12:56Leveling Up with Eric Siu · Tutorial
6 INSANE Codex Use Cases for Business
A 13-minute live screen-share of six Codex agent workflows actively generating revenue at Single Grain — running autonomously for days at a time.
June 18thThe argument in one line.
Most creators use /goal at its default single-mission level, but upgrading it with custom presets like /goal overnight, /goal night-queue, and /goal batch—wired into Slack with approval gates—lets you run autonomous revenue loops while you sleep.
Read if. Skip if.
READ IF YOU ARE…
- A founder or operator running a business who's using /goal at default settings and wants to unlock parallel agent execution and overnight task completion without manual oversight.
- A technical operator managing Slack-resident AI agents (Hermes, Claude Code, or similar) who needs approval gates and risk-stratified autonomy to scale task delegation safely.
- An agency or product team running multiple concurrent experiments (landing pages, ad variations, design iterations) and looking to batch and queue /goal operations instead of executing them serially.
SKIP IF…
- You're not using Claude Code, Hermes, or similar Slack-integrated agent frameworks — this breakdown assumes that specific technical stack and won't translate to other LLM interfaces.
- You're still in the single-agent, single-task phase of AI adoption — the revenue unlock here depends on already running multiple parallel missions and needing orchestration, not foundational /goal usage.
The full version, fast.
The default /goal slash command in Claude Code and similar agent stacks is underused, and treating it as a single-shot task runner leaves most of its revenue leverage on the table. The upgrade is to wrap /goal in custom presets � overnight for unattended long builds with a morning executive packet, night-queue for ranking and launching the safest high-leverage work autonomously, batch for splitting open Slack threads into parallel lanes, plus continue, pause, and status for state control � all routed through a Slack-resident control tower. The non-negotiable is a definition of done on every mission and approval gates around anything that touches money, customers, or production, so agents move fast internally while risky actions stay packaged for your review.
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 – 01:00
01 · Cold open + revenue framing
/goal has shipped in Claude Code, Codex, Hermes, Open Claw — Eric reframes it as a revenue tool, not just a dev convenience, and previews that most people are using it at the base level.
01:00 – 02:50
02 · Live example — /goal'ing open Slack threads
Walks through using /goal inside Slack with his Hermes agent (Codex as the frontier model). Asks 'what are the open threads, can we /goal them?' to drain a queue of half-spec'd work without re-checking each one.
02:50 – 04:20
03 · Defining 'done' for a /goal
Core principle: every /goal needs an explicit definition-of-done outcome. Examples given: 'design 50 ads in this style, this copy' or 'three landing page variants pulled from Mobbin MCP'.
04:20 – 06:10
04 · Upgrading /goal — the custom preset registry
Shows his Hermes /goal registry: /goal overnight, /goal status, /goal pause, /goal resume, /goal remove, /goal business-operator, /goal big-op, /goal batch open-threads, /goal continue, /goal open-thread scope. Calls out the historical Ralph (Wiggum) loop as the predecessor, now baked in.
06:10 – 07:40
05 · Batch and continue — multiple goals at once
Default is one goal at a time, but /goal batch lets you fan out across Slack threads, and /goal continue lets a paused goal resume from a clarifying question without manual reprompting. Argues the Kanban board most teams use is too messy.
07:40 – 08:40
06 · Single Brain sponsor break
Mid-roll for singlebrainwithab.com / singlebrain.com — unified intelligence layer in Slack/Teams, ad-creative agents, data pulls from Meta/Google/SEO, team-visible execution.
08:40 – 10:00
07 · Diagram 02 — /goal overnight + 04 — AI Optimization Lab
Walks through two operating diagrams: 'Agents work the night shift with proof and approval gates' (Eric → Sleep → Morning approval; Tracks: Builder/Reviewer/Operator/Escalator; Proof: deliverable monitors / risk filter / executive packet), and the AI Optimization Lab compounding revenue loop (metrics → diagnose → hypotheses → eval scoring → approval → test → readback).
