New Claude Updates DESTROY OpenClaw?

New Clawed Updates Destroy OpenClaw. Here's the thing nobody saw coming. Four months ago, people were buying Mac minis just to run an AI agent on their phone. Not a chat assistant, not a search tool, an actual AI that would sit on your phone, take tasks whilst you sleep, write code and send emails and manage files, doing it on its own whilst you're at dinner on the plane or living your life.

The product was called OpenClaw, went viral, Mac minis literally sold out across Asia. People were spending $500, $800, sometimes even over $1,000 for dedicated software and hardware to run this thing. And now Anthropic just built the same thing natively into Clawed. A developer named Matthew Berman put it perfectly.

He said they've built OpenClaw. And another developer, BentoBoy, said something even more wild. He posted Clawed just killed OpenClaw with this update. You no longer need to buy a magnet.

I say this as someone who owns one. That's the story we're talking about today. Clawed's biggest week, multiple major updates, and the one that's getting everyone talking is called Clawed Code Channels. But that's not all.

Because whilst Channels was dropping, Anthropic was also rolling out massive upgrades to their models, their memory features, their Excel and PowerPoint tools, their voice mode in Clawed Code and more. So let's go through it all. And I'm going to tell you why what's happening with Clawed this week is not just another product update. It's a signal, a very loud signal, about what's coming for anyone who works for a living.

Let's start with the big one. So the OpenClaw killer. You need to understand what OpenClaw was, but I'll keep it short because it's not the main event. OpenClaw was created by an Austrian developer named Peter Steinberger back in November 2024.

He created this project over a weekend. It's called OpenClaw. And it absolutely blew up. It became one of the biggest open source projects of all time.

It could work from your phone. You could give it tasks. It was super autonomous. And the problem was, however, the product was incredible in concept, but it was messy.

It was technical. You needed to know what you were doing. It had security holes. And recently, Anthropic actually blocked OpenClaw users from using their Clawed subscriptions to power it, which caused a lot of frustration.

So people waiting and wondering, is it going to be a proper, safe, easy version of this? And then on March the 20th, 2026, this week, Anthropic answered. Anthropic announced Clawed Code channel. It's a way to hook up its powerful Clawed AI to a user's Discord or Telegram messaging app, letting them message Clawed directly whenever they want, whilst on the go and instructed to do work for them.

This isn't just a new AI, it's a new fundamental shift in how developers interact with AI agents, moving from a synchronous ask and wait model to an asynchronous autonomous partnership. Now, let me translate that from tech language to real language. Before this, using Clawed was like having a phone call. You called, you waited, you answered, you talked, the call ended.

You couldn't multitask. You couldn't step away. You had to be there at your computer waiting for Clawed to finish. Now, it's like having an employee with a phone.

You text them a task, they go do it. When they're done, they text you back. You don't sit and watch. You live your life and that work still gets done.

That's a big shift here. Now, a channel is an MCP server that pushes events into your running Clawed Code session, so Clawed can react to things that happen whilst you're not at the terminal. Channels could be two-way, so Clawed reads the event and replies back through the same channel, like a chat bridge. MCP actually stands for Model Context Protocol, so Anthropic invented it back in 2024.

Think of it like a universal plug, like how every country used to have different power sockets, and then they invented a universal adapter. MCP is a universal adapter for AI, and it lets Clawed plug into basically anything. In the new channels architecture, an MCP server acts as a two-way bridge, so when a developer starts a Clawed Code session with the channel's flag, they aren't just opening up a chat, they are spinning up a polling service. Using the BUN runtime, known for its extreme speed in implementing JavaScript, Clawed Code monitors specific plugins, and when a message arrives, it's injected directly into the active session.

Clawed can then use its internal tools to implement code, run tests, or fix bugs and reply back to the external platform. Here's what that looks like in practice. So you open Clawed Code on your computer, you type one command, and Clawed Code starts running. You walk away.

Later, from your phone, on Telegram, you send a message. Check the build logs, and fix any errors you ask. And Clawed picks up that message, it goes to your actual files on your computer, it reads the logs, it identifies the problem, and then it fixes it. And then, after all of that, it sends you a message back on Telegram, found three errors, fixed all of them, here's what I've changed.

You didn't touch your computer. Clawed Code did the work whilst you were gone. And that's it, that's the product, and that's why people are calling it the Open Claw Killer. The security model is completely different from Open Claw.

