WEBVTT Kind: captions Language: en 00:00:01.500 --> 00:00:03.706 Somewhere in the last twelve months, 00:00:03.746 --> 00:00:08.386 five products shipped that answer the same question in five different ways. 00:00:08.426 --> 00:00:12.006 What happens when a chatbot finally gets its own computer? 00:00:12.046 --> 00:00:13.753 Not a window on your screen. 00:00:13.793 --> 00:00:17.122 A machine, with a browser, a file system, and a login, 00:00:17.162 --> 00:00:19.806 that keeps working after you close the tab. 00:00:19.846 --> 00:00:23.924 This is the story of that shift, and of the five agents making it. 00:00:25.164 --> 00:00:27.058 Here is the part that matters. 00:00:27.098 --> 00:00:31.893 Every one of these five gives its agent the same thing: a virtual computer, 00:00:31.933 --> 00:00:36.406 out in the cloud, with its own browser and its own signed-in sessions. 00:00:36.446 --> 00:00:37.953 That is the whole trick. 00:00:37.993 --> 00:00:41.434 Before this, to get a machine to do something for you, 00:00:41.474 --> 00:00:43.755 the machine had to have a clean API. 00:00:43.795 --> 00:00:46.849 An interface built for programs, not for people. 00:00:46.889 --> 00:00:51.104 A lot of the software you need to get work done has no such thing. 00:00:51.144 --> 00:00:53.812 So these agents do what a person would do. 00:00:53.852 --> 00:00:55.036 They open the page. 00:00:55.076 --> 00:00:55.746 They click. 00:00:55.786 --> 00:00:56.519 They log in. 00:00:56.559 --> 00:00:58.969 The computer is the universal adapter. 00:00:59.009 --> 00:01:03.675 It is also the biggest security decision any of these companies has made, 00:01:03.715 --> 00:01:05.480 and we will come back to it. 00:01:06.720 --> 00:01:08.941 Start with the oldest of the five. 00:01:08.981 --> 00:01:13.861 On the eleventh of August, twenty twenty-six, xAI opened Grok Bot in beta, 00:01:13.901 --> 00:01:14.991 alongside Cursor. 00:01:15.031 --> 00:01:20.243 The word Bot does not mean the reply bot on X, and it is not the Grok chat app. 00:01:20.283 --> 00:01:22.504 It is a set of named AI teammates. 00:01:22.544 --> 00:01:27.490 Each one gets its own cloud computer with a desktop, a browser, a terminal, 00:01:27.530 --> 00:01:28.155 and files. 00:01:28.195 --> 00:01:30.216 You name it, you give it a job, 00:01:30.256 --> 00:01:34.006 and you message it the way you would message a colleague. 00:01:34.046 --> 00:01:36.200 On phone or desktop, same thread. 00:01:36.740 --> 00:01:40.039 Three things make Grok Bot different from the rest. 00:01:40.079 --> 00:01:41.938 First, it learns by watching. 00:01:41.978 --> 00:01:47.110 You record yourself doing a task once, and it turns that into a reusable skill. 00:01:47.150 --> 00:01:50.842 Second, it has routines, which is a polite word for cron, 00:01:50.882 --> 00:01:54.377 so a bot can work on a schedule with no prompt at all. 00:01:54.417 --> 00:01:56.669 Third, the bots talk to each other. 00:01:56.709 --> 00:02:00.663 One can ask another for help, they share context in a thread, 00:02:00.703 --> 00:02:02.299 and one can spin up more. 00:02:02.339 --> 00:02:04.460 That is a team, not an assistant. 00:02:04.500 --> 00:02:06.162 And there is no free tier. 00:02:06.202 --> 00:02:10.876 It rides on SuperGrok and on Cursor plans, with usage billed separately. 00:02:10.916 --> 00:02:14.608 It passed four hundred thousand users in its first month. 00:02:15.848 --> 00:02:19.863 A month later, on the eighth of September, Meta shipped Muse. 00:02:19.903 --> 00:02:21.924 Internally it was called Hatch, 00:02:21.964 --> 00:02:25.912 and Meta spent extra months on security before releasing it. 00:02:25.952 --> 00:02:28.903 Muse runs on something called Muse Secure VM, 00:02:28.943 --> 00:02:33.756 a dedicated machine in the cloud that holds both the agent and your data. 00:02:33.796 --> 00:02:37.678 Sitting on that same machine, isolated at the system level, 00:02:37.718 --> 00:02:39.938 is a second agent called Sentinel. 00:02:39.978 --> 00:02:44.392 Nothing Muse does reaches the internet unless Sentinel approves it. 00:02:44.932 --> 00:02:48.223 Meta is the most aggressive here on one specific point. 