24:23Mark Kashef · Tutorial
Master All 5 Layers of Every Agentic OS
A 24-minute Earth-layers framework for building AI operating systems that don't decay.
June 30thThe argument in one line.
UltraCode is not a smarter model but a project manager that forks your task across up to 100 simultaneous agents, each verified by a separate review agent before results return to you.
Read if. Skip if.
READ IF YOU ARE…
- You use Claude Code regularly and want to understand when UltraCode is worth the cost.
- You saw the UltraCode option in the /effort menu and assumed it was just a step above Max.
- You are planning tasks that involve gathering many parallel threads of information simultaneously.
- You want to know the actual cost difference between effort levels before committing to a session.
SKIP IF…
- You are already familiar with agentic multi-worker architectures and just want the API docs.
- You need a deep technical breakdown of how dynamic workflows are implemented internally.
The full version, fast.
UltraCode bundles two distinct capabilities: xhigh thinking (one step below max effort per message) and dynamic workflows (Claude plans the task, forks it across up to 100 agents simultaneously, and routes output through a verification agent). The key distinction from Max is that Max is purely a thinking depth setting while UltraCode adds the parallel team on top of near-max thinking. The trade-off is direct: better output on complex parallel tasks, dramatically higher token cost.
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 – 00:33
01 · Cold open and agenda
One-worker vs. 100-worker analogy; cost tease planted; three-part agenda card shown.
00:34 – 01:05
02 · The /effort menu
Shows the /effort slider from low to ultracode; explains UltraCode sits beside Max, not above it.
01:06 – 02:10
03 · The two pieces
xhigh thinking defined as one step below max per message; dynamic workflows defined as parallel agent team with built-in verification.
02:11 – 02:23
04 · Enabling UltraCode
One click from the /effort menu activates it session-wide; no per-task invocation needed.
02:24 – 02:57
05 · UltraCode vs. Max
Max is a pure thinking-depth setting; UltraCode adds the agent team on top of near-max thinking. Distinct tools, not interchangeable.
02:57 – 03:58
06 · Live demo
Food truck business research prompt; 48 agents visible in the lightning bolt counter; output is a complete multi-section PDF starter guide.
03:58 – 04:32
07 · The cost reveal
UltraCode is the most expensive Claude Code mode by a large margin -- officially acknowledged in the docs. Bar chart drives the point home.
04:32 – 04:54
08 · Recap and CTA
Two-sentence recap of both UltraCode components; subscribe ask; link to prior video.
Lines worth screenshotting.
- UltraCode is not an effort level -- it lives beside Max on the /effort menu because it is a fundamentally different execution model, not just a higher thinking setting.
- Dynamic workflows split a single task into parallel workstreams and automatically route each result through a verification agent before you see it.
- Max gives you the deepest single chain of thought; UltraCode gives you near-max thinking across a whole team of agents running at the same time.
- The agent count is visible in real time as a lightning bolt counter -- 48 agents fired in the food-truck demo.
- UltraCode is the most expensive way to run Claude Code by a large margin -- even the official docs call it out explicitly.
- One /effort command switch enables UltraCode for every subsequent task in the session -- you do not need to invoke it per prompt.
- The built-in QA layer is architectural, not optional -- a dedicated verification agent checks every UltraCode task automatically.
- /usage after any session breaks down token consumption by category, essential for UltraCode runs where costs accumulate across dozens of agents.
Two settings, one team, one big bill.
WHAT TO LEARN
UltraCode is not a step up from Max -- it is a different product: a parallel agent team with built-in QA, and the cost scales accordingly.
- The /effort menu is not a linear scale -- UltraCode sits beside Max rather than above it because it operates on a different dimension: parallelism, not just thinking depth.
- Dynamic workflows mean Claude first writes a plan, then forks execution across multiple agents simultaneously -- tasks that would take minutes serially can complete in seconds in parallel.
- Every UltraCode run routes through a dedicated verification agent before results return to you, making QA automatic rather than something you have to prompt for separately.
- If your task requires the absolute deepest single chain of reasoning, Max is the right choice -- UltraCode uses xhigh thinking, which is near-max but not identical to max.
- The /usage command after any session shows a full token breakdown -- for UltraCode in particular, checking this is essential because dozens of simultaneous agents burn credits faster than the session timer suggests.
Terms worth knowing.
- UltraCode
- A Claude Code mode that combines xhigh thinking with dynamic workflows, spinning up a coordinated team of up to 100 agents that work in parallel and verify each other output.
- Dynamic workflows
- A Claude Code capability that automatically plans a task, forks it into parallel sub-tasks assigned to separate agents, and aggregates verified results.
- xhigh
- A thinking level in Claude Code one step below max. Every message is processed with near-maximum reasoning depth, taking longer but producing more thorough responses.
- /effort
- A Claude Code slash command that opens a menu for selecting thinking depth: low, medium, high, xhigh, max, or ultracode.
- /usage
- A Claude Code slash command that displays a breakdown of token consumption for the current session, showing how many tokens each effort level burned.
Things they pointed at.
Lines you could clip.
01:10
“UltraCode is just two things that are bundled together.”
02:30
“It's actually not an effort at all.”
04:11
“UltraCode is the most expensive way to run Claude Code by a lot.”
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.
analogy
So Clot just got a new setting called ultra code, and it is awesome. Think about a task that needs to get done. Instead of an average worker grinding away for an hour to finish it, you have 100 incredibly smart workers each ready to take a small piece of that task and work it all at the same time.
That's what UltraCode does. Do you think it sounds expensive? Well, that actually might surprise you.
We're gonna get into that in a second. So in this video, I'm gonna walk you through two different pieces that make UltraCode so powerful. I'm gonna go through the big difference between UltraCode and max effort, and then I'll go through turning it on.
