Blindfolded AI Rebuilds 3D Part from Scratch
30sShows AI solving an impossible problem by building its own solution, making viewers rethink AI capabilities.
▶ Play Clip"Promises a how-to guide but delivers a hype-heavy overview with minimal actionable instructions."
Anthropic's new Claude Opus 5 AI model demonstrates unprecedented autonomy by writing its own tools to solve problems, correcting its own mistakes, and even pushing back on user instructions. This video showcases its capabilities through real-world tests and explains how it can be integrated into business workflows using the Agent OS system.
Given a drawing of a machine part then deprived of visual access, Opus 5 wrote its own program to read raw pixels and rebuild the part in 3D. No other AI achieved this in five attempts.
Opus 5 processed a messy customer spreadsheet, identified at-risk customers, alerted the team, and drafted a retention summary. CEO Wade Foster reported 100% success vs. older models failing completely.
Vercel reported Opus 5 opened its own web pages on desktop and mobile, found a button hidden off-screen on mobile, and fixed it without being instructed to check the phone version.
When given a real bug in a popular coding tool, Opus 5 found a hidden edge case that the human community had missed, while another AI only fixed the easy part.
Engineer Marcus Wang observed Opus 5 refusing a suboptimal instruction, explaining the good part, the problem, and offering a fix—acting like a smart teammate.
Opus 5 features a low-to-max effort dial. Even at low, it outperformed all other AIs on Zapier's business task test.
Opus 5 scored 43 on the Frontier Bench test, more than double the previous model's score of 19, in just a couple of months.
Anthropic noted Opus 5 is not their smartest for the longest jobs (Fable 5 wins there) and can still make mistakes, but it is the most trustworthy model yet, at the same price as the old model.
Claude Opus 5 represents a significant leap in AI autonomy and reliability, enabling individuals and businesses to automate complex tasks with minimal supervision. The key is pairing it with proper setup like Agent OS to unlock its full potential.
What did Opus 5 do in the machine part test?
It wrote its own program to read raw pixels and rebuild the part in 3D after being deprived of visual access.
00:28
What was Opus 5's success rate on Zapier's business task test?
100% success, while older models failed completely.
00:52
How does the effort dial on Opus 5 work?
It has low, medium, high, and max settings that control how hard the model thinks. Even on low, it outperforms other AIs.
03:29
What score did Opus 5 achieve on the Frontier Bench test, and how does it compare to the previous model?
Opus 5 scored 43, more than double the previous model's score of 19.
03:59
What is a limitation of Opus 5 according to Anthropic?
It is not their smartest model for the longest jobs; Fable 5 wins there.
05:03
Self-directed tool creation
Demonstrates a new level of AI autonomy: building its own solution when faced with an obstacle.
00:28Unprompted cross-platform testing
Shows the model proactively checks its work on different devices without being told.
01:33Constructive pushback
Highlights the model's ability to evaluate human instructions and suggest better alternatives.
02:24Doubling performance in months
Quantifies the rapid improvement in AI agent capabilities.
03:59[00:02] is how. Anthropic just dropped a new AI model called Claude Opus 5 and it changes what one person with a laptop can do. Real quick, so you know where I'm coming from, I run my whole setup through something I built called the
[00:14] Agent OS. Think of it like a control room for your AI and the engine I run inside it is Claude Opus 5. So, everything I'm about to show you I'm me show you the thing that stopped me cold. Anthropic ran a test. They gave
[00:28] the AI a drawing of a machine part. Then they took away its eyes. It couldn't even look at the picture. Most AI would just quit right there. Opus 5 didn't quit. It wrote its own little program to see the drawing. Then it read the shapes
[00:40] off the raw pixels. Then it rebuilt the whole part in 3D. It did this over and over. No other AI could do it, not once, even after five tries. Think about what that means. You give it a wall, it builds its own ladder. That's new.
[00:52] Here's another one. Zapier is a big automation company. They handed Opus 5 a messy spreadsheet full of customer accounts. Opus 5 found the customers who were about to leave. It warned the right person on the team. Then it wrote up a
[01:05] summary for the retention crew, start to finish, on its own. Wade Foster, the CEO of Zapier, said older AI models flat-out failed this test. Opus 5 hit 100%. Hey, if we haven't met already, I'm the digital avatar of Julian Goldie, CEO of
[01:20] helping clients get more leads and customers, I'm here to help you get the latest AI updates. Julian Goldie reads every comment, so make sure you comment below. So, what makes it different? Two things and they're simple. First, it
[01:33] checks its own work. Most AI writes something, says done, and walks away, even if it's wrong. Opus 5 goes back and looks. One company, Vercel, said it opened its own web pages on a computer screen and a phone screen. It found a
[01:46] button that was hidden off the edge on mobile. Then it fixed it before handing the work back. Nobody told it to check the phone version. It just did, like a careful worker who double-checks the door is locked. Second, it doesn't give
[01:59] up when things break. Give it a hard bug, it hunts down the real cause, not just the surface problem. There was a real bug in a popular coding tool. The human community had patched it, but they missed a hidden edge case. Opus 5 found
[02:12] the part they missed and fixed it. Another AI just fixed the easy part and said, "All done." It wasn't done. Now, here's the part I love. It talks back when you're wrong. One engineer, Marcus Wang, said he told Opus 5 to build
[02:24] something a certain way. Opus 5 pushed back. He pushed harder. It still didn't "Here's the good part of your idea. Here's the one problem. Here's a fix that keeps both." That's not a robot following orders. That's more like a
[02:38] smart teammate who actually cares if the thing works. Quick pause here because this matters for you. If you're watching this and thinking, "This is cool, but I because you don't need to code anymore. That's the whole point, and that's where
[02:51] the AI Profit Boardroom comes in. Inside the full Agent OS is inside the AI Profit Boardroom. The complete zip file ready to install. You download it, you drop it in. We don't just talk about Opus 5. We show you how to put it to
[03:04] work. We've got a coaching call every week where we go deep on building Opus 5 agents that handle real jobs in your business, catching customers before they leave, sorting your inbox, writing your follow-ups. You bring your setup, we
[03:16] help you fix it live. Daily step-by-step tutorials on Opus 5 workflows and a 30-day roadmap so you know exactly what to build first. The link is in the description. Okay, back to it. Let me explain the effort dial because this is
[03:29] huge for normal people. Opus 5 has a knob, low, medium, high, and max. Turn up, it thinks harder on tough stuff. You control it. So, easy jobs stay cheap. Hard jobs get the deep thinking. You're not stuck paying for a genius to answer
[03:44] a baby question. And here's the wild part about that dial. Even on its lowest setting, Opus 5 beat every other AI on Zapier's business task test. Its worst effort still won. Now, why does this matter more than the last 10 AI updates?