10:00 – 11:00
08 · Diagram 03 — Night Queue + 07 — Hermes Control Tower
Night Queue reads the business surface, dedupes work, ranks by impact, auto-launches safe jobs, packages risky ones for approval. Hermes Control Tower model: goals launch runners → runners create artifacts and evals → approvals touch outside world → metrics+memory improve the next goal. Mantra: 'move the ball forward'.
11:00 – 11:19
09 · Diagram 06 — Safe command surface + outro
Safe vs risky autonomy: internal docs / dry runs / evals & QA are safe to auto-run; email, CRM, social, CMS, ad-spend, deploys require Gate 4 human approval. Mantra: 'autonomy without letting the robot touch the money printer'. Outro pitches the next video on Hermes revenue growth.
Lines worth screenshotting.
- Most people are using /goal at its default single-mission level — the real revenue unlock is custom presets that let it work overnight, run in queues, and process batches.
- Defining the definition of done before running /goal is the difference between an agent that finishes a task and one that spins indefinitely on ambiguous completion criteria.
- A /goal overnight preset lets your AI agent execute one high-priority mission uninterrupted while you sleep — without any manual check-ins.
- Wiring a Hermes control tower into Slack creates an approval gate where risky actions pause for human sign-off while safe actions execute autonomously.
- The /goal batch command runs multiple independent missions in parallel — compressing what would take days of sequential work into a single overnight session.
- A /goal night-queue preset staggers multiple tasks across off-hours in sequence, so the agent is productive during the hours you're not watching.
- The more tools and MCPs you connect to your agent, the more interesting experiments you can run — the bottleneck is tool connectivity, not model capability.
- Running Hermes inside Slack rather than a terminal interface makes it accessible from your phone, enabling on-the-go oversight of autonomous overnight work.
Steal the format.
Slash-command preset playbook for JoeFlow / Mod Producer
Don't ship a slash command. Ship a registry of branded presets that each carry their own safety rails and output format — every preset becomes both a feature and a pricing anchor.
- Pick one base verb (/flow, /goal, /run) and wrap it in 6-10 named presets, each with a clear definition-of-done — overnight, batch, continue, status, pause, resume.
- Frame the upgrade narrative the same way Eric did: 'You're using X wrong, here's how the pros stack it.' The promise sells the upgrade.
- Bake a two-lane safety model in from day one — green path runs without confirmation, red path produces an approval packet a human signs off in Slack/email.
- Tutorials that show a real Slack thread / real Hermes registry crush abstract slide decks. Eric switches between his actual Slack window and his actual preset definitions — that's the proof.
- Slot the sponsor / upsell at the 65-70% mark (~7:40 of 11:19) right before the highest-density payoff section, not at the end where viewers have already bounced.
- Close every framework with a one-line memorable mantra ('move the ball forward', 'don't let the robot touch the money printer'). Mantras are the only thing viewers can quote in the comments.
- Visual recipe: bookshelf talking head for emotional beats, full-screen Slack for proof, orange-on-black operating-diagram slides with PiP host in the corner for the framework block. Three modes, no slop.
Terms worth knowing.
- /goal command
- A slash command in Claude Code and similar AI coding environments that lets a user specify a high-level objective and have the agent autonomously plan and execute a multi-step path to achieve it.
- Hermes (agent)
- An open-source AI agent framework that can be configured to run autonomously on a schedule, receive instructions via messaging platforms like Slack, and orchestrate other AI agents or tools.
- Codex (agent)
- OpenAI's autonomous coding agent capable of independently writing, testing, and refining code toward a stated goal without step-by-step human guidance.
- OpenClaw
- A community-built AI coding agent environment that supports slash commands and autonomous task execution, compatible with multiple language model backends.
- Night queue
- A scheduled batch of AI agent tasks configured to run overnight while the operator is offline, maximizing compute time without requiring human oversight during execution.
- Approval gate
- A checkpoint in an autonomous AI workflow that requires a human to review and confirm before the agent proceeds with a high-risk or irreversible action.