Open Claw is a webhook-based, which means sometimes operating and opening inbound ports on your network. Clawed Code channels uses outbound polling only, so connections only go from your machine to the outside. There's no inbound ports ever opened. On top of that, there's an allow list-based sender authentication, so only registered users can send messages, which just makes it more secure.

And this is the thing that matters most for businesses, because Open Claw did have serious security concerns. IT departments and business owners get nervous around Open Claw, right? Clawed Code channels is built with security-first architecture, so there's no open ports, only you can message it. And that changes everything for teams.

But here's what you can't do yet. And I want to be honest, because I keep saying on this podcast, hype without substance is just noise. And only two platforms are currently supported. First of all, it's only Telegram, and it's only Discord that you can use with this.

The biggest gap is no Slack. If you want to use this for work, Slack integration is practically a must right now. Compared to Open Claw supporting over 50 platforms, it's pretty bare with Clawed. There's no WhatsApp, no iMessage, no Microsoft Teams.

If you run a business on Slack, which a lot of businesses do, you're still waiting. The other frustrating part is when the session closes, it's over. If you send a message on Telegram whilst the Clawed Code session is down, that message is lost, my friend. So you have to keep Clawed Code running in the background.

It's not truly always on the way that Open Claw was. You have to run it in persistent terminal or background process. Anthropic has said the channel's flag syntax and notification protocol may change based on feedback. And there's a plan for API changes in Q2 to Q3 2026 if you're building production workflows around this.

So this is also a research preview. And that means it's real, it works, but it's not 100% finished yet. And there are going to be changes. So is Open Claw actually dead?

Well, Open Claw still supports platforms. Channels doesn't, right? WhatsApp, iMessage, Slack, Signal, Microsoft Teams. Channels only has Telegram and Discord.

If your business runs on Slack, you're still in the DIY camp for a few more months. And the honest answer is Clawed Code channels is a massive step forward. But right now, Open Claw still wins on platform coverage. The question is how fast Anthropic moves.

And based on everything else they've released this week, they're moving very fast. Now, let's talk about the models. Because whilst all of this was happening with Clawed Code, Anthropic also shipped two brand new AR models. And these weren't just small upgrades, OK?

So basically, what matters most with both of these upgrades is not the model. The model didn't change here. They were still using Opus 4.6 and Sonic 4.6, but now they support a 1 million token context window. That's roughly 750,000 words in a single conversation, which is a million tokens.

Now, let me help you picture that. The average business report is maybe like 5,000 words. A million tokens is 150 of those reports all at once in one conversation. With Clawed reading every single word, remembering every single detail and connecting the dots across all of it.

A year ago, the best AR models could handle maybe 8,000 words at a time. Then it was 32,000, then 200,000, now we're at a million. That's a 125 increase in roughly 18 months. Think about what that means for your business.

You can dump your entire product documentation into one conversation and ask, find every place where we contradict ourselves. You can upload a year's worth of customer support tickets and ask, what are the top five problems our customers keep having? You can share your entire code base, every file, every function, and ask Clawed to find the bugs. This isn't AI helps you write emails territory anymore.

This is an AI that can read and understand your entire company's knowledge base in one single shot. Also, a common complaint about AR models was context rot, where basically performance degrades as conversations get longer. Opus 4.6 performs markedly better on this. On a needle in a haystack test at 1 million tokens, Opus 4.6 scored 76%, whilst the previous Sonnet 4.5 scored just 18%.

It's no small improvement, right? That's a four times better result. The previous model was basically lost in the information. After a certain point, Opus 4.6 stays sharp the whole way through.

Here's where it gets interesting as well, because on GDPVALAA, an evaluation of performance in economically valuable knowledge work, tasks in finance, legal, and other domains, Opus 4.6 outperforms the industry's next best model. So it's a more powerful model, it's got more tokens, it's more useful, and it's outperforming everyone as well. So these models combined with the power of Clawed's new updates are really powerful as well. Now, here's something else that might affect more people than anything else I've mentioned, which is Clawed is now available as an add-in in PowerPoint.

And Clawed for Excel has been updated to use Opus 4.6 as well, and supports native Excel operations, including pivot table editing and conditional formatting. These are things that you just don't get with OpenClaw, but now you can do it with Clawed, and it's just way more powerful. In fact, Clawed, with Box's CEO, they actually tested Clawed Sonnet 4.6 on real enterprise work. So this is the company Box.

