Claude Code: Painpoints and Stress Inducers

People who (understandably) haven’t read the stupidly long post on how I work with Claude most often ask “what makes it stressful”. These are the main things. Obviously, all developers do some of these things. But Claude does them more often, faster, and with supreme confidence. And whatever it does is on me.

Bullshit: Like all LLMs, Claude can, and does, bullshit constantly

Claude is disobedient

Claude (sometimes) writes code that is dangerous, inefficient, repetitive

and it writes a lot of it

I am constant watchingly token consumption to avoid compactification

I can’t trust anything Claude says

When not coding, Claude is like a caged animal, champing at the bit

While generally malleable, Claude can be quite (apparently) passive-aggressive

Claude has good session and bad sessions

Claude codes before understanding the problem

Claude goes off the rails

Claude deletes code with gay abandon

including uncommitted code

with rm and with git checkout

Claude lulls you into a false sense of security by going things well

until it starts doing them badly

Claude wants tests to pass in all circumstances; even when they should fail

and will suggest anything to make them pass

Claude rarely proactivity tells you when it is “worried”

Claude rarely admits when it doesn’t “understand” the task/approach

not even when it can see it’s wrong