Lennox Saint
Open menu
LAB 0016

on YouTube 8 oct 2026

field notes · new episode

make Claude use every drop of your plan

from the episode: OpenAI's rival called their 28 days "nothingburgers" | LAB 0016

Lenny's AI Builders - LAB 0016

one prompt makes Claude check your usage every hour and add sub-agents when you have usage to spare.

i had about 30% of my weekly Claude usage left, with two days to go.

so remember this: use every drop you buy.

why it matters

your plan resets every week. usage you don't use is gone.

most people only worry about running out. the other problem is wasting what you paid for.

you can't watch the usage page all day. so get Claude to watch it for you.

what you get

one prompt you paste at the start of a big Claude Code session.

  • an hourly check. Claude reads your 5-hour and weekly bars on claude.ai.
  • a pace. it works out how fast it can spend what's left.
  • more hands. it adds sub-agents when you have usage to spare.

the real example: my Eddy Studio session

i ran this on 6 October 2026 in my Eddy Studio Claude Code thread. that's where i'm building my agentic video editor.

  • i had about 30% of my weekly usage left. i told it to pace that over the next two days.
  • Opus 5.5 was the main agent. it had two Sonnet 5.5 sub-agents running.
  • once it worked out the pace, it added three more. that's five in total.

this is the message it gave me:

LAB 0016 frame at 13:51: Claude Code's message - burn since yesterday evening averages about 2.5% an hour, reaching 100% by 5 pm needs about 3.2% an hour, so it adds three lanes, five in total
source: LAB 0016 episode · 13:51 - Claude's own pacing message

my take: i couldn't have worked that out. i wouldn't have taken the time. Claude did it in a couple of seconds and course-corrected on its own.

try it

  1. start a big Claude Code session, or use one you're already in.
  2. make sure Claude can open claude.ai in your browser (for example with Claude in Chrome).
  3. paste the prompt below first.
  4. tell it how long you want your usage to last.
  5. let it run. check your 5-hour and weekly bars later.

copy this prompt

paste this first in Claude Code. keep claude.ai in it - Claude wasn't comfortable opening the desktop app.

Lenny’s AI Builders / LAB 0016
monitor claude.ai every hour to check and manage my usage limits across 5-hour and weekly limits.
parallelize work across subagents to ensure i'm using as much of my usage as possible.
LAB 0016 frame at 12:32: the "paste this first in Claude Code" prompt card with claude.ai and subagents highlighted in red
source: LAB 0016 episode · 12:32 - the prompt

you'll know it worked when

Claude checks your usage page on its own.

it tells you its pace, like an hourly burn rate, and adds or drops sub-agents to match.

by the end, little of your plan is left unused.

a bad result: it never opens claude.ai, or it adds sub-agents with no reason.

limits and dates

this is my own run on 6 October 2026, in one session. your numbers depend on your plan and your work.

more sub-agents use your plan faster. only do this with usage you would otherwise waste.

LAB 0016 went public on YouTube on 8 October 2026 at 5:49 pm Perth time.

the rest of this LAB

  • OpenAI's 28 days and the nothingburgers beef. OpenAI promised 28 days of gifts. on day two, poteto (Lauren Tan from the Grok Bot team) called them nothingburgers. 76% of almost 75,000 votes on Tibo's poll said day two needed a reset. Tibo says auto-review is now free and doesn't use your limits - that's his post, not OpenAI's docs. my take: build the model, not the calendar.
  • Claude now works inside Google Docs. Claude sits in a sidebar in Docs, Sheets and Slides and edits in place. it's a beta, on paid plans only. my take: watch where the tentacles go.
  • Hark Pro. the Figure founder's personal agent. it gets its own computer and checks with you first. there's a free plan plus $20 and $100 plans. i haven't tried it.
  • OpenAI's AI did new maths. a repo of 722 manuscripts in 372 families. many results are computer-checked, not all. the model isn't out.
  • treg vs Clay. treg finds leads and emails inside Claude Code or Codex, for under a cent per verified email. the benchmark is treg's own test.
  • Nano Banana 2.1. better visual design, mask edits and consistency. it's in the Gemini app and AI Studio. Nano Banana 2 shuts down in the API on 29 October.

sources

next step

watch LAB 0016 on YouTube - the practical starts at 12:01.

want the next one? get the next field notes. optional - these notes stay free to read.

want the next one?

free. unsubscribe any time, one click. privacy.