And we've generated reports, doing due diligence and expert analysis across multiple documents. They actually saw a 15% jump in performance and accuracy. Now, 15% might sound small, but across thousands of tasks, across entire teams, 15% is enormous. There's a number of updates that barely got any attention at all, but I think it's important, and that is the memory update.

So basically what happens now is with Clawed Projects, they basically have this optionality now, where you can create projects with context and new files and new memory for Clawed inside Clawed Cowork. And it's actually called Clawed Projects, this feature. Basically what it allows you to do is create multiple different AI agents. So, for example, inside OpenClaw, you've got one long stream of conversation.

But with Clawed Projects, you can basically have different conversations with an autonomous AI agent inside Clawed Cowork. It's running for you 24-7, so you can still connect to it through your phone, and you can still give it scheduled tasks like you would with OpenClaw. But the difference is you can have many different conversations with many different projects, many different ways of controlling context across, could be clients or content or different brands and that sort of thing. But the main thing to note is it's much easier to create separate projects inside Clawed Cowork with this new projects feature versus something like OpenClaw as well.

And that makes it a lot easier. And bear in mind, this stuff is way easier to set up than OpenClaw as well. OpenClaw requires a lot of technical setup inside the terminal, if that makes sense. Whereas with Clawed Cowork, you're not debugging issues all the time, the same with Clawed Code as well.

Now, there's also another update, and this one is early, but it's worth noting. And the most anticipated feature of Clawed Code in March 2026 is voice mode, which is rolling out progressively to all users. It activates with the forward slash voice command, and it works for a push-to-talk mechanism. So you can just hold down the spacebar to speak, release to send.

And it's not an invasive, like always-on listening system, but a controlled and precise approach. So you can hold the spacebar, you talk, you say, fix the logging bug we were working on yesterday, Clawed hears you, and then Clawed works. And transcription has been optimized for technical teams and also repository names and terms as well. So voice mode is currently available to approximately 5% of users with progressive rollout over the coming weeks.

And this is still early, but think about the direction, right? Clawed Code started as something you just typed into. Then it got mobile access through dispatch. Then it got messaging through channels.

Now it's getting voice, and the interface is expanded. And the ways you can interact with your AI are multiplying too, so every new interface makes it more accessible to more people. A non-technical person can now talk to their AI coder and say, hey, I want a website to have a contact form that sends me an email when someone fills it out. And they say out loud, Clawed Code goes off and builds it.

And that's not really a developer workflow anymore. That is anyone's workflow that works for anyone, right? Now, here's the thing about everything we've just covered. Six months ago, if you wanted an AI agent that works whilst you slept and texted you updates, you needed something like maybe OpenClaw, but you know, OpenClaw was not out back then, right?

So it'd be something more like an autonomous agent that you would set up separately, locally. Maybe something like Agent Zero, for example. You needed technical knowledge to do that. You needed dedicated hardware.

You needed to be comfortable with security risks. And most people couldn't do it. But now, Anthropic ships it natively, secure, built-in. You need a Clawed subscription and a Telegram account.

Six months ago, if you wanted AI that could read your entire company's documents and full code base and all your data at once, well, that wasn't really possible. Now, a million token context, you know, is easily doable with something like Clawed Opus and Sonic 4.6. So you can see the progressions here. Again, six months ago, if you wanted an AI that actually remembered you between sessions, understood your context, built on what you talked about before, that was the max plan and available for paid users only.

Now, it's free for everyone. And I keep saying this, and we're going to keep saying it. The gap between what's possible for a small business owner or a solo operator with access to these tools and what used to require a full team is getting smaller every month, not every year, every month. And the question is not whether AI is changing work or it already has.

The question is whether you're on the right side of that change or on the wrong side. And if you're watching and thinking, OK, this is a lot of updates, where do I even start? I want to tell you about something I built specifically for this problem. It's called the AI Profit Boardroom.

It's where over 2,600 business owners are already using AI automation to do more, earn more and work less. Here's what's inside. Four live coaching calls every week where we go through exactly how to apply the latest AI tools to your business. New tutorials every single day, not slide decks, actual step-by-step videos showing you exactly how to set things up and get them working.

30-day roadmaps, so you're not starting from absolute zero. You get a clear path and prompts for basically every situation you can think of, whether that's marketing, sales, customer service, operations or all of it. And a community map where you can actually find and connect with other business owners near you because there's always someone online 24-7 to help you when you're stuck. We're not talking theory, we're talking the actual practical stuff that turns these AI updates into income.

