the first prompt every OpenAI dot needs

from the episode: OpenAI's $500 plan feels like the old $200 plan - but dots are free | LAB 0011

Lenny's AI Builders - LAB 0011

before your dot does any work, make it restate your goals and audit every job you already run.

before it does any work, give it this one prompt. so it has to prove it understands you and stops what's wasting your usage.

initial impressions are pretty good. i just hope it stays free.

why it matters

a dot is OpenAI's new always-on agent. it runs on GPT-6 Astra, gets its own computer in the cloud and its own browser, and connects to over 4,000 apps.

right now dots come with Pro plans. talking to your dot doesn't use your limits. but Codex or work jobs it starts can.

and the $200 plan goes down to 10x Plus usage once the old limits end on 29 October. it was 20x.

so if your dot stacks new jobs on top of old ones, it eats the usage you have left. check what's already running first.

what you get

you see if your dot really gets your goals before it touches anything.
you find the jobs that just burn usage, each marked keep, kill or trigger.
you start with one job and a weekly budget, not a pile of new schedules.

the real example: my dot, Clarity

i set up my dot live on my Pro plan, gave it access to everything and called it Clarity. then i pasted the prompt below.

its first reply took about three minutes. more than 20 minutes after i sent the prompt, it had:

restated my goals: "you want Threadify, your content, and your AI expertise to produce dependable income without you coordinating every agent and handoff."
given its diagnosis: "you have overlapping systems with unclear ownership and weak proof of completion."
found 84 Codex entries, 25 labelled active, and 6 of those with no future run. all 9 old wiki jobs were paused.
picked one job: "one follow through for one low risk PR using the existing process rather than adding another monitor."

it's waiting for my approval. nothing was approved on camera.

LAB 0011 frame at 12:56: "audit before you automate" - what my dot found on day one, sorted into keep, kill and trigger — source: LAB 0011 episode · 12:56 - the keep, kill and trigger card

try it

set up your dot on desktop.
connect only the tools you really use.
paste the prompt below before you give it any job.
check if it got your goals right.
kill the schedules it flags, once you agree with each one.
approve one job, with a weekly budget.

copy this prompt

this is the prompt i read out at 10:40. paste it into your dot before it does any work.

Lenny’s AI Builders / LAB 0011

Before you do anything, read everything I've connected you to: my repos, Slack, Notion, calendar, and every scheduled task or agent I already run.

1. Restate in your own words what you think my goals are and the problem I'm trying to solve.
2. List every recurring job and scheduled agent I run. Mark each: keep, kill, or replace with a trigger.
3. Pick the ONE job you'd take off my plate this week. Give me the outcome, the tools you need, what you must ask me first, and a weekly budget.

Don't start work or create any schedule until I approve.

LAB illustration: "audit first" - a dot sorting jobs into ticks and crosses — source: LAB illustration made for these notes with ImageGen · not a screenshot from the episode

you'll know it worked when

your dot writes back your goals in words you'd agree with.

you get a list of every recurring job and scheduled agent, each marked keep, kill or replace with a trigger.

it names one job with the outcome, the tools, what it must ask you first, and a weekly budget. and it hasn't started work or made a schedule.

a bad result: it starts building before you approve, it gets your goals wrong, or it misses jobs you know you run.

limits and dates

recorded on 30 September 2026, Perth time. in the US, that was still DevDay, 29 September. the dot test ran live during filming, on day one of dots.

LAB 0011 went public on YouTube on 1 October 2026 at 12:48 am Perth time.

this is one day-one run, on my own setup. the audit is my dot's opinion, not a measurement. your dot will say different things.

nothing was approved on camera. this shows the audit, not the job getting done.

dots come with Pro plans for now, so you're paying at least $100 a month. they're not in the EU or the UK yet.

the prompt is built from what builders shared on X: poteto's restate prompt, plus people's Grok Bot and dots setups.

the news in this LAB is as it stood when i recorded. check the linked sources for where things are now.

the rest of this LAB

the $500 plan and the $200 cut. the new $500 Pro plan gets 25x Plus usage, plus Ultrafast (up to 8x faster). the $200 plan goes down to 10x Plus. it was 20x. existing Pro users keep their current limits through 29 October 2026, plus a one-time grant of 62,500 usage credits ($2,500 value). "feels like the old $200 plan" is my reading, not OpenAI's words. my take: if you're on the old $200 plan, don't upgrade yet.
GPT-6.1 Sol. $2/$10 per million tokens, a fifth of GPT-6 Astra's $10/$50. on Artificial Analysis: Sol 52, Opus 5.5 58, Sonnet 5.5 56. my take: cheap, but not the best. it's my new default in Codex.
dots. always-on GPT-6 Astra agents with their own cloud computer and browser. free with Pro plans for now, with an asterisk: Codex or work jobs your dot starts can use your limits.
Codex cloud. close your laptop and your agents keep working, on your normal plan usage. not in my app yet, so i haven't tested it.
Sign in with ChatGPT, plus two new APIs. your ChatGPT plan pays for the AI in over 16 partner products like Devin, OpenCode and Notion, per Tibo, and you can cap each app. plus the Agents API and the Decisions API (about 150 milliseconds, limited preview).
ChatGPT Space and pages. a shared home for you, your team and AI, with live pages your dot can keep updated. my take: ChatGPT wants to be Notion.

sources

next step

watch LAB 0011 on YouTube - the prompt starts at 10:27, and my dot's answer at 12:07.

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