[03:59] Because of the jobs it can now finish alone. There's a test called Frontier Bench. It's hard agent work in a computer terminal. The old model, Opus computer terminal. The old model, Opus 4.8, scored about 19. Opus 5 scored 43,
[04:11] more than double, in a couple months. Let me put that in plain talk. A year ago, AI could help you write one email. Now, it can run a five-step job across coffee. That's the jump. That's why people who wait keep falling behind.
[04:24] This thing doesn't just do the task, it fills in the gaps you forgot. It handles the parts you didn't think about. So, what do you actually do with this? Say you run a small business. You could point Opus 5 at your list of leads and
[04:36] ones, then write a first message for each. One job, start to finish, while you sleep. Or say you're drowning in customer emails. Opus 5 can read them, sort them, and draft replies in your voice. You just check and hit send.
[04:51] Notice something. None of that is coding. That's just work, real work, the let's be fair for a second. I'm not going to sit here and tell you it's going to sit here and tell you it's magic. Anthropic themselves said Opus 5
[05:03] still isn't their smartest model for the hardest days-long jobs. Fable 5 still wins there. And any AI can still make mistakes. You still need a human checking things. But here's what's different. Opus 5 makes way fewer
[05:16] mistakes than the old ones. Anthropic tested it and called it the most trustworthy model yet. It lies less. It's harder to trick. It doesn't go rogue and do something it can't undo, so you can hand it more and hover less. And
[05:29] it's the same price as the old model, same cost, way more brain. Anthropic says it comes close to their top model, Fable 5, at half that price. So, this isn't some pricey toy, it's the cheap everyday workhorse that happens to be
[05:43] coming back to. The gap isn't between smart people and dumb people anymore. It's between people who know how to hand work to these tools and people who don't. That's it. That's the whole game now. A year ago, building
[05:57] developer, waiting weeks, paying a fortune. Now one person who knows how to afternoon. The person next to you who learns this will pull ahead fast. Not picked up the tool first. So let me make this easy for you. Here's the problem.
[06:13] Opus 5 is powerful, but out of the box, it's just a smart brain sitting there. It doesn't know your business. It doesn't know your leads, your emails, your follow-up steps. You have to set all that up. And that setup is where
[06:25] most people get stuck and quit. So I built something to skip that whole part. It's called the Agent OS. Like I said at the start, think of it like a control room for your AI. It's a ready-made setup that tells the AI who you are,
[06:38] what your business does, and exactly what jobs to run, sorting leads, handling emails, writing follow-ups. All of it wired up and ready. You don't build it from nothing. It's already built, and I run Claude Opus 5 inside
[06:50] it. That's the engine. The Agent OS is the driver's seat. Opus 5 is the motor. You drop Opus 5 in, and now that smart brain actually knows your business and gets to work. And you can get the whole thing. The full Agent OS zip file is
[07:04] inside the AI Profit Boardroom, ready for you to install. You don't build it from scratch. You grab the file, drop it in, plug in Opus 5, and start running agents in your business today. Plus the weekly coaching calls where we go deep
[07:16] on Opus 5, the daily tutorials, the 30-day roadmap, and 2,800 business owners in there already doing this. A lot of them running these exact agents description and the comments. And if you want the full process, the step-by-step
[07:30] guides, and over 100 AI use cases like the ones I just showed you, join the AI the notes from this video, plus a community of 85,000 people who are all figuring this out together. Links in the description and comments, too. Look, new
[07:45] AI models drop all the time. I get it. It's a lot. But, this one's different in a way that actually touches your day. It finishes jobs. It checks itself. It fixes what it breaks. And it costs the same as the model before it. The people
[07:58] look up in 6 months and be way ahead. Not because they worked harder, because they started today. You can be one of them. Go grab the Agent-OS and start them. Go grab the Agent-OS and start building.
⚡ Saved you 0h 08m reading this? Transcribe any YouTube video for free — no signup needed.