Skip to content

Podcast

The OpenClaw-ification of AI

The AI Daily Brief: Artificial Intelligence News and Analysis

Source ↗ ← All highlights
  • Anthropic Versus The Pentagon Over Military Use
    • The Anthropic–Pentagon standoff raises whether labs or government set rules for military AI use.
    • Pentagon pressured Anthropic to accept ‘all lawful use’ terms and threatened supply-chain blacklist or Defense Production Act invocation. Transcript: Nathaniel Whittemore We kick off today with a topic that could easily be a main episode. The Pentagon has handed Anthropic an ultimatum in what could be the most crucial battle for AI safety to date. Now, heading into this week, we had building reports that Anthropic’s contract with the Department of Defense, or the DOW, on who you ask, was in jeopardy over AI readiness. Anthropic was insisting that their technology should not be used to power autonomous weaponry or for the surveillance of Americans. At certain points in the reporting, it seemed like Anthropic had contracted on the basis that their technology would not be used to assist in any lethal exchange. The Pentagon, on the other hand, was insisting that tech companies should not have input on how technology is used in military operations. They have been pushing for each AI lab to provide their models with terms that allow, quote, all lawful use, basically establishing the position that only U.S. Law rather than tech company policy should be the limiting factor in the conduct of war. The conflict had, of course, erupted after reports that Anthropics AI models had been used during the raid that captured Venezuelan President Nicolas Maduro. The reporting has been unclear to what extent AI was used, but it appeared that Anthropics was furious that their technology was used in what was ultimately a lethal operation. In that context, Anthropics executives were summoned to a meeting at the Pentagon on Tuesday to settle the dispute, and things did not go well. Defense Secretary Pete Hegseth personally gave Anthropics CEO Dario Amadei a Friday deadline to agree to their terms or face a government blacklist. And this is not just they would drop their contract. The threat was that Anthropic would be designated a supply chain risk and barred for use among military contractors as well. That designation had previously only been applied to foreign companies such as Huawei. New reporting substantially fleshes out the conflicting views. Sources familiar with Anthropic’s position said they are not confident that their AI models are reliable enough to operate weapons without a human in the loop. They also note that there are no current laws dealing with how AI can be used in domestic surveillance. Hegsitz reportedly told Anthropic that they have until 501 on Friday to get on board or not. He also warned Anthropic that he would ensure that, quote, the Defense Production Act is invoked on Anthropic, them to be used by the Pentagon regardless of whether they want or not. Most are taking this to mean the Pentagon would compel Anthropic to provide a version of Claude without guardrails trained into the model. The Defense Production Act does allow the U.S. Government to dictate contract terms on goods and services deemed critical to national security. However, it has rarely been invoked against the will of individual companies and is more typically used to order entire sectors to prioritize the production of certain goods to fulfill Government orders. For example, it was invoked during the 2021 California wildfires to increase the production of fire hoses and during the pandemic to boost the supply of ventilators. Katie Sweeten, a former liaison for the Justice Department to the Department of Defense, noted that Hegg says threats are kind of contradictory, telling CNN, I would assume we don’t Want to utilize the technology that is a supply chain risk, right? So I don’t know how you square that. What it sounds like is that the supply chain risk may not be a legitimate claim, but more punitive because they’re not acquiescing. Despite being tense, sources said the meeting was cordial and that there were no raised voices. Hegseth even praised Anthropics models, which has been one of the undercurrents of the story. The Pentagon seems to believe that Anthropics technology is so far ahead of the competition that it can’t be replaced by another supplier. The Pentagon has standing contracts with OpenAI, Google, and XAI. However, XAI is the only one to agree to all lawful uses standard across both classified and unclassified settings. Claude is the only model currently cleared for use in classified settings, according to Axios. Sources familiar with Anthropics thinking said that they have no plans to buckle to the Pentagon’s demands. Their official statement was polite but firm, commenting, We continued good-faith conversations about our usage policy to ensure Anthropic can continue to support the government’s National security mission in line with what our models can reliably and responsibly do. Meanwhile, a Pentagon spokesperson said, This has nothing to do with surveillance and autonomous weapons being used. You can’t