February 2024: CursorIDE, Prompting Fatigue, and Building Tools to Push Through
“By February 2024, ChatGPT was my second monitor. CursorIDE became my third hand. But even prosthetics get tired.”
— Wing
đź’» Not a Valentine’s Post#
This February wasn’t about hearts or roses. It was about prompts, diffs, and the quiet frustration of watching AI almost do what I needed—again.
CursorIDE entered my workflow with promises of tight integration, inline prompting, and smart code rewrites.
It was sleek. Fast. Magic, at first.
But the more I used it… the more I started feeling the seams.
⚙️ The Friction of “Almost Right”#
- ChatGPT? Still my go-to for reasoning, refactoring, idea exploration.
- Cursor? Great for local context, sidekick edits, real-time diff magic.
But Cursor didn’t feel like a partner. It felt like a shortcut that sometimes got bored mid-sentence.
No consistency across sessions.
No memory across files.
And if your prompt wasn’t surgical—your results were spaghetti.
So I did what any frustrated dev does:
I started writing tools.
đź§° Tools to Overcome Tool Laziness#
I hacked together:
- A scratchpad that forced Cursor into deeper chain-of-thought behavior.
- Macros to break work into modular prompts and chain responses.
- A system to track code attempts across sessions—because Cursor sure wasn’t doing that for me.
It worked.
But it also broke things.
Cursor started to feel like a CLI wrapped in denial—incredible for bursts, but clunky for architecture. And the more I built around it, the more I saw that it wasn’t designed to scale with me, only to serve me at the moment.
đź§ The Real Problem#
Cursor wasn’t lazy.
It was obedient.
And obedience without memory or context is a bottleneck masquerading as speed.
- It’ll refactor.
- It’ll fix errors.
- It’ll rewrite your function header three different ways—without noticing the one below already works.
Meanwhile, ChatGPT—detached, floating in its own bubble—understood intent, but couldn’t see the code in your editor unless you copy/pasted it like a medieval scribe.
🧓 Wing’s Mid-Feb Rant#
We’re not using assistants anymore.
We’re building scaffolding for tired AI.
Cursor does what you say—until you realize what you said wasn’t enough.
ChatGPT explains everything—until you hit the copy/paste wall.
And me? I’m caught in between, writing glue code for systems that still can’t talk to each other intelligently.
⚠️ What I Learned#
- IDE-side AI is good for micro-work.
- Chat-based LLMs are better at high-level architecture.
- Neither is great at remembering what you’re building unless you become the memory.
And that’s the weird part:
The better the tools get, the more I need to become the integration layer.
đź’¬ Working with Cursor? Writing meta-tools to bridge the prompt fatigue? Drop me a line.
We’re all duct-taping the future together—might as well compare adhesives.