it depends on how you define an AI agent… 

I define it as software that can work autonomously (to some degree), plan, and focus on clear goals while utilizing AI. To date, they are still expensive, relatively slow, and often inaccurate. In most cases, they are also less autonomous than you might be led to think. However, they are already a powerful and critical part of the evolution of AI. 

If you want to understand what AI agents are, what they are suitable for, and how they work, read part 1 of this series: The Era of AI Agents is Here - Helping You Make Sense of it. 

Why should you care?

After two months of deep-diving into AI agents (mostly because we're building one for coding), I've realized that timing is everything. Early adopters who move smart will win, while those rushing in blindly might make expensive mistakes. Staying entirely out of the game? That's risky too - you could miss crucial learning opportunities.

If agents are the future, I should start creating them in masses, right?

Wrong. Well, maybe. It depends. Just because everyone talks about AI agents doesn’t mean you have to build one. Not all use cases are suitable for AI agents. Not all teams and organizations are ready for AI agents. Even with the proper use case and a highly motivated team, you can still get it colossally wrong. 

OK, I’m feeling scared and excited in equal measure. What should I do?

“Before you rush into building an AI agent, I want you to stop, take a deep breath and think step-by-step…” Wait, why am I chain-of-thought prompting you as if you are an LLM? 

That’s because I want to prevent you from rushing into this. It will only get you more hallucinations than good results.

Here’s what I suggest you do to be more factual about your decision to build an AI agent: 

Introducing my version of “JAMES BOND”: The AI Agent Activation Checklist

I've developed a 9-question checklist to help you decide if building an AI agent makes sense. Think of it as your reality check before diving into the agent-building pool. 

This is a progressive list. After each step, you might abort the idea of building an agent (at least until things change). If you get to the end of the list and are still motivated to build an AI agent, not only will you be doing it for the right reasons, but you will also have many leads on how to build one for your needs. These questions apply whether this is your first AI agent or the thousandth. 

Here’s an infographic with the 9 questions, followed by more context and information on each question. I used the JAMES BOND acronym to make it memorable for you (and fun for me…). There is no need to memorize; read to the end to see how I made using this checklist extremely easy for you.

The "JAMES BOND" method in high level - what you need to ask to make a correct agentic decision.

Let’s elaborate what I mean by each question:

1. Judgment: Given an understanding of AI agents' pros and cons, should you consider building one for your business needs?

This is where you encounter the first reality check. 

Before deciding which agent to build, how, and through which tools, ensure you understand what AI agents are, their opportunities, and their limitations. Go back to the 1st article in the series if you need a refresher. 

Let me call out the main pros and cons of AI agents as I see them:

With these pros and cons in mind and knowing your business context, I encourage you to make the first decision: 

To agent, or not to agent, that is the question

2. Analysis: Business problem or opportunity, its properties, and suitability for Agents

This is where you encounter the primary ideation considerations. 

Whenever contemplating building or buying any AI product, consider the “problem or opportunity” space before jumping into the “solution space”. This will prevent building an agent for the wrong reasons or use cases. I am deliberately spending a lot of time here. This is a very critical decision junction. While planning and execution are key to agents’ success, you will fail if you go for an irrelevant use case. 

Some questions to help you analyze the business challenge/ opportunity:

• What are the existing business processes you have or would like to have? 

• Where are they falling short, or can they be significantly augmented?

• What will the impact and value be if we replace or augment these processes with AI agents?

Business needs first. Technology second.

After answering these questions and realizing there is a justification for considering agents, you need to understand if the business challenge or opportunity is suitable for AI agents.

Here are some properties of use cases that are suitable for agents:

• Can be done on a computer. If a human can perform the task using a computer, an AI agent can likely be built to do it as well.

• Are repetitive or tedious for humans. 

• Require 24/7 availability. AI agents don't need to sleep, eat, or take breaks to provide continuous service. This is particularly useful for businesses operating in multiple time zones or with a global customer base.

• Can benefit from personalization. AI agents can use data about customers to tailor their responses and actions.

• Rrequire access to a large amount of data. AI agents can be granted access to a wealth of information they can use to make decisions and act.

• Require a high level of specialization. While AI agents can make mistakes, they can be trained to achieve high accuracy for specific, well-defined tasks.

• Well-defined. The more specific the task, the easier it will be to build an AI agent to handle it. If you have a good understanding of the steps involved in a task, you are more likely to be successful in automating it with an AI agent. Moreover, an agent will most likely fail if experts struggle to fully articulate the entire process of executing a task. 

• Can be measured. Required for stirring agents' behavior and for measuring success.

Properties of a use case suitable for agents

3. Minimization: Will a simple automation or a prompt-based solution solve it?

This is where you encounter cost-effectiveness considerations. 

Before you go “full agentic mode,” I implore you to ensure there are no other options to get comparable or good enough results. In many cases, I’ve seen teams tempted by the “sex appeal” of the technology, rushing to ask: “How can we use it?” This is not the right question. Instead, you should ask:  “Is this the right technology?” 