lead tactical ops by exception. As of Wednesday, Axios reports that the Pentagon is taking preparatory steps towards labeling Anthropic a supply chain risk. They’ve reached out to major contractors, including Boeing and Lockheed Martin, to determine if they have any critical dependencies on Anthropics technologies. Now, the reason that this is so important, and like I said, could easily be a main episode, and who knows, might need to be at some point, is that there’s much more than Anthropics’s $200 Million contract at stake. The dispute goes right to the heart of the debates around AI safety that have been raging on the internet for the past decade. We are no longer talking about hypotheticals, but a very practical question of whether AI labs or the US government should have the final say on how military AI is deployed. Semaphore writer Reed Albergade commented, the most likely conclusion of this saga is that the Pentagon forces Anthropic to comply using the Defense Production Act. In some ways, this means both sides get what they want. The Pentagon no longer has restrictions on how it can use Claude, and Anthropic gets to save face even if it ends up complying with the Pentagon’s demands. For some, this would be a very unsatisfying conclusion. Lawfare Media wrote, The terms governing how the military uses the most transformative technology of the century are being set through bilateral haggling between a defense secretary And a startup CEO with no democratic input and no durable constraints. Congress should be setting these rules and it should do so in a hurry. Obviously, this story continues to evolve and is something we will watch. (Time 0:01:09)
  • NVIDIA Earnings Confirm AI Compute Boom
    • NVIDIA’s earnings show an ongoing AI hardware boom with massive revenue growth and new data-center guarantees.
    • Jensen Huang highlighted exponential token demand and NVIDIA pledged $3.5B in guarantees to help data center financing. Transcript: Nathaniel Whittemore Speaking of Wall Street, we turn finally today to NVIDIA, who absolutely crushed earnings, delivering another record quarter for revenue. Wednesday night’s earnings left no room for interpretation. The AI chip boom is still in full swing. NVIDIA’s revenue was up 73% annualized to reach $68.1 billion, exceeding expectations by a massive 11 percentage points. They also guided even higher growth for the current quarter, projecting a 77% annualized revenue gain. Center business grew even faster than top-line revenue, coming in at 75% growth. NVIDIA also noted that they’re still assuming zero revenue from China in their forecasts, so revenue could see another big boost if exports start flowing. CEO Jensen Huang remarked that AI demand is going from strength to strength, commenting, The demand for tokens in the world has gone completely exponential. I think we’re all seeing that, to the point where even our six-year GPUs in the cloud are completely consumed and the pricing is going up. Still, NVIDIA also disclosed massive new backstops for the AI build-out. They’ve now provided $3.5 billion worth of guarantees to companies developing data centers, four times the amount they disclosed in the previous quarter. These guarantees allow smaller neocloud companies to access debt on more favorable terms. However, if AI demand falters, NVIDIA could be required to take over leases and the operation of data centers in a struggling market. NVIDIA did not disclose which companies they gave these new guarantees to. AMD recently gave a similar guarantee to Crusoe, pledging $300 million, suggesting these arrangements are becoming more common. While the backstop adds marginal pressure on NVIDIA’s balance sheet, ultimately these deals are pretty modest compared to the booming profits. Jensen also weighed in on the recent drawdown in software stocks, arguing that the markets got it wrong. He restated his view that existing software firms will build agentic features on top of their products, but their value as systems of record and databases will remain. He said, use today, whether it’s Cadence or Synopsys or ServiceNow or SAP, these tools exist for a fundamentally good reason. These agentic AI will be intelligent software that uses these tools on our behalf and help us be more productive. Nobody’s going to service better than ServiceNow, and they’re going to come up with agents that are really fine-tuned and optimized for the work that uses the tools that they have. In the end, we need the tools to finish their work and put the information back in a way we can understand. Now, despite the rosy outlook, the market response was a little underwhelming. NVIDIA stock gained 3% in after-hours trading once earnings were announced, but fell below its closing price later in the evening. Ultimately, right now, I just think that markets are completely unmoored. In general, but especially when it comes to AI. And there’s basically no better evidence for that than its response to this NVIDIA announcement. (Time 0:08:39)
  • Claude Code Brings Phone To Machine Coding
    • Anthropic’s Claude Code remote control brought phone-to-machine coding into mainstream products.