- Slash command
- A shortcut typed as /<keyword> in a chat or terminal interface that triggers a specific preconfigured action or workflow, such as starting a task or switching an agent's mode.
Things they pointed at.
01:18toolCodex
01:18toolHermes (agent)
01:18toolOpen Claw
03:35toolMobbin MCP
05:25conceptRalph Wiggum loop (predecessor pattern)
Lines you could clip.
03:16
“When you're running slash goal, you need to make sure that you're defining an outcome.”
10:15
“Move the ball forward — that's ultimately gonna build the most leverage for you.”
10:45
“Autonomy without letting the robot touch the money printer.”
02:38
“I could've been continued to work in those eight hours where you're sleeping.”
09:27
“It's no different than when you hire someone that's really amazing that's able to kinda figure things out on their own.”
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.
metaphoranalogystory
00:00So you've probably heard about how slash goal has dropped across Claude code and has dropped with codex, with Hermes, with open claw, you name it, and it's going to change everything in terms of growing your revenues or making more money. So especially good if you're running a business right now because literally you can build anything that you want, whatever you want.
00:19And I'm gonna show you how you could actually go about upgrading this because people are using kind of the slash goal command at a base level. But I wanna talk about using it from a business standpoint, from a revenue generating standpoint so you can actually more money. Let's get into it.
00:30So first and foremost, I wanna show you an example. This is me working with my Hermes agent inside of Slack, and I'm using Codex as the main agent here or more so the the main LLM. So that's the main frontier model that I'm using here.
00:42And within this, you can see that I'm actually having to execute a full a full list of different things that I've already been working on. Okay? So I just ask, hey, what are the open threads that we have right now?
00:51Can we slash goal them? Okay? So I'm just I'm lazy.
00:53I wanna finish these things off, and I've already specked them out pretty well at at the beginning of it, I wanted to continue. Okay? So the whole idea with slash goal is that you're able to not have to check up as much as before whenever you are building something, and you can even have it work through the night.
01:08And I'm gonna show you how you can do that in a second because that doesn't come in necessarily built in, But this does make it easier to run multiple parallel agents and get things done faster. K?
01:18So if you look at this, slash goal overnight finished the execution panel. So this is one of the things that we're working on. And build, verify the monitor packet generator, run, dry, run verification gets opened item.
01:27So this is basically kinda the the definition of done. Right? Because when you're running slash goal, you need to make sure that you're defining an outcome.
01:32And ideally, it's something that's a little clear like, hey, I want you to design 50 ads for me in using this initial design over here, and I want you to use this text, or maybe I want you to do three landing pages, three landing page variations that I wanna I want you to pull from this mobbin m c p, which actually has a bunch of designs and and screens that you can pull from.
01:52So really a bunch of ideas as the the more you connect to your your your agent, like an Open Claw agent or Hermes, the more you're gonna find that you can run more interesting experiments with it. Okay? So the way slash go works is that you are basically able to define what you want and you should define specifically what the definition of done is, and it can get that thing done for you.
02:10Okay? Now slash goal doesn't come with all these upgrades that you might need in terms of hey, maybe you might want it to work overnight, overnight. Maybe you want it to optimize itself overnight, but I decided that I wanted to upgrade my slash goal, and I'm using my Hermes agent to do that.
02:23So you can see over here, these are new commands and presets that I've set up for for myself. And so the what I wanna call out here is slash goal overnight. So this allows you to run one high priority mission as an overnight operator loop with safe internal execution, verification, and a morning packet.
02:37I want the morning packet to tell me exactly what was done, and I want this to, you know, I want this to be safe, I don't want this to run any destructive commands. We want this to verify to make sure that the work was actually done. Right?
02:47And then you can define kinda what the mission is. What's the standing objective that Hermes keeps working on across turns until it's done, blocked, paused, or out of budget? And that way you don't need to keep coming back to it all the time.
02:56Right? And so now, you know, this this used to be called a Ralph loop because it was a Ralph Wiggum plugin before. Now this is built into almost all of these, you know, Frontier Frontier Labs and then which is built into these these these autonomous agents as well.
