Claude Opus 4.8 :Build and Automate Anything!

Añadido: 2026-06-01

[00:00:00]Claude Opus 4.8, build and automate anything. Prop just put out a brand new AI on Thursday. It's called Claude Opus 4.8 and it does one thing that most AI tools still won't do. Tells you when it's not sure. I'll say that again a different way because it matters. When this new AI does a job for you and part of that job is shaky, it flags it. It says, "Hey, I'm not confident about this bit." Most AI just hands you the work and acts like it nailed it when it didn't. That's the headline for me, not the speed, not the benchmarks, the honesty. And I'll prove it with a real number. Tropic ran their own tests. They found Opus 4.8 is around four times less likely than the last version to let mistakes slip by without saying anything. Four times. So, if you give it a job and there's a flaw in the work, the old AI might just stay quiet and hand it over. This new one is way more likely to stop and point it out. Why does that matter to you? Because if you run a business, the scary part of AI was never that it's dumb. The scary part was that it sounds confident when it's wrong. You can't trust a worker who lies to your face with a smile. Now you've got an AI that owns up. That changes how much you can hand to it. Stick with me because the honesty thing is just the start. There's a part later where this AI splits one big job into hundreds of small jobs and does them all at the same time. Get to that. It's wild in the best way. First, let me set the table. Quick.

[00:01:14]Tropic makes Claude. Claude is the AI a lot of business owners use to write, plan, sort, and build stuff.

[00:01:20]It's their top model, the smart one, the one you give the hard jobs to. 4.8 is the newest version and it just replaced the last one, 4.7. Here's the part that tells you how fast this is moving. Last version, 4.7, only came out 41 days ago.

[00:01:34]41 days. And now there's a better one. A year ago, AI updates like this came every few months. Now it's every few weeks. The gap between good and great keeps shrinking and it's shrinking fast.

[00:01:44]So, what actually got better? Tropic says three things. Sharper judgment, more honesty about its own work, and the ability to work on its own for longer than before. Let me take those one at a time because each one matters for a normal person trying to get stuff done.

[00:01:58]Sharper judgment first. People who tested it early said it's more reliable and sharper in its judgment on real tasks. Plain English, that means it makes fewer dumb calls. It picks the right next step more often. Think of it like the difference between a new hire who needs you to check everything and a seasoned one who just gets it. You spend less time fixing and more time moving.

[00:02:17]Quick example, you could ask it to clean up a messy list of people who signed up for AI Profit Boardroom and pull out only the ones who asked a question.

[00:02:24]Better judgment means it actually gets the right people, not a random mess you have to redo. Now, honesty. I touched on this, but go deeper with me. Testers found Opus 4.8 is more likely to flag the stuff it's unsure about and less likely to make claims it can't back up.

[00:02:38]When it writes you something, it's pickier about what it states as fact. If it doesn't know, it's more likely to say so instead of guessing and sounding sure. Picture the worst kind of helper, one who never says, "I don't know." They just make something up so they look smart and you find out later it was all wrong. That helper costs you more than no helper at all.

[00:02:55]4.8 is built to be the opposite of that.

[00:02:57]Would rather tell you, "I'm not certain here." than fake it. For you, that's trust. Trust is the whole game. The more you can trust the work, the less you have to baby sit it. Less you baby sit, the more you can hand off. That's how you actually save hours instead of just playing with a toy. Works on its own for longer. The old versions would do a chunk of work and then sort of run out of steam on big jobs. This one can keep going. It can stick with a long task from start to finish without you holding its hand the whole way. Here's why that's a big deal for a busy owner. Most of us don't have a big block of free time problem. We have a I keep getting pulled away problem. So, an AI that needs you every 2 minutes isn't real help. An AI that can take a job and run with it while you do something else, that's the dream. You set it off, you step away, you come back to finish work.

[00:03:38]Oh, quick, before I show you the best part. If you're watching this and thinking, "Okay, I want to actually use Claude Opus 4.8 to take work off my plate, but I don't know where to start."

[00:03:46]That's exactly what we do inside the AI Profit Boardroom. We run four live coaching calls every single week and right now a big chunk of them are about setting up Claude to run your repeat tasks for you. We've got daily step-by-step tutorials showing you how to use the new effort control and the longer work sessions on real business jobs. And we've built a 30-day road map just for getting Claude doing your busy work, sorting leads, writing your follow-ups, building your pages, so you stop doing it by hand. Plus a prompt library full of prompts that already work, and a member map so you can find people near you doing the same thing and swap what's working. Links in the comments in the description or just go to aiprofitboardroom.com.

[00:04:22]The part I teased.

[00:04:24]8 came out with a new feature called dynamic workflows. It's in early preview right now, so it's fresh. And here's what it does. It can take one huge job, break it into hundreds of smaller jobs, and run all those small jobs at the same time, single session. Anthropic showed this off with big coding jobs, where it can rework a giant project across hundreds of thousands of lines from start to finish. Now, most of you watching aren't coders, and that's fine.

[00:04:47]The point isn't the code. The point is the shape of it. The shape is a job that used to be too big for one tool is now doable because it doesn't work like one tool anymore. Works like a crew. And that shape spreads to normal work over time. Big slow jobs you've been avoiding because they'd eat your whole week, those are the ones that start becoming a set it and walk away thing. Now, let me hit the speed because they fixed something that used to drive people nuts. The old version had a habit.

