07:42Leveling Up with Eric Siu · Talking Head
How Fable 5 Can Run Your Business For You
A 7-minute breakdown of five concrete revenue plays unlocked by the new Claude Opus 5 and Sonnet 5 models.
June 11thThe argument in one line.
Manually prompting AI agents one-by-one is already the old way of working; the practitioners pulling ahead are designing loops — autonomous programs that trigger, gather context, act, evaluate output, and iterate until a defined stop condition is met.
Read if. Skip if.
READ IF YOU ARE…
- You use AI coding or marketing tools and still manually re-prompt to move work forward.
- You run a small team with recurring workflows in sales, content, or recruiting that eat significant time each week.
- You want a plain business-owner mental model for what the AI agent hype actually means in practice.
- You are skeptical of the loop narrative but want to understand what engineers and founders mean when they say prompting is dead.
SKIP IF…
- You are an engineer already building agentic pipelines — this is the business-owner orientation, not a technical implementation guide.
- You need code examples or step-by-step tooling — the video points to Matt Van Horn's article for that depth.
The full version, fast.
The shift from prompting to loops means you stop being the person typing instructions and start being the author of a system that does the prompting for you. Every business loop has five parts: a trigger that signals work needs to happen, a signal/context pull that gathers the right data before acting, an action the agent executes (draft, score, route), an eval gate that checks whether the output is accurate and safe, and a stop condition that kills the loop when the work is done. Applied to sales this means auto-reviving stale deals; applied to content it means drafting, reviewing, publishing, and capturing engagement as a single continuous loop; applied to recruiting it means sourcing, scoring, and outreach without manual re-initiation. Without all five parts — particularly evals and stop conditions — loops break and work rots in open Slack threads.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:41
01 · Why Loops Are The Future
Hook with social proof from Boris Cherny and Peter Steinberger
00:41 – 02:00
02 · The Receipts
Peter Steinberger 7.6M-view tweet and Matt Van Horn 2.8M-view post cited as proof of massive interest
02:00 – 04:00
03 · What A Loop Actually Is
Boris Cherny plain definition via Matthew Berman; IDE deletion story; evolution from cron jobs and slash goal command
04:00 – 05:00
04 · Five Tips For Running Agents Autonomously
Boris tips: auto mode, dynamic workflows, slash loop, cloud execution, self-verify — tip five is the practitioner obsession
05:00 – 06:00
05 · The Loop Is The Expensive Part
Reality check on token costs; the loop is now the finance problem, not just a velocity win; sponsor read
06:00 – 07:30
06 · Business Examples: Stale Deals, Speed To Lead, Content
Three concrete loop walkthroughs in sales and content with summarize-draft-approve-ship-track pattern
07:30 – 09:03
07 · The Five Parts Of Every Business Loop
Trigger, Signal, Action, Eval Gate, Stop Condition defined and illustrated with lower-third graphics
09:03 – 10:30
08 · Broken Loops In Slack and Telegram
Open threads = triggers with no actions, no signal pull, no eval, no kill criteria — where work goes to die
10:30 – 10:57
09 · How To Compound With Loops
Loop audit framing and closing argument: compound interest for AI — prompts die, loops persist
Lines worth screenshotting.
- The creator of Claude Code says he no longer prompts — he builds loops, and the loops do the rest of the job.
- A loop is a small program that prompts the agent, reads the output, decides if it is done, and if not, prompts again — you become the author of the loop, not the typist inside it.
- Every business loop has exactly five parts: trigger, signal, action, eval gate, and stop condition.
- Broken loops are already everywhere — open Slack threads with no action, no signal pull, no eval, no kill criteria are where work goes to die.
- The loop is now the expensive part: every AI agent is a for-loop, an LLM call, and a try-catch; the only agentic thing is the token bill at the end of the month.
- An eval gate is just a definition of what success looks like — without one, you are guessing whether the output is good enough to send.
- Every loop needs a stop condition; without kill criteria, stale items rot in the queue forever.
- Speed-to-lead is a natural loop: lead comes in, score it, assign an owner, draft outreach, track response time — no human trigger needed for each step.
- One-off prompts go to the graveyard; loops persist and compound with each iteration when feedback and evals are wired correctly.
- Building tighter loops around revenue, content, recruiting, and ops right now puts a business ahead of over 99% of competitors who have not started thinking about this.
Five parts that make an AI workflow actually autonomous.
WHAT TO LEARN
A loop without an eval gate is guessing; a loop without a stop condition runs up bills forever — most broken AI workflows are missing one or both.
- Prompting an AI agent manually every time you want it to advance is the same as being the human inside the loop — the whole point of a loop is to remove that dependency.
- Every autonomous business workflow needs a trigger (what starts it), a signal pull (what data it gathers first), an action (what the agent does), an eval gate (how it decides if the output is good), and a stop condition (when it ends).
