Using AI Agents to Research and Prototype Ideas

I think what I like most about these coding agents is how they can reduce the friction of things. Today I created a prototype for my gaming website that would retrieve my Xbox profile, presence, recently played games, and achievements.

The Old Way

Being a Senior Software Engineer - the old way to do this would have been:

• See if an API exists
• Hope it has a good "getting started" section and decent documentation.
• Set up any access needed (Microsoft / Azure account etc)
• Create some Python code to connect to the endpoint. Start basic.
• Add more code to validate each endpoint / feature I need to prove out.

I can do all this. This is my bread and butter and I've done this at many jobs over the years. But going back and forth with the documentation, the editor, and the tests takes time. It can get frustrating if you run into an error and need to troubleshoot and the documentation is lacking help in that problem area.

Enter the Agent

But the agent really turbocharges this process! I can create a loose (or heavily detailed if I have it in mind!) plan, and have the agent review it and help expand it into an actionable to-do list. Then I can have the agent scaffold the prototype, and even write the first pass of the code if I want it to. When there is a problem, we work it together. I have insights from years of troubleshooting and app development so I can usually spot or have a good sense of what the problem could be. If it's quick I investigate it and fix it right there. If it may be a little more involved, I can relay the info to the agent and set it on fixing it. Soon enough we have working code.

The Result

I think I was able to do all this in less than an hour. I think Copilot says it cost ~$2.00 with Sonnet 4.6 (we can save the 'which model is best' talk for another time).

As someone with a full-time day job, having an idea and actually seeing it through the same evening, without worrying about a possible huge timesink is genuinely energizing.