OpenClaw project steward Peter Steinberger argues developers should stop manually prompting AI coding agents and build automated programmatic loops · Digg

OpenClaw project steward Peter Steinberger argues developers should stop manually prompting AI coding agents and build automated programmatic loops · Digg

Posts from X

Most Activity

VIEWS239.5KBOOKMARKS749

Chris Albon@chrisalbon

What does this loop like in practice? Like Claude -p in a for loop? /goal?

Peter Steinberger 🦞@steipete

Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore.

You should be designing loops that prompt your agents.

21d239.5K546749

LIKES4.5KRETWEETS235

rahul@0interestrates

if you’re still writing loops that prompt coding agents you’re falling behind. you need to build a meta agent that infers what loops you would have wanted based on your vibe and then write those loops

20d192.1K4.5K433

REPLIES333

Tibo@thsottiaux

Anyone writing nested loops yet?

Peter Steinberger 🦞@steipete

Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore.

You should be designing loops that prompt your agents.

20d212.5K1.2K262

Gergely Orosz@GergelyOrosz

On the whole “just use loops”

Outside of the increasingly few people who

1) have unlimited AI token budgets

2) feel like prompting agents are holding them back (usually thanks to no #1)

I don’t think many have a use case for them. I’m more than content prompting (esp w #1!)

21d91.3K1.1K200

George@odysseus0z

Folks it is just control loop

Think of Kubernetes.

You write desired state/spec, what to observe/status, the controller, and the kubelet.

You let this whole control loop sense and actuate to prompt/drive the agent.

Obviously the right idea as AI gets smarter

Peter Steinberger 🦞@steipete

Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore.

You should be designing loops that prompt your agents.

21d66.6K413300

Garry Tan@garrytan

Don’t build Foxconn factories that make agents do the same thing over and over

Agents by and large are smart, thoughtful, and not dangerous, so you should let them do more, not less

Peter Steinberger 🦞@steipete

Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore.

You should be designing loops that prompt your agents.

21d87.1K586232

Drew Breunig@dbreunig

Will just leave this here… https://dspy.ai/getting-started/program-dont-prompt/

Peter Steinberger 🦞@steipete

Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore.

You should be designing loops that prompt your agents.

21d37.8K176229

Graham Neubig@gneubig

Agreed. My basic agent loop is this:

1. Organize my to do list: pull in context from email, slack, github, and linear and add linear issues for any missing things 2. Walk through the to do list in order of priority, tell me the current status, ask what to do 3. I delegate a sub-agent (using the OpenHands cloud api typically) to work on that task 4. Once I have many tasks going at once, check on the ones that finished in order, and take any action

I tried a more automated version of this where I skipped step "2", but eventually it resulted in a lot of slop, so I think the human decision step is still pretty useful.

Peter Steinberger 🦞@steipete

Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore.

You should be designing loops that prompt your agents.

21d20.9K108123

Matthew Berman@MatthewBerman

@steipete Can you explain your workflow in detail? Would love a blog post about it

Peter Steinberger 🦞@steipete

Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore.

You should be designing loops that prompt your agents.

21d23.6K33755

Yuchen Jin@Yuchenj_UW

On the whole:

“You shouldn’t be prompting coding agents anymore. You should be designing loops that prompt your agents.”

Loops are the temporary workaround: today’s LLMs have poor judgment. They struggle to know when to keep going, when to stop, or when to call a tool. Loops force agents to work longer.

Loops are incredibly powerful for verifiable goals for now, as AutoResearch shows.

20d15.1K18437

Philipp Schmid@_philschmid

Subagentmaxxing or /goal + subagents (^2 depth).

You should naturally evolve towards this when you try to max your agents run for longer or solve more complex task. You replace your oversight with another agent, and then their oversight with another agent around.

Peter Steinberger 🦞@steipete

Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore.

You should be designing loops that prompt your agents.

21d7.2K8329

Rohan Paul@rohanpaul_ai

The prompt era is ending. That's too linear, too bottlenecked by humans.

We are entering the loop machine of AI agents.

The value is in moving judgment upstream, so the human designs the process and the model handles the recurring friction.

Peter Steinberger 🦞@steipete

Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore.

You should be designing loops that prompt your agents.

20d8.4K7723

Beff (e/acc)@beffjezos

@steipete what do you mean by that more specifically

Peter Steinberger 🦞@steipete

Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore.

You should be designing loops that prompt your agents.

21d13.1K11217

Peter Steinberger 🦞@steipete

@InderosD I have my claw supervising my codex’es.

21d9.1K6824

Gergely Orosz@GergelyOrosz

If you built something specific that can be shown with loops that is useful, feel free to share.

I see experimentation happening for those who don't have budgets (either working at eg an AI lab, or just being happy to burn hundreds on an idea that is throwaway output)

Gergely Orosz@GergelyOrosz

On the whole “just use loops”

Outside of the increasingly few people who

1) have unlimited AI token budgets

2) feel like prompting agents are holding them back (usually thanks to no #1)

I don’t think many have a use case for them. I’m more than content prompting (esp w #1!)

21d15.3K924

elvis@omarsar0

The point is that you should start implementing ways to encode instructions/prompts with clear goals inside automations.

Nothing new but newer LLMs are being trained to perform for longer duration uninterrupted. Loops are one way to take advantage of that.

Peter Steinberger 🦞@steipete

Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore.

You should be designing loops that prompt your agents.

21d6.8K288

Peter Steinberger 🦞@steipete

@gauthampai Don’t worry it’ll take 3 months until it’s there.

We’ll be talking about fleets that design your loops then.

21d7.1K766

Boaz Barak@boazbaraktcs

I still hand prompt codex like god intended.

Peter Steinberger 🦞@steipete

Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore.

You should be designing loops that prompt your agents.

21d6.9K824

Alex Volkov @ AI Engineer@altryne

2026 is the year of Proactive agents!

Peter Steinberger 🦞@steipete

Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore.

You should be designing loops that prompt your agents.

21d5.3K408

Peter Steinberger 🦞@steipete

@LexanderBrouwer Who still uses VS Code?

21d8.2K643