How I stopped hitting Claude's limits?

Saw a few posts this week about tracking Claude usage, and I get it, the limit anxiety is real. But the thing that actually fixed it for me wasn't watching the meter closer. It was changing how I work so I stop burning through it.

Four things that made the biggest difference:

1. Plan before it builds anything.

Most wasted tokens come from one long session piling up dead ends. I make it lay out the plan in a few bullets first, then approve it. A minute of planning saves a pile of failed retries.

2. Push heavy reading into a subagent.

Instead of dumping a huge file or a load of research into my main chat, I send it to a subagent. It reads the 50k tokens and hands me back a 2k summary. My main context stays clean and cheap.

3. Start fresh before the context fills up.

Quality drops as the window fills, and a bloated session costs more per message. Around 70-80% full I save a short checkpoint, clear, and pick up from there. Feels backwards but it's faster and cheaper.

4. Keep instructions in files, not in every prompt.

Anything reusable lives in a file the agent loads only when it needs it, instead of me re-explaining it every session.

Net effect, I almost never hit the wall now, and the meter went from something I anxiously refresh to a signal that a session is getting bloated.

What eats your limit the fastest? Curious if it's research, long builds, or something else.

4 comments

How I stopped hitting Claude's limits?