00:02:48.263 --> 00:02:51.615 Muse has no visibility into your passwords or your card. 00:02:51.655 --> 00:02:54.825 You hand it credentials, they go into a secure store, 00:02:54.865 --> 00:02:57.611 and it can use them without ever reading them. 00:02:57.651 --> 00:03:02.577 It asks before it sends an email or spends money, and it keeps a full audit trail. 00:03:02.617 --> 00:03:05.424 It can check out through Link, built by Stripe, 00:03:05.464 --> 00:03:09.482 which makes it the first agent covered by that purchase protection. 00:03:09.522 --> 00:03:12.025 And Meta says a Confidential VM is coming, 00:03:12.065 --> 00:03:15.780 where the whole machine is encrypted with a key only you hold, 00:03:15.820 --> 00:03:17.537 so not even Meta can open it. 00:03:17.577 --> 00:03:19.717 It launched free with a usage limit, 00:03:19.757 --> 00:03:23.775 and it hit the number one free app on the US iOS chart within days. 00:03:23.815 --> 00:03:26.379 Then Amazon blocked it, on privacy grounds. 00:03:26.419 --> 00:03:27.954 Shopify partnered with it. 00:03:27.994 --> 00:03:29.892 That is the market talking back. 00:03:31.132 --> 00:03:34.919 Five weeks after Muse, on the twenty-ninth of September, 00:03:34.959 --> 00:03:37.105 OpenAI announced Dots at DevDay. 00:03:37.145 --> 00:03:39.087 Same promise, new vocabulary. 00:03:39.127 --> 00:03:41.410 A dot is powered by GPT six Astra, 00:03:41.450 --> 00:03:44.827 it has its own cloud computer and its own browser, 00:03:44.867 --> 00:03:48.312 and it can connect to more than four thousand apps. 00:03:48.352 --> 00:03:52.343 You talk to it inside ChatGPT, and also in Slack and Teams, 00:03:52.383 --> 00:03:56.580 and it carries the same context into every one of those rooms. 00:03:57.120 --> 00:04:01.252 What is different about Dots is the layer above the agent. 00:04:01.292 --> 00:04:03.482 OpenAI shipped specialist dots. 00:04:03.522 --> 00:04:05.352 These are not generalists. 00:04:05.392 --> 00:04:09.596 A specialist dot has its own identity, its own credentials, 00:04:09.636 --> 00:04:12.258 hardware provisioned by your IT team, 00:04:12.298 --> 00:04:15.854 and deep integrations with your systems of record. 00:04:15.894 --> 00:04:18.516 It is an employee-shaped abstraction. 00:04:18.556 --> 00:04:22.184 And OpenAI is wiring them into Microsoft Agent 365, 00:04:22.224 --> 00:04:24.630 which is how it gets into offices. 00:04:24.670 --> 00:04:30.241 Dots start on Pro and Business Premium plans, one per user, with more to come. 00:04:30.281 --> 00:04:35.276 Pro users in the EEA, Switzerland, and the UK were excluded at launch. 00:04:36.516 --> 00:04:37.323 Now the turn. 00:04:37.363 --> 00:04:40.778 On the twenty-fourth of November, twenty twenty-five, 00:04:40.818 --> 00:04:45.536 a developer named Peter Steinberger released something he called Warelay. 00:04:45.576 --> 00:04:48.860 It got renamed twice, to Clawdbot and then Moltbot, 00:04:48.900 --> 00:04:51.401 before settling on OpenClaw in January. 00:04:51.441 --> 00:04:54.725 It is the opposite move from the four you just saw. 00:04:54.765 --> 00:04:58.506 There is no company behind it and nothing to subscribe to. 00:04:58.546 --> 00:05:02.351 It is MIT licensed, stewarded by an independent non-profit, 00:05:02.391 --> 00:05:06.392 and it has three hundred and ninety-one thousand GitHub stars. 00:05:06.932 --> 00:05:10.673 The design decision that defines OpenClaw is the Gateway. 00:05:10.713 --> 00:05:14.652 One process, running on your own machine or your own server, 00:05:14.692 --> 00:05:19.693 is the control plane for everything: your sessions, your tools, your events, 00:05:19.733 --> 00:05:21.418 and every chat connection. 00:05:21.458 --> 00:05:22.811 Channels are plugins. 00:05:22.851 --> 00:05:27.454 WhatsApp, Telegram, Discord, Slack, Signal, iMessage, Teams, and more. 00:05:27.494 --> 00:05:29.444 And the model is a plugin too. 00:05:29.484 --> 00:05:33.755 Claude, GPT, DeepSeek, or something running on your own hardware, 00:05:33.795 --> 00:05:36.209 and swapping it changes nothing else. 00:05:36.249 --> 00:05:40.123 The catch is the price of that freedom, and it is not zero. 00:05:40.163 --> 00:05:41.914 It will not hold your hand. 00:05:41.954 --> 00:05:45.628 You secure it yourself, and the project says so plainly. 