← Back to blog

How I Use Claude Code as My AI Teammate, Not Just Another AI Agent

How I Use Claude Code as My AI Teammate, Not Just Another AI Agent

Everywhere I look, people are talking about AI agents.

Build an agent to do this. Create an agent to automate that. Build an entire team of AI agents that can run your business while you sleep.

And while I’m incredibly excited about what agents can do, I’ve started to realize that I approach AI a little differently.

I don’t think of Claude Code as my agent. I think of Claude as my teammate.

And there’s an important distinction.

An agent is typically something you give a task to and expect it to execute with some level of autonomy.

A teammate is someone you collaborate with. Someone you communicate with, plan with, challenge, and learn to trust.

That’s how I approach building software with Claude Code.

I still bring the ideas, the vision, the experience, and the responsibility for what we’re building.

Claude helps me take those ideas and turn them into working software, often at a speed that would have been unimaginable just a few years ago.

I don’t use AI to replace my thinking, I actually use AI to amplify my ability to execute on what I am thinking about. That’s a key distinction.


But there’s a process behind how I do this.

And I want to share five of the most important parts of that process because I think understanding how to work WITH AI is becoming just as important as understanding what AI can do.

1. Clarity Comes First: I Direct the Vision

One of the biggest mistakes I think people make when working with AI is jumping straight into asking it to build something.

They get an idea, open an AI tool, and immediately start prompting.

But I’ve found that the more clarity I have before I start building, the better the result.

When I get an idea, I usually start by thinking about what I’m actually trying to accomplish. 

  • What problem am I solving?
  • Who am I solving it for?
  • What should the experience look like?
  • How should it work?
  • And how does it fit into the bigger system I’m already building?

Sometimes I need to talk through an idea before I even know exactly what I want.

That’s okay.

I might capture my thoughts, organize them, and work through the possibilities before I start asking Claude to implement anything.

Because I don’t want AI to make every decision for me.

I want AI to help me execute the decisions I’m intentionally making.

And that starts with clarity.

2. Claude Is My Teammate: We Plan and Build Together

Once I have a clearer understanding of what I want, I bring Claude into the process.

But I don’t always immediately tell it to start writing code.

Often, I’ll ask it to create a plan first.

I want to understand how Claude proposes solving the problem.

  • What components does it need to build?
  • What existing systems will it interact with?
  • What changes need to happen?
  • Are there security considerations?
  • Is there a simpler or more efficient way to accomplish the same thing?

Then we review the plan together.

Sometimes I agree with everything Claude recommends.

Other times, I’ll challenge a decision, ask for another approach, or explain why I want something built differently (actually do the last part quite often).

And this back-and-forth is where I think a lot of the real value happens.

I’m bringing my experience as a software engineer, my understanding of the business, and my vision for the product.

Claude is bringing its ability to analyze code, propose solutions, and rapidly implement technical changes.

Together, we can accomplish much more than I could by doing every step manually.

And here’s another important point.

Claude can spin up additional agents to work on specific tasks when it makes sense.

But I still think of Claude as my primary teammate.

I’m directing the work, reviewing the approach, and maintaining an understanding of how everything fits together.

The goal isn’t to remove myself from the process. It’s to increase what I’m capable of accomplishing within it.

3. Trust Is Essential, but Verification Is My Responsibility

This is something I’ve been thinking about more and more.

When you’re working with AI to build real software, trust becomes incredibly important.

Especially when that software is being used by actual customers.

I can’t afford to blindly accept code, push it into production, and hope everything works.

There are real consequences when something breaks.

Customer experiences can be disrupted. Sensitive information can be exposed. Businesses can lose money.

And ultimately, I’m responsible for the systems I build and operate.

So when I say I trust Claude, I don’t mean that I assume everything it produces is automatically correct.

I mean that I’ve developed confidence in how we work together.

I understand its strengths, and I know where I need to pay closer attention.

I can review its proposed architecture, ask questions, request tests, and evaluate the changes it’s making.

And I can build processes that help catch problems before they reach production.

