These NEW Claude Updates are INSANE!

This new Claude update is genuinely insane. Claude can now dream. Yeah, I know how that sounds. Stay with me.

Thropic just rolled out 4 big upgrades to Claude managed agents, and one of them lets the AI literally review its past work while you sleep, find its own mistakes, and get better on its own. Harvey, the legal AI company tested it and saw task completion rates go up around 6 times. 6 times. Not from a smarter model, just from the AI learning from itself.

Let me say that again in plain English. Before this update, AI agents were like a new hire on day 1 every single day. They forgot everything. They made the same mistakes over and over.

You had to babysit them. Now, Claude looks back at its own past sessions, spots patterns, fixes its own memory, and shows up sharper the next time you use it. That's the headline feature. Call it dreaming, but there are 3 more upgrades that matter just as much.

Outcomes, multi-agent orchestration, and webhooks. I'm going to walk you through all 4 in plain English and show you exactly why this changes how you can use AI in your business. The announcement happened at Anthropx code with Claude event on May 6th. Dreaming is in research preview right now.

Outcomes, multi-agent orchestration, and webhooks are in public beta. So this is live. Developers can request access today. Let's start with dreaming because that's the wild one.

Hey, if we haven't met already, I'm the digital avatar of Julian Goldie, CEO of SEO agency Goldie Agency. Whilst he's helping clients get more leads and customers, I'm here to help you get the latest AI updates. And Julian Goldie reads every comment so make sure you comment below. Here's how it actually works.

When your Claude agent finishes a task, it writes notes to its memory. Normal stuff. Over many sessions, that memory gets messy. Duplicates, contradictions, old info that's no longer useful.

Sound familiar? It's like a notes app you never clean out. Dreaming is a scheduled background process. Runs between your sessions.

Reads through all the past work. Pulls out the patterns. Cleans up the memory. And it writes new notes that the agent can use next time.

Thropic gave a really clear example at the event. They ran a simulation across six landing sites for a space mission. The first run was good but not perfect. They pressed one button to trigger a dreaming session.

For night, the dreaming agent reviewed all the past sessions and wrote a detailed playbook full of lessons learned. The next morning, they ran a new simulation. Results were way better. Especially on the sites that had failed the night before.

One button. That's it. Harvey is the big legal AI company. Their agents handle long documents and legal drafting.

With dreaming turned on, their agents started remembering small but important things like file format quirks and tool specific tricks. Completion rates went up around six times in their tests. That wasn't from a new model. That was from the agent carrying lessons across sessions.

Model. Way better results. Just because it stopped forgetting. Quick pause.

If you're sitting there thinking, okay, this is cool, but how do I actually use this for my business? That's the whole reason I made this video. Because most people are going to watch this and do nothing. And in six months, they're going to wonder why their competitors are five steps ahead.

Inside the AI profit boardroom, we've already started building playbooks around the new Claude managed agents features. We have weekly coaching calls where we go deep on how to use Claude's dreaming and multi-agent setup to grow your business without you having to be there. Daily tutorials, walking you through every new Claude feature as it ships. 30 day roadmap to set up your first self-improving Claude agent for client work, lead generation, content support, you name it.

It's in there right now. A lot of them already running Claude agents in their day-to-day. Prompt library built around Claude workflows and a member map so you can connect with other Claude users near you. Link in the description or go to AIprofitboardroom.com.

Okay, back to it. The second feature is called outcomes. This one is just as important as dreaming, even though the name is boring. Here's the simple version.

With outcomes, you write a checklist of what a good result looks like. Then a separate Claude agent grades the work against that checklist. If the work doesn't meet the bar, the grader says exactly what's wrong and the main agent tries again. It keeps trying until it gets it right.

Think about that for a second. The AI is checking its own homework and it won't hand it in until it passes. Why does this matter? Because the grader runs in its own fresh context.

It doesn't see how the work was made. It only sees the final result. So it's not biased by the agent's own thinking. Here's a real-world example.