03:09K? You can also run for me, I can run slash goal status. So I wanna know how these goals are going right now.
03:14I can pause a goal as well without deleting it. I can restart a pause goal from its saved state, and I can remove the current goal so the session stops chasing it. So the other thing I'll say is typically you can only run one goal at a time, but I've actually added an upgrade as of now.
03:28I actually need to we're we're testing it right now, but the whole thing is that you can actually run batch goals. Right? So literally, I think the command is like slash goal batch, and then you can Because I have all these these open loops, which I'll show you in a second, there's all these things that you're working on all the time, especially if you're working in Hermes sorry, Hermes is where your your Open Call is working within Slack or even Telegram, you better forget things.
03:48And sure, you can use the Kanban board that they have, but oftentimes, it's just like it it's it's The Kanban board itself that they have is is pretty messy too. Right?
03:56And so I'd rather just say ask, hey, what are the open slides right now? What do I need to review right now? What do we need to continue?
04:01And so I actually have a command in here that's a slash goal continue. So once something finishes off and it it's asking for more feedback from me, I might just type slow slash goal continue or ask me any clarifying questions. Right?
04:12And then it just continues there. You can see here, I also have this one over here. So slash goal open thread scope.
04:17So shortcut for open thread batch mode, then we can we can batch a bunch of these together. And you can see I have batch threads over here. Okay?
04:24So splits open Slack threads into separate work lanes, attracts each lane independently, and produces a consolidated executive summary. Now the cool thing is if you're running this, you can also have maybe you're collaborating with a partner on your on your team. You can you can work with them on this.
04:36Right? Maybe you don't wanna have too many people involved, too many cooks in the kitchen. And NightQ allows you to look across open work, ranks the safest high leverage task.
04:44So I like talking about highest leverage because, you know, little bit of output, maybe $1 of output yields you $10 or a $100 back. Runs what it can and turns risky work into approval packets. Right?
04:54So the whole idea here is that you have slash goal to keep work moving, but you want it to work overnight. Maybe you want it to focus on continue to focus on high leverage task through through the evening. And then maybe you have opened their into maybe you need to batch it out.
05:05Right? So this helps my workflow. What I would suggest for you is that whenever you're you're working with your agent, in this case I'm working with Hermes over here, I'll say, hey, based on how I work, what upgrade should we add add to slash goal?
05:18And you could even take a screenshot of what I have over here on the screen and just say, hey, like here are some ideas for you to start with. What else should I add? Okay?
05:25So I'm gonna give you an example here on how it looks on my screen in terms of the things that I'm asking it to to do, and this will hopefully help give you some ideas too. Okay? So in Slack, I'm just like, hey, like, you know, what are the open threads and you know, can we slash goal to finish them off?
05:40Uh, have we enabled the slash goal upgrades like continue? Okay. So it continues over here, and then it just talks about all the open threads that I have.
05:47And these are a lot of things that I'm working on. It could be from a product standpoint, could be from a customer development standpoint, weekly connector, just the collector in terms of like all the open threads that I have. Ambient single brain control plane, that's something that we're working on for our single brain ambient agents.
05:59And so a lot of stuff like fleet cleanup, gateway token health, like all these things over here. So again, you're going to forget because you're you're doing so much work, you're doing this so quickly, you have this infinite intelligence in your hands right now, and you wanna keep it moving, and you feel like, oh, when you wake up in the morning, man, I I could've been continued to work in those those eight hours where you're sleeping.
06:16Well, this is a way for you to continue to work, and this is how it's gonna continue to help you drive more revenue ultimately. Right? And so the way I look at this is I always have a bunch of ideas.
06:25Well, I just want these ideas moving forward, and then maybe I'm I'm probably wanna slash goal to to just audit what I'm working on too to make sure that's tying in with my goals, to make sure that I'm working on super high leverage things. So I'm gonna then show you a couple of a couple of diagrams that I put together for this, and let's take a look at it.