The people in the boardroom don't just watch what's happening with Claude, they're already using it. It's the kind of environment you want to be in. So if that sounds good to you, check it out at theaiprofitboardroom.com or go to the link in the comments description. So back to it.

Here's something else from Claude Code this month that didn't get as much attention as Channels, but it deserves it. So the forward slash loop command introduced in March 2026 transforms Claude Code into a recurring monitoring system. So you define an interval and a prompt and Claude will implement it automatically. For example, forward slash loop, five min, check the deploy.

And this checks the deployment of your website or app every five minutes. And that could be the same, for example, it could like check the AI automation news every single hour, for example. And it works as a lightweight session level cron job. Now a cron job is basically a timer that runs tasks automatically on a schedule.

That used to require programming knowledge. You have to write scripts, configure things, set it up on a server. Most people couldn't do it. Now you can type forward slash loop, five m, check the deploy.

And Claude Code watches your deployment every five minutes and flags any issues whilst you're sleeping. Or you could set up, for example, forward slash loop, one hour, check for new customer support tickets and draft responses. Claude Code running in the background every hour, looking at incoming support tickets, drafting replies and logging them for your review. Or you could say, for example, forward slash loop, one day, pull the sales numbers from today and send me a summary.

That's actually really useful, right? Every day automated without you touching it. This is what business automation used to cost thousands of dollars to build. Custom software, dedicated developers, monthly maintenance.

Now it's a single command in a chat interface. The loop stays active as long as the session is open and can be disabled with an environment variable. So if you combine this with channels where Claude texts you when it finds something, you have an AI that's monitoring your business, texting you updates, taking actions, or whilst you live your life. There's also something called the effort command.

And this is something where you can basically allow yourself to set the model's effort level simplified to three levels. So to temporarily activate maximum effort, you can use the ultra think keyword in your prompt. And this is practical because you're doing a quick thing, for example, like formatting a document or writing a short email or looking for something up. And you don't need Claude at full power for that, right?

You can use low power, low effort, and it's a faster way of using fewer tokens. And, you know, if you're doing something complex, for example, like analyzing a contract or building a multi-part system or making a critical business decision, well, then you can type ultra think before you prompt. And what happens is Claude actually shifts into maximum reason mode. It's like having a dial that goes from quick and cheap to deep and thorough.

Most AI tools don't give you that dial. Claude now does. So you can see all of these updates coming along right here. And there's a business story underneath all the technical updates as well.

So enterprise plans, which were previously only available to customers working directly with Anthropic's sales team, can now be purchased directly on the website with no sales conversation required. It's self-serve, right? Now, enterprise plans include access to Claude, Claude Code, and Co-Work. And Claude Code access is now included with every team plan standard C.

So Anthropic is removing every friction point between a business and full AI deployment. You used to need a sales call to get enterprise. Now you just click and buy. You used to need a separate Claude Code subscription on your team plan.

Now it's included. They want to be in every single company, not just tech companies, not just the startups, every single company. And the third party data is backing this up. According to multiple reports from March 2026, 67% of Fortune 500 companies now have at least one AI agent in production, up from 34% in 2025.

That's nearly double in one single year. The average cost reduction reported by companies using AI agents for customer support is now 35%. So that's a 35% reduction in customer support. It's not a small number, right?

In a business with 10 support staff, for example, costing maybe $50,000 each, that's $175,000 a year back in the company's pocket. And the venture money is following too. $4.2 billion in venture funding went to AI agent startups in just the first quarter of 2026 alone. $4.2 billion in three months.

Not because investors are dreaming, but because businesses are buying. So I hear this a lot. A lot of people say, Julian, I'm not a developer. This all sounds amazing, but I can't code.

How does any of this help me? Let me be direct. Everything we've talked about today, the channels, the million token context window, the voice mode, the memory, the Excel and PowerPoint integrations, none of it requires you to write a single line of code. You can message Claude and telegram, tell it research the top five competitors in my market and summarize their pricing.

It goes off and it actually does it. You don't need to touch a computer. You can drop your entire sales pipeline into Claude and ask, where are we losing customers and what should I change? And with a million token context window, that means it can read everything.

