I'm researching the real friction points in agentic coding — specifically with tools like Cursor. Just five honest scenarios.
Pick what resonates, and drop a comment with the why.
🧠 The Context Tax
You're mid-flow. The agent asks for more context — again. You stop coding and start explaining.
How much of your session is spent re-explaining instead of building?
- A) < 10%
- B) 10–30%
- C) 30–50%
- D) > 50%
🤖 Autonomy Gone Rogue
You asked for one thing. The agent touched five files you didn't expect.
How often does your agent overstep?
- A) Rarely
- B) Sometimes
- C) Often
- D) Almost always
🪟 Session Overload
Multiple agents, multiple threads — and now you're the project manager, not the engineer.
Managing parallel agents feels like...
- A) Air traffic control
- B) Herding cats
- C) Smooth sailing
- D) I avoid it entirely
🎯 The Trust Gap
Great output, but you have no idea what it actually did under the hood.
How much do you trust agent decisions without reviewing them?
- A) Fully
- B) Mostly
- C) Barely
- D) Never
🚨 The Rules Rebellion
You set up .cursorrules. You wrote clear project instructions. The AI read them, acknowledged them — and then did whatever it wanted anyway.
How reliably does Cursor actually follow your custom rules?
- A) Always — it's rock solid
- B) Mostly — minor slips here and there
- C) Hit or miss — can't rely on it
- D) It ignores them almost entirely
What is it that breaks your flow?
Top comments (0)