06:41By the way, if you wanna grow faster, you need to have a single brain unified intelligence sitting inside of your chat. So it could be inside of Slack or Teams, but you can see right here this bot's working and it's creating ad creatives. It is doing data pulls from Meta, from Google.
06:55You can pull your SEO data. So imagine having all these data connectors. You you can ask whatever question you want, get the data pulls a lot faster.
07:01Your team can see it as well. They can choose 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, there's all the agents that can do a bunch of things.
07:10And 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. So check it out.
07:18Just go to singlebrainwithab.com. Singlebrain.com. We'll see you inside.
07:21Alright. So I put together a few diagrams here for you to look at. So business growth workflows enabled by slash goal.
07:27So seven screenshot ready operating diagrams for showing how vague AI tasks become measurable growth loops with proof, gates, and executive review. So we have I think what I'm gonna bring your attention to is on the slash goal overnight. And so agents work the night shift with proof and approval gates.
07:41So Eric sets the mission before bed, tracks with the work, morning starts with an executive packet, not a prompt box. Right? Because I wanna know what what happened and what next steps are.
07:49So I'll set the mission over here, this is before I go to bed, then I go to sleep, morning approval. Tracks, we have a builder, you have a review, an operator, an escalator, and then you have proof deliverable, monitors, risk filter, and executive packet.
08:00That's basically how it works, and that's that's how I keep the ball moving. Because what you don't wanna do is like, oh, okay. It it keeps checking in with you, and maybe it needs more approval through the night.
08:09But this because the thing is I've learned with with Hermes when I when I type slash goal, it usually gets the job done in about, I don't know, call it ten minutes, fifteen minutes or so. But when I run slash goal overnight, it actually truly overnights it, and I want it to be Typically, it might be a longer a longer task.
08:23Right? It might be a longer build. And you might not need to use overnight that much, but it's still helpful to have.
08:29That's what I would say. Okay? The next one would be night queue, which is slash goal night queue.
08:33So execution control plane beyond manual prompt copy and paste. So night queue reads the business surface, dedubs work, ranks by impact, launches safe jobs, and packages risky work for approval. So I have a Kanban board with my Hermes, no open threads, recent priorities, done, running blocks.
08:48So this will run on autopilot basically. It'll it'll just figure out what it can actually do safely. It'll auto launch it, it'll queue it, and then safe internal launch, risky approval packet.
08:56Right? So you just gotta think through what workflows make sense for you and what you're trying to accomplish. And then we have this AI optimization lab one, which is growth becomes a compounding revenue loop.
09:04So the lab turns metrics into hypotheses, approved tests, read backs, and a learning store that improves the next experiment. So you have metric inputs from the different sources that you have. It diagnoses the different bottlenecks that you have, hypotheses and variants, and eval scoring, and then approval then test.
09:18So you can basically run this AI optimization lab over and over. So the whole idea here is that you want a system that recursively self improves, um, because if you have to keep checking in all the time, that that that makes it tougher.
09:30Right? And so it's no different than when you hire someone that's really amazing that's able to kinda figure things out on their own. This AI optimization lab piece is something that you can try.
09:38And it's not gonna be perfect the first time, but the whole idea is that you're constantly testing this stuff. Okay? So anyway, all that to say is you wanna make sure I mean, the way I look at this is my Hermes agent is a .
09:50Right? It's not just a pile of prompts. So goal launches runners, and then you have runners create artifacts and evals, approvals, such outside world, back stress and memory, improve the next goal.
09:58So we want My whole thing here is just three words, , and that's ultimately gonna build the most leverage for you. Right? So autonomy is a safe command service.
10:07So autonomy without letting the robot touch the money printer. So a clear command service lets agents move fast internally while gate four protects money, reputation, production systems. What are really important.
10:15Right? If you're talking about running a business or growing a business faster. So you have internal docs which are safe, dry runs, evals, and q and a.
10:22So it'll dry run. We'll it'll test things. We have evals and q and a that that we should continue, again, to define what a good setup looks like.