- Open Slack or Telegram threads with AI agents are broken loops by default: triggers exist, but there is no signal pull, no eval, and no kill criteria — work just sits there.
- An eval gate is not a technical concept — it is just a written definition of what a good output looks like so the agent or a human reviewer can check against it.
- Loops compound: if each iteration feeds its outcome back as signal for the next run, the system gets better over time; one-off prompts cannot do this because they die at execution.
- Before building a new loop, run an audit: where does work currently sit idle, what is the most expensive manual trigger, and is there a human decision that could be codified as an eval rule?
Things they pointed at.
00:00channelBoris Cherny (Claude Code creator)
00:43linkPeter Steinberger tweet (7.6M views)
00:58linkMatt Van Horn article (2.8M views)
10:00linkMatt Van Horn livestream on loops
Lines you could clip.
01:58
“You stop being the thing inside the loop typing prompts. You become the author of the loop. The model becomes a subroutine.”
04:19
“Every AI agent I shipped this year is a for loop, an LLM call, and a try catch around the JSON parsing. The only thing agentic about it is the entropic bill at the end of the month.”
09:55
“One-off prompts end up going to die, go to the graveyard, and loops will persist.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogy
00:00Everybody is talking about how loops are the next big thing when it comes to AI. For example, you have the founder of Claude Coats saying it. The thing that I've been finding myself using more and more is loop.
00:10Just like the coolest thing. It's like the simplest thing that works. And you also have the founder of open clause saying it.
00:14So let's go into what that is exactly, and more specifically, let's go into how this will make more money for you, how this will make more money for your business. So let's take a look. First and foremost, the receipts are here.
00:24If you look at Boris here, he's the founder or creator of Claude Code, and he has said, don't prompt Claude anymore. What I mostly use now is loops. There's that word loops.
00:34I create loops, they do the rest of the job. In twenty four minutes, Boris reveals his real daily Claude Code setup, and then Claude plus loops plus routines plus dynamic workflows. So that's one proof point.
00:45Okay? And then you have Peter Steinberger who said this recently. By the way, this has 7,600,000 views on it.
00:49Okay? So this exploded. And here's your monthly reminder that you shouldn't be prompting coding agents anymore.
00:54You should be designing loops that prompt your agent. And so what the heck does this mean exactly? So what I did is so I I I have this this guy that I know, Matt Van Horn.
01:02He wrote a post that got 2,800,000 views. And so between these, we're talking millions and millions of views. Right?
01:08And Matt here, he I'm actually By the way, if you check my channel on on YouTube, I'm actually doing a livestream with him, so you should check that out where we actually go into a little more detail. But I'm talking about this in the context of making more money, and and driving more revenue for your business, because that's ultimately what I care about, and that's that's what my company does, so that's what we're talking about.
01:26Right? So I think he wrote some really good things here around what is a loop actually. And so it it's it's pretty detailed, but I think the thing to call out here is that when you when you think about a loop, it's there's a lot of different people responding as to what that is.
01:40Right? Is it a Ralph loop? Is it just like a slash goal command?
01:42What is it exactly? And so what Matt did here was that he ran the last thirty days. That's a skill that you can run to see what everyone was fighting about, because nobody has a clear definition right now.
01:53And I think he he did a really good job with this article here. So here's the monthly reminders that this one got the the 7,600,000 views, and then nobody knows but him and Boris.
02:00Right? Somebody Matthew Berman, who's another creator, said this. What a loop actually is okay.
02:05If you look at the plain version, a loop is a small program that you write that prompts the coding agent for you. It reads what it produced, decides whether it is done, and if not, prompts it again. You stop being the thing inside the loop typing prompts.
02:17You become the author of the loop. The the . He says in the last thirty days, this is December twenty twenty five, one hundred percent of my contributions to Cloud Code were in my Cloud Code.
02:27I landed 259 PR. Then he said he deleted his IDE in November.
02:31It hasn't since he hasn't opened it since. So I think the thing to call out here is, like, people are just like, oh, know, it's it's just a rebranded version of a of a cron job, right, as to what what this loop actually means. And so what Matt's saying here is that a loop is actually an evolution of a cron jobs.
02:46People are saying, oh, well, isn't this just an evolution of the the Ralph loop or the slash goal command? And it is. And and so maybe this is perhaps the step before recursive self improvement where the the the agent improves itself over time.
02:59And so the way I think about this is a loop is I like, for example, I run my Hermes agent and I have a bunch of different Slack threads open with it, and oftentimes I have to come back and I have to prompt it. Right? Or with my codec, for example, my clog code, I have to prompt it.
03:12I have to I have to be the human in the loop that's constantly reviewing all the time. But the idea with the loop is that it's just gonna keep working on it over and over. And, you know, people say that slash goal can do it, like slash goal to an extent can do it now now that it's been updated.
