This post is the start of a new series, Building Successful AI Agents, where I explore practical frameworks, methods, and tools for creating effective AI agents. We begin with insights from Chapter 8 A Practical Guide For Building Successfule AI Agents of Agentic AI: Harnessing AI Agents to Reinvent Business, Work and Life by Pascal Bornet and colleagues. That chapter 8 offers a step-by-step approach to building AI agents from a business perspective. In this post, I’ll briefly cover Steps 1 to 4. You could compare and contrast project management’s approach and their Five Project Phases. You will find similarities.

We have another series of posts called Building Agentic AI.

Step 1 – Finding the Right Agentic Opportunities

The journey starts with identifying where an AI agent can create real value. This means looking for processes or workflows that could benefit from autonomy, proactivity, and adaptability. The right opportunity is one where an agent can either make something faster, cheaper, more accurate, or open up entirely new possibilities.

Step 2 – Defining the AI Agent’s Role and Capabilities

Once the opportunity is chosen, clearly define what the agent will do — and just as importantly, what it will not do. This includes deciding on its goals, scope, and key capabilities. Without this clarity, the project risks scope creep and misaligned expectations. This is an appropriate time to look into the levels. Pascal Bornet has an Agentic AI Progression Framework.

Step 3 – Designing AI Agents for Success

Bornet’s book lists five design principles for creating effective agents:

1. Start with the End in Mind: Be clear about the agent’s goals and success metrics. “To-be”
2. Understand Your Current State: Know your existing processes, tools, and constraints. “As-is”
3. Design the Target Process: Plan the new workflow with the AI agent fully integrated.
4. Choose the Right Architecture: Select a technical setup that fits the problem, whether no-code, low-code, or full-code.
5. Build in Human–AI Collaboration: Design for humans and agents to work together, each doing what they do best.

Step 4 – Implementing Your AI Agents

This is where the planning meets execution. Implementation involves building the agent, testing it iteratively, and integrating it into real workflows. Bornet emphasizes the importance of starting small, running controlled pilots, and scaling up once the agent proves its value. Implementation is not just about coding — it includes training staff, adjusting processes, and monitoring for unexpected outcomes.

Deploying agents is a lot more difficult than building them, according tp Pascal Bornet.

Plan Do Check Act (PDCA)

We have a whole series of posts on plan, do, check, and act starting with PDCA Cycle: A Model for Learning and Improvement.