[00:05:10]Sometimes it would think way too hard about a simple question. You'd ask it something easy, and it would sit there chewing on it like it was solving world peace. It wasted your time. So, Anthropic added a control. It's a little dial in the menu where you pick your model. You choose how much effort you want the AI to put in. Low effort for quick stuff. Then medium, high, and max for the heavy jobs. You're in charge now. Think of it like talking to a worker. Sometimes you want a fast gut take. Sometimes you say, "Take your time. Get this perfect. I'll wait." Now you get to tell the AI which one you want instead of it guessing and getting it wrong. And on top of that, the fast mode runs about 2.5 times quicker than before. So, when you want speed, you really get speed. You ask, it moves.

[00:05:47]Those together and you see the picture.

[00:05:49]You can flip it to fast and low effort to bang out quick stuff all morning.

[00:05:52]Then flip it to max effort and walk away while it grinds on the one big project you've been dreading. I two totally different gears you drive. Here's a clean way to use that. You could set it to high effort and ask it to build a simple page that explains AI Profit Boardroom, what people get, why it helps, who it's for, and let it take its time to get it right. Now let's talk about how it stacks up because the numbers back the story. Anthropic put Opus 4.8 up against the top models out there. GPT 5.5, Gemini 3.1 Pro, best the other labs have. On a hard test for real coding work called SWE-Bench Pro, Opus 4.8 scored 69.2%.

[00:06:28]Last version scored 64.3. Jumped about five points in 41 days. And on that test, it beat both GPT 5.5 and Gemini 3.1 Pro. Got better at reasoning across different subjects while using tools.

[00:06:41]That jumped from 54.7 up to 57.9. Plain words, it's better at handling messy real jobs that mix a bunch of things together. Most of your actual work is exactly that kind of messy mix. Now I want to be straight with you because hype helps nobody. Anthropic didn't oversell this. Their own words, they called it a modest but tangible improvement over the last version.

[00:07:00]Modest but real. They're not screaming that it changes the world overnight.

[00:07:04]Some reviewers agreed. They said you'll notice it on bigger jobs, but it's not a total game changer on day one. Respect that they said it plainly. And honestly, that's the right way to read every one of these. The point isn't that Thursday's release flipped a switch. The point is the direction. Sharper, more honest, works longer on its own, and a new better one lands every six weeks now instead of every six months. Get the pushback, by the way. Hear it. You are tired of a new AI dropping every couple of weeks. Think, why bother learning this one if there's another one next month?

[00:07:32]Felt that, too. Here's the thing. The pattern doesn't change. The skills don't reset. Once you know how to hand off a task, how to set the effort, how to let it run, how to check the parts it flags.

[00:07:42]That skill carries over to every single version after this. You're not learning a tool, you're learning a way of working. And that way of working only gets more powerful as the tools get better, which brings me to the open loop I'll leave you with. Tropic also dropped a hint at the end of this release.

[00:07:56]They've got an even bigger class of model code named Mythos that's only gone out to a few companies so far. They held it back over safety worries.

[00:08:04]They said they expect to bring Mythos class models to everyone in the coming weeks. Opus 4.8 is the strongest one you can use today, and there's a bigger one waiting in the wings. People who get good at handing work to the AI now are the ones who'll be ready the second that bigger model lands.

[00:08:20]Starts from zero again while the early folks are already flying. Let me bring this all the way back down to your day.

[00:08:26]You're busy. You're pulled in 10 directions. You've got jobs piling up that you keep pushing to later.

[00:08:31]Sign-ups, writing the follow-ups, building the page, turning one call into a week of posts. None of it is hard.

[00:08:37]It's just time, and time is the thing you never have. 4.8 is the first version where I'd actually trust it with that pile because it works longer on its own, because it tells you when it's unsure instead of faking it, because you can dial it up for the big jobs and down for the quick ones, and because it's clearly going to keep getting better fast. You don't need to be a tech person. You don't need to code. You just need to learn how to hand off work and check the parts it flags. That's it. That's the whole skill, and it's the most valuable skill you can build right now. So, here's what I do today.

[00:09:06]Claude, pick one job you hate doing.

[00:09:08]Just one. The effort to hire. Get over.

[00:09:11]Watch what it gives back, then check the bits it flagged. That's your first rep.

[00:09:15]Do that once and you'll feel it click.

[00:09:16]If you want help doing exactly that, taking the work off your plate with Claude Opus 4.8, come join us inside the AI profit boardroom. We've got four live coaching calls a week where we go deep on getting Claude to run your repeat jobs, and you can ask about your own setup live and get answers on the spot.

[00:09:31]There's a fresh step-by-step tutorial every day showing real ways to use the new effort control and the longer work sessions on actual business tasks. We've got a 30-day roadmap built around getting Claude doing your busy work, so you stop trading hours for stuff a machine can handle. Prompt library full of prompts that already work, so you're not staring at a blank box. And a member map with 2,800 other owners doing the same thing near you, so you've always got someone to ask. Links in the comments and the description, or head to aiprofitboardroom.com.

[00:09:58]And if you want the full process, the SOPs, the step-by-step breakdowns, and 100-plus AI use cases just like this one, come join the AI Success Lab. It's our free community.

[00:10:08]In the comments and the description.

[00:10:10]You'll get all the notes from this video over there, plus access to 67,000 members who are using this stuff every single day. Come hang out. Help you get started.