03:25The Ralph loop back then, a couple months ago, couldn't exactly do that. Like, it could work for a little longer, but I at least for me, I didn't see it working through the night. Right?
03:33And so slash goal command can actually do that, and slash loop is kind of the the evolution of that. Right? And so Boris posted five tips for running Opus autonomously for hours or days, and so that's what we we all want.
03:43Right? We want this to continue to to move for us. And so five tips, in his words, use auto mode for permission so Claude doesn't ask for approval.
03:50This is if you're using Claude. Use dynamic workflows to have Claude orchestrate hundreds or thousands of agents to get a task done. I don't think you usually need to to to use dynamic workflows.
03:57Also, it's very token intensive as well. And then use slash goal or loop slash loop to nudge Claude to keep going until it's done. Use Claude code in the cloud so you can close your laptop and make sure Claude has a way to self verify its work end to end.
04:09Okay? So that's what we're looking for. Right?
04:11Tip five is the one that hype skips and the practitioners obsess over. So we you want a loop that's that's trustworthy and it has the ability to check its own work. And so if we move on here, the loop is now the expensive part.
04:23So here's where the research turned from velocity to a finance problem. So the sharpest deflation of the whole agent's mythology came from a working engineer. So every AI agent I shipped this year is a for loop, an LLM call, and a try catch around the JSON par parsing.
04:38. So this again, this is very token intensive. Right?
04:43And so when I start to think about this from a from a business standpoint, I'm gonna give you a couple examples here. Okay? By the way, if you wanna grow faster, you need to have a single brain unified intelligence sitting inside of your chat.
04:55So it could be inside of Slack or Teams, but you can see right here the spot's working and it's creating ad creatives. It is doing data pulls from Meta, from Google. You can pull your SEO data.
05:03So imagine having all these data connectors. You you can ask whatever question you want, get the data pulls a lot faster. Your team can see it as well.
05:09They can choose to execute, and then you can run your other specialist agents that you have inside. So we have ad creative agents. We have email infrastructure agents.
05:16There's all these agents that can do a bunch of things, and the whole idea is they're all playing together. They're playing with your team as well. That way you're gonna be able to just move a lot faster and then grow a lot faster.
05:24So check it out. Just go to singlebrainwithab.com. Singlebrain.com.
05:28We'll see you inside. So a couple of of of examples when it comes to to business here is you think about let's say you have a stale deal, for example.
05:36Okay? So you you might have a stale deal, then you could summarize the context. Right?
05:40This is talking about in the context of sales. So you have a stale deal. Okay?
05:43It's not going anywhere right now. It's a it's a lead that you might have talked to a while back. And now you're gonna what you're gonna do is you're gonna summarize the context of the deal, draft the follow-up, approve it, ship it, and then you're gonna track replies, and then you're gonna you're gonna keep looping this as needed until you maybe clear up all your still deals.
05:58Right? Or even talk about reviving deals as well for deals that you might have lost a while back, maybe sixty days out to you you can go out even like three, four years or so, and just try to keep reviving these deals. Right?
06:08Because because these are just things that you need to repeat over time. And then speed to lead. So let's say a lead comes in, so then you basically need to score that lead very quickly to see if it's a fit.
06:16You wanna assign an owner, you wanna draft outreach, and you wanna track time to first response. Okay? So that is something that you could loop as well.
06:23Content, for example, think about all the topics that you can cover. Maybe use that last thirty day scale where you can see the last thirty days, what has been talked about for that particular topic.
06:32Um, and then you can draft for different channels. You can peer review, you can publish, you can capture engagement, and then feedback for the next cycle.
06:39So you can continue these loops. Alright? So by the way, when it comes to business standpoint, there are there are five parts over here.
06:44Right? So number one is you have a trigger. So let's say something signals that work needs to happen.
06:50Okay? Number two is you have signal and context. So pull the right info before acting.
06:53Right? I just gave you a couple examples over there. Number three is an action that you're gonna take.
06:57So an agent does something. So for example, you might wanna have it draft an email, for example, or you might wanna draft content for you, and then maybe it's scoring something or it's routing something. Okay?
07:05Number four, you have an eval gate. So you wanna make sure that an eval is just a definition of what success looks like, and so so then it helps your your your agent decide if something's actually good or not good. And so a human or system can ask if if something is accurate, if it's safe, if if it's worth sending.
07:19So evals are very important. Part five, you want a stop condition. So if it's shipped, approved, killed after seven days, like, you you you want to decide what that what that actually looks like, what that stop condition looks like.
07:29And so those are the five parts of of every business loop ultimately. Right? Again, you can apply it to content, you can apply it to sales, you can apply it to you can apply it to paid media, you can apply it all these different you can apply it to recruiting as well.