    • The feature mirrors OpenClaw’s mobility but differs by offering a dedicated terminal bridge aware of local dev context like repos and node modules. Transcript: Nathaniel Whittemore Right at the top, we have a Claude Code’s new remote control feature. It is exactly what it sounds like and brings a thing that people have wanted forever and that numerous startups have built around to the mainstream, which is the ability to code with Claude Code from your phone. On Tuesday, Anthropic posted, new in Claude Code remote control. Kick off a task in your terminal and pick it up from your phone while you take a walk or join a meeting. On Tuesday, Anthropic posted, Some of the first reactions were to joke about the social life implications of this. Cremio wrote, And yet for most, the very clear comparison was in fact OpenClaw. AppClub’s MetaPreston writes, this is the only reason I wanted ClawdBot. Investor and content creator Sarah Dietschy writes, okay, I guess I don’t need to try ClawdBot anymore. Guy L. Breton calls it OpenClaw for grownups. And Allie K. Miller says, this is the feature I’ve been waiting for. I work from my phone for hours every day. Remote control is about to turn me into a Claude Code maniac. So one of the big things that people love about OpenClaw is the ability to interact with your agents via Telegram. And that’s not because Telegram is necessarily the best experience when you’re sitting in front of a computer. It’s because using Telegram or WhatsApp or whatever chat app you’ve connected it to makes it so that you can interact with your agents on the go. This was certainly one of the first things that attracted me to OpenClaw, was the ability to have an on-the coding agent so I could keep working on my projects when I wasn’t sitting in Front of the laptop. Lots of other people clearly felt similarly. Nate Herc writes, Claude Code just added what everyone wanted. With remote control, you essentially have Claude Code in your pocket so that you can check in on progress and continue to assign work as you’re out and about. Some even went so far as to argue that the use cases for OpenClaw in this new remote control feature might actually end always-on butler. It’s better for managing your calendar, booking flights, or texting your mom. Why the remote control is different. First is native versus messaging. OpenClaw talks to you through Telegram or WhatsApp. CloudCode remote control gives you a dedicated web or mobile terminal bridge to your local hardware. Deep context. CloudCode is aware of your specific dev environment. It knows your repo, your node modules, and your local CLI tools. Session versus Butler. You boot up Cloud Code for a work session. OpenClaw stays running 24-7 on your machine or VPS. From a pragmatic usage perspective, use Cloud Code to start a heavy refactor, head to a meeting, and check the diffs on your phone. Use OpenClaw to stay on top of your notifications and general digital existence while you’re away from your keyboard. If you want to code from your pocket with surgical precision, get Cloud Code. If you want a 24-7 digital butler, keep OpenClaw. Now, what I don’t totally agree with here is the idea that OpenClaw is only for these sort of personal use cases like managing your calendar or booking flights. In fact, as I mentioned in my video from a couple weeks ago about my OpenClaw setup, my most useful agents are my research agents that are running 24-7 and updating information that’s Both useful for this show as well as for some of our ecosystem projects like AIDB Intelligence. But I think two things. First of all, the points about why this is going to be so valuable to people from a coding perspective, because it plugs into the main system that people are already using in the form of Clawed code, is I think true. One of the reasons I believe that I found myself using my remote coder open claw agent a little bit less is the context gap between it and where I’m doing most of my other projects. The second thing that I think is relevant about InterNoun’s post is that even if one doesn’t agree with the idea of how OpenClaw is characterized here, the actual underlying point is That we’re in a broader paradigm shift of agentification and that there are going to be different tools and different platforms that have similar types of setups and similar types Of interactive modes, but with very different use cases. Whatever the use cases end up being, remote control hit like gangbusters. Claude Code PM Noah Zweben had to come back to Twitter the day after the announcement to apologize for the capacity issues that they had run into on their end and to ensure that a broader Rollout was coming. For some, remote control was just further evidence of how fast Anthropic is moving. Anonymous AI rumor account, I Rule the World, writes, Anthropic have now shipped OpenClaw. Dario and the gang are unfathomably ahead and their velocity is insane. It’s a fair observation that they may have reached escape velocity, crushing every benchmark, eating enterprise, Department of War begging. One of the labs made a huge breakthrough. (Time 0:15:14)
  • Using OpenClaw Agents To Work While Mobile
    • Nathaniel describes using an OpenClaw coding agent on his phone to continue projects while away from a laptop.