WiseDocs is a medical document review company. They built a quality check agent using outcomes. WiseDocs cut its document review time by 50% using outcomes half the time. Quality bar.

The third upgrade is multi-agent orchestration. This is where Claude becomes a team instead of one worker. Here's how it works. You have one lead agent.

The lead agent breaks a big job into smaller pieces. Then it hands each piece off to a specialist agent. Each specialist has its own model, its own prompt and its own tools. They all work at the same time.

Share files. They share notes. Lead agent checks in on the mid-job because everything is saved. So instead of one Claude doing everything in a slow line, you get a team working in parallel.

The fourth feature is webhooks. This sounds boring but it's actually huge. A webhook is just a way for one app to ping another app when something happens. With webhooks now built in, you can kick off a Claude agent, walk away and get a notification in Slack, in your email or in your own app the second it finishes the job.

Why does this matter? Because before you had to sit there and wait or you had to check back. Now Claude runs in the background and tells you when it's done. Now zoom out.

Here's what all four features look like put together. Multi-agent orchestration breaks the job up. Each agent gets a piece. Outcomes grade the work and force a rewrite if it's not good enough.

Memory lets each agent learn during the job. Dreaming reviews all that work overnight and updates the memory for next time. Webhooks tell you when the whole thing is done. That's an agent team that works in parallel, checks itself, learns and reports back.

Without you in the room, the Harvey result is what to remember. Six times more tasks finished. Model. Just because the agents stop repeating their old mistakes.

Now I want to address something. Some of you are watching this and thinking, this sounds like science fiction and I'll never use it. Get that. I really do.

Here's the thing. A year ago, AI agents would forget what you told them five minutes earlier. Now they review their own past work overnight and come back smarter. The tools are getting easier to set up, not harder.

You don't need to be a coder to put this together. You need a clear job, a checklist for what good looks like and a starting prompt. That's it. And the businesses using this stuff right now are getting results in months that used to take years.

Half the document review time at WiseDocs. Six times the completion rate at Harvey. Netflix processing logs they could never get through manually. If you're a small business owner, a marketer, a coach, a service provider, the question is not, will this affect me?

It's, when will I learn to use it? Here's a simple way to think about where this goes next. Right now, most people use Claude like a smart assistant. You ask, it answers.

This update, you can use Claude like a team. You give it a goal. The team runs in the background. They check each other's work.

They get smarter every week. They ping you when it's done. That's the shift. From one chat at a time to a team of agents running your repeat tasks while you sleep.

If you already use Claude, look at the managed agents docs. Outcomes is the easiest feature to try first. Pick one task you do every week. Write a rubric, run it through outcomes.

See the difference. If you're running a business, pick one process that eats up hours every week. Customer replies, weekly reports, content, lead follow-up. That's your first agent.

And if you want help setting all of this up, this is exactly what we do inside the AI Profit Boardroom. Four weekly coaching calls focused on building Claude agents that actually save you time and bring in more clients. Daily step-by-step tutorials on how to use the new Claude managed agents features as they roll out. 30-day Claude roadmap that takes you from zero to your first running agent.

Playbooks for using dreaming, outcomes, and multi-agent setups in real businesses. Members already inside, many of them building Claude agents for their own client work. A prompt library built around Claude. And a member map so you can connect with other Claude users near you.

Someone's always online to help. Link in the description or go to aiprofitboardroom.com. And if you want a free spot to start, join the AI Success Lab. It's our free community.

You'll get the full notes from this video plus over 100 AI use cases like this one. Around 67,000 members in there crushing it with AI. Links are in the comments and description. As thought, the line that sticks with me from the Anthropic announcement is the simplest one.

Memory lets each agent capture what it learns. Dreaming refines that memory between sessions. That's it. That's the breakthrough.

For years, AI was a goldfish. Smart in the moment. Useless an hour later. Now it remembers.

Now it improves. Now it works while you sleep. The people who learn to use this in the next 90 days are going to look very different from the people who don't in a year. That's the real story.

See you in the next one.