Agentic Engineering: Collaborate with AI

I remember when I first started integrating AI tools into my workflow—like opening a door to a whole new world. But here's the thing: it wasn't just about using AI; it was about working with it. That's where agentic engineering comes into play. In today's rapidly evolving tech landscape, leveraging AI isn't just about automation; it's about collaboration. Engineers are moving beyond traditional usage to a more integrated, agentic approach. In this article, I'll walk you through how I and many others are making this shift. We're talking about adopting and using AI tools, the evolution of AI models in software engineering, and the importance of context management in AI engineering. And, crucially, how we're orchestrating this with well-configured AI agents. Because yes, we're at that point: no longer passive users, we're active collaborators.

Adopting AI Tools: A Practical Approach

As an engineer, I often experience the excitement surrounding AI tools. However, while 90% of engineers have used these tools, only 50% use them regularly. Why not more? The answer often lies in practical integration into daily workflows. For me, adopting an AI tool is like adding a new team member. It's not just about installing software, but about integrating it into the existing workflow.

To start, I mapped out repetitive and tedious tasks in my workflow. Then, I explored how tools like GitHub Co-pilot could automate them. For instance, in code generation, Co-pilot has revolutionized my productivity by suggesting entire functions. But watch out, adoption isn't without pitfalls.

• Don't jump on every AI suggestion indiscriminately.
• Avoid over-reliance on automated suggestions.
• Constantly evaluate the relevance of suggestions to the project context.

Evolving AI Models in Software Engineering

With the evolution of AI models, we've moved from static systems to dynamic systems capable of executing tasks autonomously. This impacts our daily engineering tasks directly. For instance, in 2025, AI models began executing tasks independently, including submitting pull requests. In my experience, this transformed how we approach projects.

I recall a project where AI suggested not just code, but also performed tests and proposed changes. It was a real game changer. However, it's crucial to balance innovation with stability during AI model updates.

• Monitor frequent updates to avoid unexpected disruptions.
• Consider compatibility with existing systems.
• Ensure the team is trained on the new capabilities of AI models.

Mastering Context Management in AI Engineering

Context management is a subtle yet essential art in AI engineering. Poor management can lead to quality degradation of over 50%. The Model Context Protocol (MCP) plays a crucial role here. In my work, I learned to fill the context window with only relevant information.

For instance, during a project, I had to manage parallel agents, which has become increasingly common. Adding incorrect or excessive context can make the model less effective. Here are some practical tips:

• Limit context to essential information.
• Avoid overloading the agent with unnecessary details.
• Use MCP to structure interaction with tools like the GitHub API.

Agentic Engineering: Collaborating with AI

Agentic engineering represents a fundamental shift in our interaction with AI. Instead of just using these tools, we collaborate with them. For example, in a recent project, I worked with an AI agent capable of reading and applying best practices from Stack Overflow.

However, AI lacks judgment. It doesn't understand business context or specific architectural decisions. Armand, the creator of Flask, gained more than 30% of his time thanks to AI, but he knows what to keep for himself. That's the difference between using AI and collaborating with it.

• Direct the AI's work rather than blindly accepting every suggestion.
• Identify the limits of the agentic approach and necessary trade-offs.

Implementing the Research Plan Implement Loop

The research-plan-implement loop is a powerful method for enhancing AI engineering processes. It starts with thorough research, followed by structured planning, then implementation. In my practice, I've seen how this approach improves accuracy and efficiency.

During a project, I first mapped tasks into small steps, which allowed for better management and error reduction. However, challenges remain:

• Planning can be lengthy but is essential to avoid costly errors.
• Adjustments are often necessary along the way.
• Iteration is key: each cycle improves the next.

Agentic engineering has been a game changer for me, reshaping how I collaborate with AI in daily workflows. First, I realized it's not just about using tools but truly engaging with them. That's a big mindset shift. Then, I noticed that only about half of engineers are using AI tools regularly—there's so much untapped potential for efficiency here. By managing context more strategically, I saw real gains. It's like having four different ways to work with my internal platform APIs. But watch out—you need to know the limits to avoid performance pitfalls. Now, I'm convinced that integrating these strategies into your workflow can really make a difference. So, share your experiences and let's evolve together. For those wanting a deeper dive, check out Brendan O'Leary's full video: Agentic Engineering: Working With AI, Not Just Using It. It's a real eye-opener.