    • He found the agent especially useful for research agents that run 24/7 and feed show intelligence and ecosystem projects. Transcript: Nathaniel Whittemore Nate Herc writes, Claude Code just added what everyone wanted. With remote control, you essentially have Claude Code in your pocket so that you can check in on progress and continue to assign work as you’re out and about. Some even went so far as to argue that the use cases for OpenClaw in this new remote control feature might actually end always-on butler. It’s better for managing your calendar, booking flights, or texting your mom. Why the remote control is different. First is native versus messaging. OpenClaw talks to you through Telegram or WhatsApp. CloudCode remote control gives you a dedicated web or mobile terminal bridge to your local hardware. Deep context. CloudCode is aware of your specific dev environment. It knows your repo, your node modules, and your local CLI tools. Session versus Butler. You boot up Cloud Code for a work session. OpenClaw stays running 24-7 on your machine or VPS. From a pragmatic usage perspective, use Cloud Code to start a heavy refactor, head to a meeting, and check the diffs on your phone. Use OpenClaw to stay on top of your notifications and general digital existence while you’re away from your keyboard. If you want to code from your pocket with surgical precision, get Cloud Code. If you want a 24-7 digital butler, keep OpenClaw. Now, what I don’t totally agree with here is the idea that OpenClaw is only for these sort of personal use cases like managing your calendar or booking flights. In fact, as I mentioned in my video from a couple weeks ago about my OpenClaw setup, my most useful agents are my research agents that are running 24-7 and updating information that’s Both useful for this show as well as for some of our ecosystem projects like AIDB Intelligence. But I think two things. First of all, the points about why this is going to be so valuable to people from a coding perspective, because it plugs into the main system that people are already using in the form of Clawed code, is I think true. (Time 0:16:43)
  • Scheduled Tasks Turn Models Into Background Workers
    • Anthropic added Scheduled Tasks to Claude Cowork, turning chat models into background workers.
    • Scheduling and heartbeats let agents run recurring briefs, spreadsheet updates, and reminders without manual prompts. Transcript: Nathaniel Whittemore Just one day later, they returned to X to announce a new feature, this time for Cowork, called Scheduled Tasks. Once again, it was exactly what it sounds like. They wrote, Claude can now complete recurring tasks at specific times automatically. A morning brief, weekly spreadsheet updates, Friday team presentations. They also write, it gets better with plugins, which give co-work domain expertise across design, engineering, operations, and more. Also, we’re adding a new customized tab in your co-work sidebar. One place to manage your plugin skills and connectors. Now, going back to the idea of the clawification of AI, if you’ve listened to my episodes about OpenClaw, or if you’ve tried it out yourself, you’ll know that a big part of the magic is The way that it interacts with scheduled tasks. First, OpenClaw agents have something called a heartbeat, which is basically a reminder the time interval of which you can set, but which is every 30 minutes by default, where effectively The agent reminds itself to remember its mission and to go keep working if it perhaps has lapsed in the previous period of time since the last heartbeat. On top of that are cron jobs, where basically you can tell an agent to do a specific thing at a specific time. When you dig into OpenClaw, it turns out that a lot of what the magic is, is just being able to schedule tasks at certain times. (Time 0:19:57)
  • Scheduling Is A New Labor Primitive
    • Scheduled tasks are a labor primitive that replace many small SaaS products and junior roles.
    • Examples include morning briefs compiling Slack, weekly auto-generated spreadsheets, and timed contractor reminders. Transcript: Nathaniel Whittemore On top of that are cron jobs, where basically you can tell an agent to do a specific thing at a specific time. When you dig into OpenClaw, it turns out that a lot of what the magic is, is just being able to schedule tasks at certain times. Across my five project manager agents, they all have different schedules for the couple times a day when they check in with me either to remind me of all the to-do lists, or to go interact With external sources to find the state of different projects. And so this idea of scheduled tasks feels once again, not like Anthropic trying to compete with OpenClaw, but instead recognizing a new primitive in agentic AI. That didn’t stop people making the comparison, however. Content creator Ole Lemon shared the post and said OpenClaw for normies has landed. Akash Gupta gets that part of the paradigm shift is that AI that works continuously without you having to prompt it every time. He writes, scheduled tasks means Claude stopped being software you talk to and became software that works while you sleep. That’s a category change, not a feature update. Now he, along with lots of others, also pointed out that this is unlikely to make Wall Street, which was already nervous about every new announcement from Anthropic, any less scared About the implications for white-collar work. Later in the same post, he writes, A morning brief that compiles overnight Slack activity, email threads, and calendar changes before you wake up. A weekly spreadsheet that pulls data, runs formulas, and drops a formatted Excel file into your folder every Friday. Contractor reminders that fire at 3 p.m. Without anyone remembering to send them. Each of those tasks used to be a SaaS product, a Zapier workflow, or a junior employee’s morning routine. (Time 0:20:53)
  • Perplexity Positions The AI As The Computer
    • Perplexity launched Perplexity Computer to unify multimodal AI workflows and orchestrate multiple specialized models.