I'll give you a demo, and then I'll talk about the cost before we wrap it up. So let's get into it. Well, let's just look at the menu really quick because the menu explains most of it.
Inside of Claude code, if you type slash effort, this little list will pop up. These are all the effort levels. That's basically how hard you want Claude to think before it answers your question or does the task that you want to do.
And down here, you'll see low. Low effort is really fast, but it also thinks the least. And as you move up that list, Claude thinks a little bit harder and harder.
It does take a little bit longer, but at the top, you have max, which is the slowest, but also puts the most thought into every single answer. But right beside max is UltraCode, and notice what it says underneath here.
It says extra high, and it says dynamic workflows. And that's the whole secret to UltraCode. UltraCode is just two things that are bundled together.
And before we turn it on, let's go through each one of those separately. First, we have extra high. Extra high is a thinking level just below max.
So every single message that you send, Claude is thinking extra hard. Not the absolute highest, but it's close. The second piece here is dynamic workflows.
And what's cool about this is when you give Claude a task with dynamic workflows, it's gonna stop for a second. It's gonna come up with a game plan before it does anything, and then it's gonna spin up a bunch of agents. And each one of these agents is gonna grab a different chunk of your task and complete it all at the same time.
And the coolest thing about this is it doesn't just complete the task. It hands it back to a separate agent that's gonna check its work before it gives it to you. And on a big task, we're talking about hundreds of agents running at the same time if your computer can support it.
So that's the combination here. Extra high for deep thinking on every single step, and then dynamic workflows, which is a whole team running the job and then checking the work that it's supposed to be doing. But when you put those together, that is UltraCode.
And to switch it on, it's pretty easy. All you're gonna do is pick this right here. And once it's on, it's on for everything.
You don't have to ask for it again. Every single task that you give it for the rest of the session, it's gonna build a team of agents for you. And one thing I found really interesting here is UltraCode lives right here in the effort menu.
It's sitting right next to Max. So initially, when I saw it, I figured that it was just a higher effort than Max, but it's actually not an effort at all. So Max is an effort level.
Yes. And if all you care about is the deepest train of thought, then go with Max. What Ultracode gives you that Max doesn't is the team of agents.
And then, of course, the effort is extra high. Next, we're gonna show you a demo. But before we jump into that, if this has been helpful so far, do me a favor.
Please subscribe, like the video, Drop a comment. This really helps out the channel, and it's something that I really appreciate. But let's jump into a quick demo here.
Okay. So I have UltraCode on, and let's say that I'm thinking about starting a business like a food truck business. We know that if I'm starting a food truck business, there's a lot of stuff that I had figure out.
I had to figure out the startup cost, the licenses, the permits, where to get the equipment, where to get supplies, maybe mistakes that people are always making. Normally, if I was gonna do that manually, I'd be digging through hundreds of tabs. But instead, I'm gonna give this whole thing to UltraCode and ask it to research all of it for me and give me one clear starter guide.
Now take a look at the bottom of the screen. You see that little lightning bolt with a number that is climbing? Every single one of those is an agent that UltraCode has spun up.
Each one is chasing down something just a little bit different. One's looking for permits. The other one's looking for equipment cost.
The other one's looking for suppliers, and it's doing it all at the same time. If I wasn't using UltraCode, Claude would run this for us, but it would do one search at a time, and it could take a very long time. Let's take a look at what it gave me here.
Alright. So this is a clean starter guide. It's showing me what things actually cost, the permits I need, where to source everything, and all of this was put together, then fact checked.
Alright. Just a few moments ago, I did ask you if you thought that this was going to sound expensive, and I said it might surprise you. Well, the surprise is actually how expensive this thing really is.
This thing uses a lot of credits. It even calls it out in the documentation itself. Matter of fact, UltraCode is the most expensive way to run Claude code by a lot.
And it makes sense when you think about it. You're not just running Claude. You're running up to a 100 agents at the same time, burning tokens in a deep thinking level.
The whole idea behind this is to get the best possible answer, but the output is amazing. You If ever wanna see what UltraCode is costing you after a run here, you can just type in slash usage, and it'll break down the whole thing for you.
So that is UltraCode. There's two pieces, extra high thinking on every single step, and then a whole team of agents that's gonna split the work, and then it checks itself. If this was helpful, do me a favor.
Please subscribe. If you haven't seen my last video, you could check it out here. And leave a comment on what you think about Ultracode, the good, bad, or the ugly.
I don't care. Alright. I will see you in the next one.
The Hook
The bait, then the rug-pull.
What if instead of one worker grinding through a task for an hour, you could hand it to a hundred workers who each take a piece and run simultaneously? That is the premise the video opens with -- and it turns out to be a literal description of how UltraCode operates, not just a metaphor.
Named ideas worth stealing.
01:06model
xhigh + Dynamic Workflows = UltraCode
- xhigh thinking (near-max reasoning per step)
- Dynamic workflows (parallel agent team)
- Built-in verification agent
The three-layer architecture that defines how UltraCode differs from all other effort levels.
Steal forExplaining any product that bundles two distinct capabilities both necessary for the outcome
00:34list
Effort spectrum
- low
- medium
- high
- xhigh
- max
- ultracode
The /effort menu hierarchy, with ultracode sitting outside the linear thinking-depth progression.
Steal forProduct positioning where a premium tier is categorically different, not just incrementally better
How they asked for the click.
VERBAL ASK
04:32subscribe
“If this was helpful, do me a favor. Please subscribe.”
Standard mid-video ask at ~2:50 before the demo, plus a closing subscribe + related-video pair. The mid-video timing slightly undercuts demo momentum.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
Visual structure at a glance.