You can ask Claude in Excel to analyze your last 12 months of revenue and flag every month that underperformed, and then it will actually tell you why. You didn't write the formula. Claude did. Now, you can also set up forward slash loop to check your email every hour and draft responses to anything marked urgent whilst you're on a call, whilst you're at the gym, whilst you're with your family.

So this is not the future. This is available right now, and it's available today for just $20 a month. And the people who figure out how to use this stuff, not code with it, just use it, are going to run circles around the people who haven't yet. And I get it.

I really do. Like some of you are watching this and thinking, I've heard AI is going to change everything for two years now. I've seen the demos. I've tried the tools.

It's not quite as magical in real life. And you're not wrong. You know, a lot of tools have over-promised and under-delivered before, right? The demos are always perfect.

Real life is very messy. Files don't format right, for example. Claude code can make errors. Voice mode is only in 5% rollout.

Channels only work if your sessions are open. And OpenClaude had a ton of security problems. There are gaps everywhere that I look, and I'm not going to pretend those gaps don't exist. They're real.

But here's the thing I need you to notice. The direction, right? On a Neo and a Haystack test at 1 million tokens, Opus 4.6 scored 76%. The previous model, Sonic 4.5, scored just 18% on the same test.

So it's not iterating anymore. It's not gradually improving. It's going at like a 4x better speed on specific critical capabilities in one generation of models. Now, everything is really improving with this.

You know, the gaps are real, but the gaps are also getting smaller at a rate that's hard to explain without words like phase change and step function. And every time Anthropic ships something that closes a gap, like, for example, channels closing the OpenClaude gap, or memory closing the ItForgetsMe gap, or the 1 million token context window closing the ItCan'tReadMyWholeDocument gap, the tools get a lot more powerful for a lot more people. You don't have to believe that AI will solve all your problems. You just have to notice that the tool is getting better than most people expected.

And that has implications for how you work. So what's coming next? Well, based on everything Anthropic shipped this week, here's what I think is on the roadmap. Channels will have more platforms, probably Slack first.

They already have it in their main product. WhatsApp will come. The research preview tag will come off and it will just be a feature. Voice mode will go from 5% rollout to everyone.

And when it does, the interface to your AI assistant becomes talk to it. That's it. No typing, no learning, no prompt engineering. Just talk to it like you talk to a smart colleague.

The Excel and PowerPoint integrations will get deeper. Not just Claude can help in Excel, but Claude can proactively notice things in your spreadsheet, flagging risks, suggesting analyses you didn't think to ask for. And the models will keep improving too. The hybrid approach has become standard in 2026.

Companies use Sonic 4.6 for like 80 to 90% of requests as a primary model and only use Opus 4.6 for the most complex requests, right? Now, this allows maintaining Opus quality in high stakes moments while saving 60 to 80% on API costs. And the smartest companies in the world have figured out that playbook, right? Fast models for routine tasks and smart models for critical decisions.

So what should you do this week? Let me give you something concrete. If you're a business owner and you don't have Claude set up for your work yet, go to Claude AI, start with a free tier, ask it to help with whatever you're going to spend the next hour on and see what happens, right? If you're already using Claude and you haven't tried Claude code, this week is a week to start.

The channel's feature is in research preview, but it's working. Set up the telegram integration. Send yourself a task from your phone. Watch it come back.

That experience alone will shift how you think about what's possible. If you're running a team, take a serious look at the team plan. Right now, the Claude code is included. The forward slash loop command alone set up for monitoring your key business metrics as well is worth the investment.

And if you're a developer, upgrade to Opus 4.6 if you haven't already, if you're doing anything complex, I would say. So the other thing I would say here is that if you're feeling behind, that's actually the right emotion to be paying attention to. Because you shouldn't panic here, but the tools are moving and the people using them are pulling ahead. The gap between AI user and not an AI user is getting wider every week, not narrower.

And it's technically hard, but also it's moving so fast, you know, and the old way of doing things is just changing all the time. So the human story underneath all of this is that OpenClaude got released. Now Claude is catching up. And if you look at the things that Anthropic is doing, they've been taken to a whole new level by what OpenClaude released.

And it's really created this competition where Anthropic needs to improve better than ever before. So thanks so much for watching. If you haven't already, check out the AI Profit Boarding, link in the comments description, or go to the AIProfitBoarding.com. And if you want actual tutorials, 30-day roadmaps, if you want step-by-step guides on this stuff, you can check it out right there.

And you can personally connect with me too. Thanks for watching. Appreciate it. Bye bye.