Trust doesn’t mean removing oversight. It means building a working relationship where you understand what you can delegate and what you still need to verify.

That’s true when working with human teammates, and I think it’s equally important when working with AI.

4. My Daily Tools: Claude Code in the Terminal and Git

Something that might surprise people is that I don’t primarily use Claude through a desktop chat application.

My go-to environment is Claude Code running directly in the terminal.

And I think that’s an important part of why I can move so quickly.

The terminal gives me a direct way to work with my projects, files, development tools, and code.

Instead of constantly copying and pasting code between different applications, I can work within the environment where the software actually lives.

Claude can inspect the project, understand existing code, make changes, and help me run the tools needed to build and test it.

But there’s another tool that’s just as important to my workflow.

Git.

I use Git every day. It’s the best, most valuable tool I know how to use.

It’s essential to my business and keeping myself sane as I build code with AI and Claude.

And I think anyone who wants to seriously build software with AI should learn the fundamentals of version control.

Because when you’re building quickly, things change quickly.

You might implement several features in a day, experiment with different approaches, or discover that something you built earlier needs to be changed.

Git allows you to track those changes, create checkpoints, work on separate branches, and return to previous versions when necessary.

It’s a critical part of maintaining control over what you’re building.

But Git alone doesn’t guarantee that your software is stable or secure.

You still need to test your code, protect sensitive information, review changes, and manage how updates reach production.

I don’t want to finish a full day of building and then worry about whether everything I’ve created is going to break.

I want a development process that allows me to move quickly while maintaining confidence in the work.

AI helps me build faster. Git helps me manage what I’m building.

And together, they’re an incredibly powerful combination.

5. Build a System Around Your AI Development Process

This might be the most important lesson of all.

As you start building more software with AI, you’ll quickly discover that writing the code is only one part of the process.

  • You need to organize your projects.
  • You need to understand where your code lives.
  • You need to manage access to your systems.
  • You need to protect credentials and customer information.
  • You need to know how your applications are deployed.

And you need a reliable way to make changes without accidentally breaking something that’s already working.

This is what I think of as digital organization and digital hygiene.

And it’s something that becomes increasingly important as you grow.

Imagine you’re working on five different projects, each with its own codebase, database, hosting environment, API credentials, and deployment process.

Without structure, that can become overwhelming very quickly.

And AI can actually accelerate the chaos if you don’t have a system for managing it.

That’s why I believe we need to teach people more than just how to prompt an AI tool.

We need to teach them the fundamentals of how software is organized, developed, tested, secured, and deployed.

They should understand what a repository is.

They should know what a branch and a commit are.

They should understand the difference between development and production environments.

They should know why permissions matter and why not every tool or agent should have access to deploy changes.

And they should understand how all of these pieces fit together.

You don’t necessarily need to become a professional software engineer to benefit from these skills.

But having a foundational understanding of how software works makes you a much more capable builder.

The more capable you become at directing and managing the process, the more effectively you can leverage AI to build.

The Future Isn’t Just About AI Agents. It’s About Learning to Work With AI.

I think we’re entering a fascinating period where the relationship between humans and technology is fundamentally changing.

For much of my career, building software meant spending enormous amounts of time translating ideas into code.

Now, more of that translation can be handled by AI.

Which means my role is changing.

I’m spending more time thinking about what should be built, why it matters, how the pieces should work together, and whether the result meets the standards I expect.

And less time manually writing every line of code.

But that doesn’t mean my expertise is becoming less valuable.

In many ways, I think it’s becoming more valuable.

Because when the ability to create becomes widely accessible, the ability to direct that creation with clarity, experience, and responsibility becomes a major differentiator.

That’s why I don’t just want people to learn how to use Claude Code.

  • I want them to learn how to become effective collaborators with AI.
  • To develop the confidence to explore their ideas.
  • To understand the systems they’re building.
  • To maintain control over their work.
  • And to create technology that actually serves people.

Because ultimately, the goal isn’t to have AI do everything for you.

The goal is to become capable of building things you previously couldn’t, while staying intentional about what you’re creating and responsible for what you put into the world.

← Back to all posts