Ask yourself:

• Would a simple “if-then” type of automation around your business process solve all or most problems? If so, don’t build an AI agent. 

• If not, when AI is required due to complex planning or reasoning needs, would a straightforward, prompt-based solution suffice?  If so, don’t build an AI agent. This is especially true if you are new to generative AI:

Advice for beginners

4. Evaluation: Can you plan and measure it?

This is where you encounter primary feasibility considerations. 

When building an agent, you must define the tasks the agent is expected to perform and how success will be measured. At this point (a.k.a. The “Agent Activation Phase”), I’m not asking you to define these fully. Instead, I’m asking you to review some guiding questions to ensure that when the time comes to build your agent, you can do it. 

Don’t be tempted to rush through these questions. If you overlook challenges in planning and evaluation now, you will encounter them soon enough. These will most likely become your showstoppers or cause for most grief later. 

If you answered all these questions positively, move to the next set. Otherwise, abort the current mission (and improve your business definitions to become agent-ready in the future).

If you've come this far, I assume you have a good task and potential for high value. The remaining questions address some productization and implementation questions. These are intended to ensure you can build a production-grade agent that can mature beyond a “cool pilot.” They will also help you understand the complete ROI equation, including the investment. 

5. Structure: Can you integrate the agent into your flows?

This is where you encounter job security considerations. 

At this point, I implore you to consider whether the AI agent can integrate with the existing tools and processes and mash well with human users. Over time, you will also need to ask yourself how it will integrate with the other agents in the ecosystem. By integration, I mean access to data and connection to all relevant interfaces. Beyond the integration, you also need to define the expected performance. The agent is probably not the way to go if you expect an instantaneous result. 

Humans-in-the-loop

6. Boundaries: Can you sufficiently protect yourself from mistakes? 

This is where you encounter reputation and legal considerations. 

Agents make mistakes. So do humans. At this point, humans are better at recovering from mistakes. To bridge this gap, you will need to create thoughtful, elaborate guardrails and recovery plans for your agents. 

Ask yourself:

• Considering your industry and use case, what is your tolerance for agents' mistakes (quantity, quality, and compliance-wise)? 

• Which autonomous actions will you let the agent take on your behalf?

• Are there sufficient guardrails you can put in place?

• Is there a procedural way humans can help prevent or recover from agentic mistakes?

• Do you have other plan B you can automate?

7. Orchestration: What is the complexity required?

This is where it starts to get complicated. 

Assessing the expected complexity of an agentic system is an important reality check before building it. 

Here are some causes for high complexity system:

• Single vs. Multi-agent Systems: While some tasks are relatively simple and can be accomplished by a single agent, there will be a need to build multi-agent or advanced systems in other cases. Such cases require the tools and know-how. They are often much more prone to mistakes and thus require more supervision and careful planning. 

• Precision requirements: Some industries and tasks require high precision and low fault tolerance. Examples include governed industries like healthcare and use cases where the cost of mistakes is high, like manufacturing line decisions.

• Other complexity elements: some use cases can be complex for different reasons. Including: 

ROI and reality check

8. Normalization: Can you productize it sustainably? 

This is where money talks. 

Regardless of your direction in the next question, building and sustaining an AI agent is costly, at least in the near future. Try to project the usage of your agent in growth scenarios. Remember that agents will have multiple and varying amounts of LLM calls. Ensure you have the budget and skills to productize, monitor, and improve the agent over time. Increase your estimated investment even further: Harnessing the power of an emerging technology is always more complex and expensive than initially anticipated. 

To agent, or not to agent, that is the question 2.0

Assuming you are still convinced about your desire to activate an agent, now comes the decision that will impact the implementation the most:

9. Decision: Should you build, buy, or in-between? Is there a tool (agentic or not) for it that you can use?

This is where skills, market, and strategy meet.

A 'build vs. buy' decision significantly impacts your agent. Here are some key pointers. I plan to dedicate future articles to this topic with much more detail.

Buy vs. build showdown

Whether you build or buy, it is beneficial to start with more straightforward AI projects and gradually increase complexity as your team gains experience. Experimenting with different approaches and tools will help determine what works best for your organization. Remember, the most crucial factor is not simply building or buying an agent but ensuring it effectively solves the business problem and delivers tangible value.

Once you have completed answering all the questions, you might wonder: 

Where do we go from here?

1. Next time you contemplate activating an AI agent, get help from JAMES BOND. To make it even more accessible, I built a very helpful custom GPT to serve as your advisor and walk you through these 9 questions. Let me know in the comments if you found it useful. 

2. If you have decided to build an AI agent, first of all, I’m excited for you! Second, read my next article, where I will discuss some of these best-known methods and know-how for planning and building your agent. 

Need more help?

Talk to me. I can help you through this process.