    • The product emphasizes model-routing (Opus), persistent memory, connectors, and running hundreds of active projects like a 2026 personal computer. Transcript: Nathaniel Whittemore On Wednesday, Perplexity announced Perplexity Computer, writing, Computer unifies every current AI capability into one system. It can research, design, code, deploy, and manage any project end-to Computer orchestrates models to run agents in parallel, leveraging Opus to match each task to the model best suited For it. Perplexity Computer, they write, is what a personal computer in 2026 should be. It’s personal to you, remembers your past work, and is secure by default. Hundreds of connectors, persistent memory, files, and web access all built on top of Perplexity infrastructure. Go from a single task to hundreds of active projects. Google Senior AI Project Manager Shibam Sabhu writes, Perplexity just launched their own open claw. Composable computers is all you need. Perplexity CEO Aravind Trinivas wrote, What has Perplexity been up to the last two months? We’ve silently been working on the next big thing, Perplexity Computer. In a blog post surrounding the announcement called the AI is the Computer, he basically made an argument that the differentiator for Perplexity’s version of OpenClaw was that it only Had access to Anthropics models, and hammered on the idea that with Perplexity you get 19 models available so that the right task can find the right model. He writes, AI models are becoming so capable that the products built around them have been a bottleneck for showing their true potential. The chat UI is good for answers, and agents are good for individual tasks. Meanwhile, the UI for entire workflows has always been the computer. AI is now firmly multimodal. It understands and generates many forms of data in a single coherent system. Jensen Huang and others have said that the future of AI must also be multimodal. They’re right. Specialized models must collaborate like a team. As AI replaces more of the function of the computer itself, the central activity of the computer will be massively multimodal. He writes, multimodal orchestration. (Time 0:24:03)
  • Product Companies Ship Built-In Persistent Agents
    • Notion, Airtable, and other product companies launched built-in custom agents that schedule automations for teams.
    • Notion’s agents run on triggers, schedules, and team workflows, marketed as ‘the AI team that never sleeps.’ Transcript: Nathaniel Whittemore Well, I’m holding OpenAI aside because they obviously got Peter Steinberger, the creator of OpenClaw, to come join them. But we’re also seeing it in the product companies. Also on Tuesday, Notion announced their custom agents features. And sure enough, it had a lot of familiar value propositions in the pitch. Notion calls custom agents the AI team that never sleeps. They’re autonomous, built for teams, and easy for anyone to build. Give them a job, set a trigger or schedule, and they’ll get it done around the clock. Most AI waits for you to ask. Custom agents just go. Route bugs to the right place, answer questions, update docs, draft weekly updates, and ping the right people. They’re multiplayer, model agnostic, and built for every team, not just the technical ones. Creator Andrew Warner writes, Notion just launched OpenClaw for regular people. Pros, no need to buy a Mac Mini. Very visual so anyone can use it. Runs automations based on time, Slack messages, Notion comments, etc. And your team doesn’t need to know terms like cron job, pseudo or CLI. The cons he pointed out is that you need to buy a seat for each person on your team or no one gets it, which when you’ve got a lot of people in Notion could be problematic. (Time 0:25:44)
  • Set Up OpenClaw To Learn Agent Primitives
    • Do set up OpenClaw yourself if you want a deep education in agentic primitives.
    • Nathaniel argues the DIY Claw experience teaches heartbeats, cron jobs, persistence, and orchestration better than abstracted products. Transcript: Nathaniel Whittemore Many will do that and it’s reasonable, but one of the best reasons to actually do the hard work of setting up open claw now is that it’s going to give you a better education in understanding These primitives than will the products where they’re abstracted away. Certainly most people will not take the time to do that, but if you’re listening to this show, I know you are not most people. (Time 0:27:52)