The Most Powerful Claude Code Feature In Months Dropped & Nobody is Talking About It
CChase AI
Computing/SoftwareInternet Technology
Transcript
00:00:00This is the most powerful Claude Code update in months, and nobody's talking about it.
00:00:04UltraCode allows Claude Code to handle extremely large, complex tasks by spinning up an army
00:00:10of agents. And I'm not just talking about a handful. I'm talking about potentially hundreds
00:00:14of agents to split up the work and get it all done in one go. But the real power of UltraCode
00:00:20isn't the amount of agents it can spawn. It's the fact that it builds a custom harness
00:00:24tuned to your specific task on the fly. Normally, Claude Code takes a one-size-fits-all
00:00:29approach to solve your problem. But with this brand new feature, that's no longer the case.
00:00:34So in this video, I'm going to break down exactly how it works so you can start leveraging this
00:00:38new feature right away. So why should you care about UltraCode and dynamic workflows? Well,
00:00:43it's pretty simple. It is the best way to handle large, complex tasks with Claude Code. That's
00:00:49it. And the way it does it is through custom harnesses. That's a fancy way of saying it's
00:00:56going to create a novel way to solve whatever it is you're trying to solve. A great example is how it
00:01:02handles something like deep research. If you have a rather complicated question, like, should we migrate
00:01:08our checkout service to a new provider? If I'm just using Claude Code as it normally operates, I just
00:01:14hop inside the prompt window and I give it this prompt. Well, it's going to use its static default
00:01:20harness. It's going to be in one single session, right? We're going to use one context window.
00:01:24It's going to do a handful of web searches. It's going to fetch the top results. It's going to verify,
00:01:30summarize, and give us a generic research report. By default, we're talking about static harness. What
00:01:34we're really saying is you just open up Claude Code here in one single context window, and we're not
00:01:38telling it to use sub-agents or do any sort of like antagonistic review of its original thought
00:01:44process. So it's rather simple. It uses less tokens, and in the end, you get a generic answer. Caveat,
00:01:51that's fine for most issues. We're talking here today about large, complex stuff. By contrast, if we use
00:01:57UltraCode in a dynamic workflow, and I'll talk about in a second the difference between those two, we get a
00:02:02custom harness that is built for that problem. We are going to have a different way of solving that issue.
00:02:09We're not just going to sit in one context window. We're going to bring in different agents, and the different
00:02:12agents are going to do different things, again, depending on the task. So in this example,
00:02:17where I say, hey, should we migrate our checkout service to a new provider, instead of just doing
00:02:22some web searches and coming up with a summary, instead with a dynamic workflow, it's going to be
00:02:26like, okay, well, what is our checkout service? So it's going to read our billing code. It's actually
00:02:30going to go deep into how we actually operate. From there, it's then going to check the features
00:02:35against the new provider's documentation. It's going to price our transaction volume. It's going to have an
00:02:40actual devil's advocate agent to say, hmm, does that actually make sense? And then in the end, we get a specific
00:02:47recommendation instead of a generic one. So the big thing with dynamic workflows is, again, the process by which
00:02:55we come up with an answer is different. It's customized, and it depends on the question we are asking. And by doing it
00:03:02this way, we get better results. That's why you should care. Now, before we talk about where UltraCode plays
00:03:07into all this, first, a message from today's sponsor, me. So not too long ago, I just released a Claude
00:03:13Code Masterclass, and it is the number one way to go from zero to AI dev, especially if you don't come from
00:03:19a technical background. I update this every week. I've recently added modules for a Codex Masterclass as
00:03:24well. And if you want to get your hands on this, you can find it inside of Chase AI Plus. I will put
00:03:29a link to that down in the pinned comment.
00:03:31So what is UltraCode? They must have cared about it to give it this fancy graphic when you bring up
00:03:35forward slash effort. So UltraCode is related to effort level. So if you do forward slash effort
00:03:41inside of Claude Code, you will see the spectrum that pops up from low all the way to UltraCode.