00:05:46.868 --> 00:05:49.945 Hermes is the other open one, from Nous Research, 00:05:49.985 --> 00:05:54.335 and it is MIT licensed with two hundred and fifty-one thousand stars. 00:05:54.375 --> 00:05:57.834 Where OpenClaw is gateway first, Hermes is agent first. 00:05:57.874 --> 00:06:00.061 Its whole pitch is a learning loop. 00:06:00.101 --> 00:06:04.832 It writes its own skills from experience, improves them while it uses them, 00:06:04.872 --> 00:06:09.731 reminds itself to save what matters, and searches its own past conversations. 00:06:09.771 --> 00:06:15.520 It is the only one of these five that claims to get better the longer you leave it running. 00:06:16.060 --> 00:06:18.806 The memory is plain files you can read. 00:06:18.846 --> 00:06:22.449 A MEMORY file, a USER file, and a personality file. 00:06:22.489 --> 00:06:27.306 Skills are markdown with front matter, the same shape OpenClaw uses. 00:06:27.346 --> 00:06:32.092 It reaches twenty-one messaging platforms, nineteen of them native, 00:06:32.132 --> 00:06:35.164 and it runs wherever you can run a process. 00:06:35.204 --> 00:06:36.450 A five dollar VPS. 00:06:36.490 --> 00:06:37.093 A laptop. 00:06:37.133 --> 00:06:37.950 A container. 00:06:37.990 --> 00:06:38.879 Your own GPU. 00:06:38.919 --> 00:06:42.665 And it is model agnostic in the same way OpenClaw is, 00:06:42.705 --> 00:06:45.879 which is why people migrate one to the other. 00:06:45.919 --> 00:06:47.951 Where it differs is the loop. 00:06:47.991 --> 00:06:49.808 OpenClaw waits to be told. 00:06:49.848 --> 00:06:52.308 Hermes writes down what it learned. 00:06:53.548 --> 00:06:55.084 So put them side by side. 00:06:55.124 --> 00:06:58.487 The one axis that separates them cleanly is ownership. 00:06:58.527 --> 00:07:00.756 Muse, Dots, and Grok Bot are rented. 00:07:00.796 --> 00:07:04.475 You do not choose the machine, you do not choose the model, 00:07:04.515 --> 00:07:06.807 and you can only leave by cancelling. 00:07:06.847 --> 00:07:08.697 OpenClaw and Hermes are owned. 00:07:08.737 --> 00:07:13.172 You choose the hardware, you choose the model, you can read the source, 00:07:13.212 --> 00:07:15.252 and leaving is deleting a folder. 00:07:15.792 --> 00:07:17.928 The second axis is who it is for. 00:07:17.968 --> 00:07:22.545 Muse and Dots are built for everyone, with the learning curve removed. 00:07:22.585 --> 00:07:27.096 Grok Bot is built for people who want automation without seeing code. 00:07:27.136 --> 00:07:31.911 OpenClaw and Hermes are built for people who will read the documentation. 00:07:31.951 --> 00:07:33.493 The third axis is price. 00:07:33.533 --> 00:07:37.912 Muse is free with limits, then twenty or a hundred dollars a month. 00:07:37.952 --> 00:07:40.287 Dots comes with a paid ChatGPT plan. 00:07:40.327 --> 00:07:44.310 Grok Bot requires SuperGrok or Cursor and bills usage on top. 00:07:44.350 --> 00:07:49.520 OpenClaw and Hermes are free, and you pay only for the model you choose to use. 00:07:50.760 --> 00:07:51.491 So which one. 00:07:51.531 --> 00:07:57.304 If you want an agent that works while you sleep and you do not want to think about infrastructure, 00:07:57.344 --> 00:08:02.049 that is Muse or Dots, and the difference is whose ecosystem you already live in. 00:08:02.089 --> 00:08:06.557 If you want agents as colleagues, learnable by demonstration, on a schedule, 00:08:06.597 --> 00:08:07.565 that is Grok Bot. 00:08:07.605 --> 00:08:12.488 If you want to own the whole thing and you are willing to run it, that is OpenClaw. 00:08:12.528 --> 00:08:16.106 And if you want an agent that teaches itself, that is Hermes. 00:08:16.146 --> 00:08:17.530 The real answer, though, 00:08:17.570 --> 00:08:22.868 is that this category is twelve months old and all five of these are the first generation. 00:08:24.108 --> 00:08:28.680 Every number in this film came from a primary source: the OpenAI, Meta, 00:08:28.720 --> 00:08:33.356 and xAI announcement posts, the OpenClaw and Hermes GitHub repositories, 00:08:33.396 --> 00:08:35.435 and the two documentation sites. 00:08:35.475 --> 00:08:39.721 Star counts were read from the GitHub API on the fifth of October, 00:08:39.761 --> 00:08:40.891 twenty twenty-six. 00:08:40.931 --> 00:08:44.333 If you want the sources, they are in the description. 00:08:44.373 --> 00:08:45.632 Thanks for watching.