07:41I I just gave you an example earlier. Like, I work with my Hermes inside of Slack. The problem with working inside of Slack or even Telegram, for example, your autonomous agents, is that those are broken loops.
07:50Right? So when I look inside of Slack, it's like, oh, you have all these open threads that are basically triggers that have no actions. Right?
07:56You you're not pulling signals. Okay? So you're basically acting blind, and you have to keep kicking it every single day.
08:00You have no eval gate as well, so you're guessing if output is good. And and you're constantly doing different things inside of Slack as well, and there's no kill criteria. Right?
08:08So there's the steel items basically rot forever. And so that's where things go to die. That's why it's important for you to start thinking about how can you set up these these loops.
08:16And you don't have to be an engineer. You have to think about how this is applied to business. Right?
08:19And I just gave you some examples there. So I'll give you some more examples. How do you apply it to recruiting and operations?
08:23And so, um, let's say you open a role, and then you need a source. Then you need a score to fit. Then a human reviews, and then you then you reach out after, and then then you track.
08:31You track the response rate, you iterate on the messaging based on how the the numbers are looking. So those are examples of of how this all applies for business, and ultimately how you can make more money with this. And so I would encourage you to think about when you're thinking about a a a loop audit, for example, you know, think about what are the most important workflows that are taking a lot of time right now or costing you a lot of money.
08:51Right? And also, like, where does work sit idle? So those are the broken loops that you wanna look at.
08:55Right? So if something's sitting sitting idle for a very long time like, for example, for me in Slack, I have so many open Hermes threads, and I and I've kind of forgotten a lot about them by by now, and so those are broken loops.
09:05That's why I've created like a Hermes execution OS that can keep the loops moving. Right? And we're still testing that right now.
09:11What data do you pull before acting? Right? So that that these are these are signals ultimately.
09:15So and and you want the loops to pull from these signals. And then, you know, who decides if the action is good enough? Is it you?
09:20Is it agent that decides it? Like, you have a really good eval for it? When does it stop?
09:23So every loop needs a kill criteria. Right? And then does the next run get better because of the last one?
09:29So that's the compounding ultimate. And a lot of people are talking about this right now, but I don't haven't seen a lot of situations where people have truly cracked it yet, at least from from my eye. So I would just go to say that if people are if you're having the creator of Clawd Code and also Open Claw talk about this right now in terms of this being the future and how basically prompting is is is, you know, having to prompt these things and sit there all the time is kinda the old way of doing the work, you should probably pay attention.
09:53The hope here is that, you know, we're no longer just one off prompting all the time. It's that we're we're compounding with these systems. You know, , and that loops will persist.
10:03Right? Especially if you have good feedback that you're continuing to give it, evals, and you need good stock conditions that that will make this these loops durable. I think ultimately, if you're gonna get on this right now, if you're having tighter loops around revenue, content recruiting ops, you're gonna be ahead of, like, 99% of businesses.
10:17Even 99.99% of businesses because they're not thinking about this stuff. And, you know, I if you think about compound interest being the eighth wonder of the world, I just look at all this as compounding.
10:27Right? And the sooner you can get not just yourself, but your business on it, and and also your team on this thinking like this, you're gonna be ahead of everyone else. Right?
10:34So I would just say this. Like, it it it all comes down to to loops now. If you're serious about business, you should learn this stuff, and, uh, you should definitely go check out that livestream that I did with Matt Van Horn where we talk about, you know, from his perspective, from a more engineering background, how he thinks about loops.
10:50So we'll see you in that video.
10:56Hi.
The Hook
The bait, then the rug-pull.
A tweet from the creator of Claude Code — just four words: 'I create loops' — has quietly made manual AI prompting look like the old way of doing business. Eric Siu built an 11-minute explainer around that shift: not the engineering mechanics, but the five-part structure every business owner needs to wire their workflows so agents keep running without a human in the loop.
Named ideas worth stealing.
07:30list
The Five-Part Business Loop
- Trigger
- Signal / Context
- Action
- Eval Gate
- Stop Condition
Every autonomous business workflow needs all five parts or it breaks — triggers with no evals produce unreviewed output, loops with no stop condition run up token bills forever.
Steal forDesigning any recurring AI workflow, auditing existing broken workflows in Slack/Telegram
08:45model
Loop Audit
- What workflows take the most time or cost the most money?
- Where does work sit idle?
- What data do you pull before acting?
- Who decides if the action is good enough?
- When does it stop? Does the next run improve on the last?
A diagnostic to find broken loops in an existing business before building new ones.
Steal forOperations review, AI strategy planning
How they asked for the click.
VERBAL ASK
10:30next-video
“Go check out that livestream that I did with Matt Van Horn where we talk about from his perspective, from a more engineering background, how he thinks about loops.”
Soft close — no card, no subscribe push, just a contextual next-step pointing to a companion livestream. Clean but easy to miss.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
