I basically never write my own /goal anymore. I ask Codex to write one for itself, and one for each agent it spawns. Like this 👇

dude... turned into a template hehe Build [THING] in [TECH/FRAMEWORK]. It should include [MAIN FEATURES], with [INTERACTION/ANIMATION/BEHAVIOR DETAILS]. Make it feel [MOOD/QUALITY], using [VISUAL DETAILS], [ENVIRONMENT DETAILS], and [EXTRA EFFECTS]. Output as [FORMAT/FILE TYPE]. For this task, write yourself a new goal and spawn agents in parallel — as many as needed to do it better and faster. Split the work into independent pieces, dispatch them concurrently, and synthesize the results as they return. Give each agent its own dedicated /goal.

Amazing

this is the way!

🫡🫡🫡

been doing the same for a few weeks now! openai actually wrote an article about how to use /goal effectively. i took it and turned it into a skill:

lol

Letting agents write their own goals only works if the goals survive past one session.

Smart! You are a source of infinite tips & tricks! Maximum respect 🫡

Thank you!

I like using plan to create the goal and then switch off plan and use goal to execute it

Hey Pietro! After you asked Fable to create the Prince, it was destined to become a meme Join our Fable Prince Community if you like, the Prince will live on forever on Solana, and you can claim Royalties off its trading fees!

People underestimate how good AI is at generating good high-stakes prompts

Smart. Let the agents code themselves.

wait until you're sending Codex to build goals in collaboration with other harness like CC, or Hermes. /plan for agentic input, /goal for agentic plans, /contract for harness-to-harness, long-term and dynamic goals.

こちらで日本語解説しています I covered this in Japanese here:

Same pattern works in Claude Code with CLAUDE.md. Let the agent define its own constraints before it starts. It knows the task better than you do at that point anyway.

I kinda do the same thing but itty bitty different, I start from a plan and then have the goals be part of the plan that way codex still can define it's own goals but it's spec'd out. Which then allows room for improvement by me and codex 🙂

prompt space is underexplored, instead of writing prompts/loops for agents you should write prompts/loops to create prompts for an agent

Same!

Meta prompting is the best way to do it

Cc @jgordijn

Nice

lol thought I was the only one doing this, its actually really good at coming up with very well thought out and clearly defined goals

letting the model draft targets saves time, but i still tweak them. without human-set constraints, spawned agents can easily drift into weird edge cases when chaining tasks. solid workflow, just needs a quick sanity check before running.

@steipete veeery interesting

I find it interesting you're using 5.5. on low to orchestrate and organize the work. I always thought the extra tokens were particularly valuable at this important stage. Can you share your reasoning behind this setup?

This is the way. Unless your goal is super long and detailed, then this is not the way

making robots do their own homework 🥴

What the fuck is next? manifesto.md?

Yep, i found it out today and this is game changer

@steipete Been rocking like this too! you intentionally got reasoning effort on low?

@Nicodemos_V finish the matching engine k thx

I should probably try this

why bother? just chat about what you need to build. After a while when everything is right for you just send /goal ok bro do it and go away from the computer

This is a lesson to always show what you do with AI. I started doing this six hours before Pietro posted the video lol

codex, copy better codex pls

but what if you dont have the meta-goal?

All you need is a request

my biggest problem is that I grasp what I want but dont know if I really want that