Moving from one AI assistant to a team of agents

Last week I shared how I started using AI as an "orchestrator." This week I want to explain why I actually did it, and why I moved away from just using one all in one assistant to building a team with OpenCode.

For me, it wasn't just about getting better code. It was about having more control.

When you use a standard AI assistant, you are basically renting someone else's brain. When you use an open source tool like OpenCode, you are building your own department.

Why I needed more than just "Autocomplete"

I started with GitHub Copilot, like most people. It is a great tool for helping you while you type, fixing small bugs, or writing unit tests. I still think it's the best tool for that.

But when I started building this portfolio, I realized I needed something more. The problem wasn't that the AI was "dumb." It was just that the chat window was too small for what I wanted to do.

Copilot is designed to be an assistant. It sits in your editor and gives you tips. But I wanted agents that could actually do things:

• Run commands in my terminal and check if the site is actually working.
• Do several steps in a row without me having to ask every time.
• Work on different parts of the project at the same time.
• Be customized so they know exactly what my project needs.

I didn't need a smarter chat bot. I needed a different tool for a different job.

Why I like the Open Source way

Switching to OpenCode felt like finally getting the keys to the car instead of just being a passenger. It's a tool that lives in my terminal and can actually run commands.

There are three main reasons why I prefer this way:

1. I can pick the best model for the job

With Copilot, you usually use whatever model they give you. With OpenCode, I use my own API keys. This means I can swap the "brain" of my agents whenever I want.

• If I need really smart logic, I use Claude.
• If I have a huge file to refactor, I use Gemini.
• If I just need something quick and cheap, I use GPT-4o.

This is important because things change fast. Recently, Anthropic started blocking some tools that were trying to use their cheap subscription for heavy work. Because I use my own keys, I don't have to worry about that. I'm not locked into one company.

2. Learning how it all works

Using OpenCode forced me to actually learn how these AI systems work. I had to think about four things:

• The Setup: My terminal is where the work happens. I had to learn how to keep my agents safe so they don't break anything.
• The Data: I learned that giving the AI too much information is actually bad. I have to pick exactly what each agent needs to see.
• Security: I hold the keys. No one else is reading my code or my prompts. I know exactly where my data is going.
• Workflows: I can set it up so that if one agent fails a test, another agent automatically tries to fix it. You can't really do that in a simple chat.

3. Having a "Team"

The best part is that I don't just have "one AI." I have a team of specialists. For this portfolio, I set up:

• Front-End Specialist: He knows Next.js 16 and Tailwind 4 inside out.
• Devops Architect: I chose Sanity.io and Vercel because they are easy to use. This agent helped me with the tricky parts, like making sure content updates without me having to redeploy the whole site.
• Content-Copywriter: This one helped me with my CV and reviewing my content.
• Code Reviewer: This is a separate agent that just checks for linting, typos or security issues. He doesn't get distracted by the other work.

Each agent has one job. It's much cleaner and I get much better results this way.

A quick reality check

I'm not saying OpenCode is perfect for everyone.

If you work in a big company, you probably have to use AWS Bedrock or the official Claude Code. Those are built for big businesses with lots of rules and support. OpenCode is still new and it's built by the community.

But I'm a solo developer. I'm building this portfolio to learn. For me, the freedom to experiment is more important than having a big company support me. I'd rather pay for what I use and have the freedom to change things when I want.

Is it too complicated?

Some people say this is too much work. "Why not just use Cursor and be done in a day?"

They are right, if you just want to finish the project fast, use the simpler tools. But my goal isn't just to finish the portfolio. My goal is to learn how to be a proficient and well informed AI Engineer.

The "extra work" is actually the point. I'm learning how to manage these agents now, so I'm ready when I need these skills for bigger professional projects.

Conclusion

GitHub Copilot is still a big part of my workflow. I use it every day for quick help and simple prompts. To be honest, their API allowance for the Pro subscription is very generous, so it makes sense to keep using it and plug into OpenCode.

But for building a whole system, I prefer having a team.

The future of coding isn't just about having a better chat bot. It's about building a workflow and a team that works for you. And I like being the one who picks the team.