00:03:46Normally, by default on Opus 4.8, we're on high, but UltraCode goes kind of a step beyond max. Well,
00:03:53kind of. What actually happens when I do forward slash UltraCode, two things occur. One,
00:04:00my effort level goes from high to extra high. We're not on max. We're just jumping to extra high.
00:04:05And secondly, I now have automatic dynamic workflow orchestration. So there's two things we're talking
00:04:11about here, right? In this video, we've talked about dynamic workflows, which is, you know, this idea that
00:04:16we're doing custom harnesses for our task, and we have UltraCode. Okay, so UltraCode changes effort
00:04:22to extra high and allows for dynamic workflow orchestration automatically. Dynamic workflows are
00:04:29also its own thing inside of Claude Code. So if I'm inside of Claude Code and I do something like
00:04:34forward slash workflows, I'm now forcing Claude Code to create a workflow for whatever prompt I give it,
00:04:42you know, like insert task. But if we're in UltraCode, which we are now, Claude Code will on its own
00:04:48decide, hey, does this need a dynamic workflow or does it not? So we have the option to always invoke
00:04:55dynamic workflows if we just do forward slash workflow or if I say something like, hey, use
00:04:59workflows. So it's almost like a skill that I can invoke, same sort of deal. Or UltraCode allows
00:05:06Claude Code to be kind of smart about it. And it's going to be like, hey, depending on the prompt,
00:05:10sometimes we'll just do the static harness. We don't always need to go crazy. Or if it's complex
00:05:15enough, hey, I'll go dynamic workflows. So that's where UltraCode comes in. It's like an automatic
00:05:20layer. You don't have to think about it. If it needs dynamic workflows, it will do it. You don't have to
00:05:23use your brain. Right? And that's great. So with that being said, to actually maximize UltraCode,
00:05:30we need to dive a little bit more into dynamic workflows because you understand the why you
00:05:34understand UltraCode versus dynamic workflows. Now, let's talk about this a little bit more.
00:05:39And Claude Code actually, Anthropic wrote an entire blog about this. So the blog, which came out about
00:05:44a week ago, is called A Harness for Every Task, Dynamic Workflows in Claude Code. We're not going to
00:05:48dive into this whole blog in this video. I'm just going to hit the parts that you need to know about.
00:05:52And I'll put a link to it down in the description as well. Now, the first thing I want to highlight is
00:05:55why dynamic workflows. We touched on it at the beginning. Hey, it's better for complex tasks.
00:05:59But why is Claude normally not good enough for this thing? Well, that is because the longer Claude
00:06:04works on a complex task in a single context window, the worse it becomes. And the three things they
00:06:09talk about here can kind of all fall under the context rot umbrella, the idea of agentic laziness.
00:06:16You know, we've all run into this where you tell Claude Code to do something that's rather large in scope,
00:06:20and it kind of does some of it. We also have self-preferential bias, referring to Claude's
00:06:25tendency to prefer its own results or findings, especially when asked to verify or judge them
00:06:30against a rubric. Hey, if you remember my video a couple of days ago where I showed you grill me
00:06:34codex, bringing codex into the equation, this is something I talked about there as well, right?
00:06:38Claude isn't great at evaluating its own work, especially if you're telling it to evaluate work
00:06:44within the same session. Like you're in the same context window, like it's not great at it.
00:06:49And then lastly, goal drift. And again, keep beating this drum, complex task, goal drift is
00:06:54going to be a big thing. We have to handle this. We can't do these things in one session.
00:06:57And so creating a workflow, a dynamic workflow helps combat these by orchestrating separate
00:07:03Claude sub-agents with their own context windows and focused, isolated goals, right? Sound familiar?
00:07:10GSD, superpowers, all these things, these last few months are all coming to this one point of like,
00:07:16how do we handle big tasks with a limited context window? It all ends up coming back to like sub-agents,
00:07:21fresh context windows, that sort of thing. Now, the last thing I'll touch on from this blog is some of
00:07:26the workflow patterns they bring up. This is not exhaustive. There's an infinite amount of workflow
00:07:30patterns, but I think it helps when we talk about custom harnesses and talk about dynamic workflows to see
00:07:35what we're actually speaking of, like visually. Now, the first example they give is classify and act.
