We’re in a time of accelerated innovation. Some say the AI bubble is going to pop, while some are betting everything on it being the future. As a participant in the development industry, sticking your head in the sand can be fatal. Yet, hopping on the hype train can seem intimidating or feel aimless at times.
I mean, there is so much to try, and fast, because it’ll be different by next month – next week even. Where does one start?
So, from one developer to another, here’s what my journey to find an agentic framework has included—along with a gut-rated list. I haven’t landed on the perfect solution yet, but that’s what’s expected with innovation: continuous learning.
To make it as tangible and digestible as possible, I’m just going to hit you with each tool’s name, a quick and informal description, and what stood out about it. Let’s go in order of discovery. That way, I think we’ll uncover a pretty clear direction of where things are going.
- IDE w/in-chat agent + MCPs (example: VSCode w/Claude Code CLI OR Cursor w/select model)
- Warp
- Claude Code
- GSD
VSCode w/ChatGPT Extension
- The classic beloved IDE with the agent, rather simply, offering code right by your mouse cursor.
- Feels like ages ago now, and I will admit I don’t prefer how it constantly showed faded code before I type, often chunks of logic that seemed grounded in little to nothing.
Warp
- I already had it as my terminal equivalent and found the agent chat feature later – there is more to it than I have tapped into as well.
- Seems more accurate with bash, git, terminal commands versus the other strengths of my in-IDE codebase knowledge setups.
Cursor
- An IDE built with agentic development in mind, where both inline and side chats are an option.
- It initially started for me in a similar way to ChatGBT, for syntax or function-by-function assistance. Then it grew and grew in autonomy and elevated to chats that offered planning as well as much larger development integration.
Leveraging MCPs
- If you haven’t heard of or leveraged MCPs yet, click the link!
- For me, I’ve really been loving the easy detail-fetching setup from these, especially for Figma — and it can offer a nice full connection if also adding Jira/Notion (from acceptance criteria, to design, to code).
Claude Code CLI
- Claude Code can come in many forms; it’s a desktop app, in your terminal, or (for an in-between of the Code CLI and the Desktop visual) as an extension in VSCode.
- This option has led to the most agent autonomy I’ve tried, and while I still don’t trust AI to handle large commits of code, I truly do see the potential of spawning multiple subagents to tackle larger, pre-defined tasks.
GSD (Get Shit Done)
- It’s an open-source library that offers a framework for spec-driven development; specifically, it helps structure the work by breaking things into certain commands and consistent terminology
- I’ve only been working with it in the past month or so, but I’ll admit that I’m not sold on it. However, I have learned a lot already, so I can see its pros and where this angle of AI-accelerated development has serious potential. (Which truly could be a spin post in and of itself.)
Do we see the pattern? Yes, constant change. But more than that, an increased amount of agent autonomy. More emphasis on planning, pairing with the agent, good documentation, and larger bites of automated code. I’m not yet signing off on this direction going full-send, because I’m also seeing it come with codebase congestion, poor version control, and engineers who could easily lose that joy (and crucial skill) of deep understanding.
It is a balance we are all striving to be at the forefront of discovering.