10:30Right? Or what what a good what the definition of done looks Then you have gate for approval. So anything that involves sending emails out or adjusting CRM or, you know, sending some stuff out to social or or adjusting your CMS, um, you're gonna need to make sure that this has approval ultimately.
10:44So there needs to be some approval gates, otherwise this thing's gonna yellow. Same thing with kind of ad spend and deploy. So these things all all here are safe to run.
10:51Gate for approval, you need to make sure that you are protecting yourself or, you know, CYA, cover your you know what. Alright. So that is how I use slash goal to grow my business faster and how you could do it as well.
11:02So you'll generate more revenue, you'll aka make more money. Um, so hope you enjoyed this one and you can check out this next video over here on how we are using Hermes to grow our revenue faster.
The Hook
The bait, then the rug-pull.
Eric Siu opens not by introducing /goal but by accusing you of running it at default. The promise is bigger than a tutorial — the title insists you are leaving money on the table, and the cold open immediately reframes /goal from a coder's convenience into a revenue lever you can stack overnight presets on top of.
Named ideas worth stealing.
04:40list
/goal preset registry
- /goal overnight — one high-priority mission as an overnight operator loop with safe internal execution, verification, and a morning packet
- /goal status — how active goals are progressing
- /goal pause — freeze active goal without deleting
- /goal resume — restart a paused goal from its saved state
- /goal remove — drop current goal so the session stops chasing it
- /goal continue — pick back up after a clarifying question
- /goal batch open-threads — split open Slack threads into separate work lanes, track independently, produce a consolidated executive summary
- /goal night-queue — read business surface, dedupe, rank by impact, auto-launch safe jobs, package risky work for approval
- /goal business-operator — operator framing for business tasks
A registry of custom /goal subcommands layered on top of the base slash command — each one a named preset that bakes in safety rails, verification, and output format.
Steal forJoeFlow / Mod Producer — bake the slash-command preset pattern into the morning batch launcher. /flow overnight, /flow batch, /flow continue, /flow status — every preset = pricing anchor and an excuse to ship a template.
10:45model
Safe vs Risky command surface (Gate 4)
- Safe to run: internal docs, briefs, research, plans, notes, summaries
- Safe to run: dry runs — preview commands, simulated sends, draft publishes
- Safe to run: evals and QA — tests, screenshots, scoring, linting, checklists
- Gate 4 approval required: email, CRM, social, CMS — anything that writes to customers or public channels
- Gate 4 approval required: ad spend and deploys — money movement and production changes
Two-lane autonomy model. One lane runs without permission; the other requires explicit human gate before it touches money, reputation, or production.
Steal forAny agent product Joe ships (Paperclip, JACE, REESE). Default-deny on anything that writes externally; default-allow on anything internal.
10:15model
AI Optimization Lab loop
- Metric inputs (Instantly, Gong, ads, SEO)
- Diagnose bottleneck
- Hypotheses & variants
- Eval scoring
- Approval-then-test
- Metric readback into learning store
Closed-loop experimentation framework — metrics in, hypothesis out, test, score, approve, ship, read back into a learning store that improves the next cycle.
Steal forFrame the MCN+ growth pipeline as a learning loop, not a list of features.
09:40concept
Hermes Control Tower
- Eric sets mission, constraints, approval posture, business priorities
- Hermes Control Tower routes work, gates risk, prepares review packets
- Runners: /goal runners, Codex, Cron, Subagents → produce artifacts and evals
- Approvals + metrics + learning store close the loop
Reframes the agent stack as a control tower, not a pile of prompts — single dispatch point, accountable runners, approval gates, observability.
Steal forJoeFlow Sessions panel positioning — Sessions is the control tower for fleet runs. Lead with the metaphor.
How they asked for the click.
VERBAL ASK
11:08next-video
“Hope you enjoyed this one and you can check out this next video over here on how we are using Hermes to grow our revenue faster.”
Soft end-card CTA — no subscribe ask, no email capture. The real conversion ask is the embedded Single Brain mid-roll at 7:40, which is harder to skip than an outro.
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.