00:07:40That's a workflow pattern where we have some sort of task. That task involves a number of sub-tasks,
00:07:45and we want to divvy those out to appropriate sub-agents. To do that, we need a classifier and using a
00:07:51dynamic workflow. Claude code will already know, this is the big part, Claude code will already know that
00:07:56this is the sort of workflow pattern you need, right? It will automatically set this up for you.
00:08:00Another one is fan out and synthesize. Think of deep research. Hey, I want you to research some sort of thing.
00:08:05I need you to go out there and get a ton of information from a ton of different sources,
00:08:09potentially hundreds of sources. I need you to bring them in. I don't just want you to summarize.
00:08:13I want you to verify. I want you to actually cross-reference it, and then at the end,
00:08:17I want a final report. Again, if you ask Claude code using ultra code or using dynamic workflows
00:08:22to do deep research on some sort of thing, well, this is the sort of workflow to come up with.
00:08:27And it continues down the line. Adversarial verification, loop until done, a tournament style
00:08:32thing where you have a bunch of different ideas and judges, and at the end, you get a final winner,
00:08:36generate and filter. The blog itself goes into detail on all these, but at least for me, it was
00:08:41helpful to kind of see what they're talking about. I'm like, all right, custom harness, what does that
00:08:44exactly mean? Well, it just means there's a number of paths to find a solution. And with a custom
00:08:50dynamic workflow, we get all these options. And the point is they're going to be custom fit to the task
00:08:55versus, you know, static harness, do a web search, ask a couple questions, summarize. We can do better.
00:09:03And lastly, before we go into the demo, I'm going to highly suggest that you take five minutes of your time,
00:09:09go to the Claude code docs and actually read what they have written up about dynamic workflows so you can get
00:09:13a better sense of how it's working under the hood versus something like Agent Teams. Hint, it's actually running
00:09:18a script at runtime execution and things like how to save workflows because you can repeat them. They're
00:09:24kind of like skills in that sense and that sort of thing. They actually have a really good write-up
00:09:28here. And in this write-up, they actually tell us about a preloaded dynamic workflow that comes with
00:09:34Claude code. It's actually a deep research workflow, similar to the deep research that's been around for
00:09:39a while on the actual web app. So we'll do two demos. The first one I'm going to show you so you can kind
00:09:44of see how this all works is this deep research one. And this is inside your Claude code right now if
00:09:48you're updated. All you have to do is do forward slash deep research. So inside of Claude code,
00:09:52we're going to do forward slash deep research. And then I'm just going to give it a prompt. Well,
00:09:56actually, it'd be kind of a meta prompt. We're going to have it do deep research on dynamic workflows
00:10:01and the best practices for creating them. Can you do some deep research on the brand new dynamic
00:10:08workflows and ultra code within Claude code? I want to report detailing the best practices.
00:10:14There's a lot of talk about there being custom harnesses.
00:10:17With the dynamic workflows, how do we make sure the custom harness that's built
00:10:22using dynamic workflows is the best one for the job? Or is that something on the user level we just
00:10:28expect Claude code to handle? So there we go. So when we did that, you see we got a couple of
00:10:32messages. Running deep research workflow, topic clear enough, proceed, workflow, dynamic workflow,
00:10:38deep research. And then it's telling us the workflow has been launched in the background. It has
00:10:43five phases, scope, search, fetch, verify, and synthesize. And we have the ability to watch it
00:10:49live. So if I go ahead and do forward slash workflows, you can see right here, all of the agents is what is
00:10:57happening in real time. And so for the scope, we just have a single agent. For search, we will have
00:11:02five agents. And as they start working, we can actually, oops, once they start working, we can
00:11:07actually see their token usage. Because one of the big costs of this, right, you can kind of see it
00:11:13right here. One of the big costs we need to think about when it comes to using dynamic workflows and
00:11:17ultra code is the token cost. It is token heavy. Now, there's definitely an argument to be made that
00:11:23we're just kind of front loading the cost. And that by virtue of us having more effective results from
00:11:30ultra code and dynamic workflows, we're probably saving tokens in the long run. But just don't be
00:11:35surprised, especially if you're using a dynamic workflow that I wasn't joking in the beginning is using 100
00:11:41plus agents. Don't be surprised if you have a wild token cost at the end of that. So we can see here for our
00:11:46five agents that are searching right now. They've used about 250,000 tokens a pop. Scope itself took
00:11:53about 40k. And then fetch over here looks like it has potentially 12 sub agents. And then we also have
00:11:59a list up here, right? Four out of 22 agents and two minutes have passed. And I'll also reference my
00:12:06total weekly usage at the end of this as well and how much I burned. So 101 agents, 3.7 million tokens
00:12:12and 11 minutes later, the workflow is complete. And in terms of usage on like my weekly max plan,
00:12:18it was 4% and I'm on the $200 a month max plan. So, you know, token usage stuff, it is no joke. Like,
00:12:25you have to know when to apply this. And then I had to go ahead and turn the report into this HTML
00:12:30asset you see here. And it's pretty much restating a lot of what we see in the actual cloud code
00:12:35documentation involving dynamic workflows. So what are some other use cases for dynamic workflows in
00:12:42UltraCode? Well, Anthropic actually spells it out for us. Things like code-based wide bug hunts,
00:12:46large migrations, and critical work that needs to be checked twice. Another example Anthropic brings up
00:12:52is rewriting Bun with dynamic workflows where they actually ported Bun from Zig to Rust over the course
00:13:00of about a week using this feature. Let's test it out with a bug hunt. I'm inside the directory
00:13:05for my AI agency website, which also doubles on the backend on an admin side. It's sort of like my
00:13:11content creation command center. So we're going to have it run a dynamic workflow to sort of do a bug hunt.
00:13:18Can you go ahead and use dynamic workflows inside this directory to run a bug hunt and see what sort
00:13:26of bugs we have in this current directory? Once you create the report, turn it into an HTML report and
00:13:33bring that up in my browser. So you can see right here calling the workflow command cluster parallel bug
00:13:38hunt across the next JS app. Adversarially verify each finding synthesize a severity ranked report.
00:13:45So it's running in the background. And as always, we can do forward slash workflows to take a look. So for this bug report, it ran in
00:13:51about half the time and took half the amount of tokens as deep research. We had 34 confirmed bugs. It had seven bugs that were false positives. And of those 34, two are high, nine are medium, and 23 are low.
00:14:04have the ability to click on them. It shows me what's wrong, the evidence, the fixed, and then
00:14:10also the adversarial verifier that's saying like, hey, this actually is a bug. Here's the problem.
00:14:15And it gives me this report for pretty much all of them, where the issue is, what it kind of defines
00:14:21the error as, what's wrong, evidence, fixed, adversarial verifier. So it's pretty deep in terms
00:14:26of what it's actually finding. And I think the best part of this is the adversarial verifier.
00:14:30Because again, one of the big issues with cloud code and complex tasks is like, can you confirm
00:14:34that it actually did what it's supposed to? So that's where I'm going to leave you for this
00:14:38video. We went over why you should care about ultra code in dynamic workflows, how they work,
00:14:43and went through a few examples. I think this is an awesome feature. I think it's extremely
00:14:47powerful. Yes, it's extremely token heavy, but sometimes we do need the big guns, especially
00:14:53for tasks we really, really care about. And before this, it was kind of hard, right? We're doing some
00:14:58hacky things. We're bringing in outside orchestration layers. And now it's all ready to go inside
00:15:03of cloud code itself. So as always, let me know what you thought. Make sure to check out Chase
00:15:08AI Plus if you want to get your hands on my cloud code masterclass, and I'll see you around.
Community Posts
No posts yet. Be the first to write about this video!
Write about this video