Grok Bot vs Muse vs Dots vs Claude Duet vs Hermes: My Honest Tier List — Transcript

Full transcript

1. 0:00Grockbot, dots, Hermes agent, Mem and
2. 0:03now even Claude Duet. It's obvious every
3. 0:06single AI lab right now is trying to
4. 0:08create a personal agent that will
5. 0:09promise to save you more money, get more
6. 0:11work done, or everything in between. The
7. 0:13problem is they kind of all do the same
8. 0:16thing and they also promise the same
9. 0:18thing. So, how do you know which ones to
10. 0:19go with? And a lot of people have been
11. 0:21feeling very overwhelmed and having what
12. 0:23they call personal agent fatigue,
13. 0:25including myself. That being said, there
14. 0:27are two agents that for me have become
15. 0:29the frontrunners. They're agents that I
16. 0:31use every single day. I think they're
17. 0:32going to become a permanent part of my
18. 0:34stack and I'm going to explain how to
19. 0:35use them and also show the pros and cons
20. 0:37between each of these agents. So, the
21. 0:39first one that I wanted to talk about is
22. 0:41Hermes agent. And lately Hermes agent
23. 0:45has been kind of getting a bad rap on
24. 0:47Twitter and on any kind of other social
25. 0:49media platform, which is kind of a shame
26. 0:51because originally Open Claw came out in
27. 0:54the beginning of the year and it was
28. 0:55pretty groundbreaking what Peter
29. 0:57Steinberger had made, but
30. 0:59to get any from meaningful work done was
31. 1:01pretty much impossible with it. It was
32. 1:02breaking down all the time.
33. 1:05All you were spending so much time
34. 1:07tinkering with it trying to get it to
35. 1:08work. So, it really didn't add any
36. 1:10productivity. But, then Hermes entered
37. 1:12the picture. And Hermes kind of worked
38. 1:15outside of the box. You weren't sitting
39. 1:16there tinkering with it. You weren't
40. 1:18trying to over-customize it. Now, it is
41. 1:20open source, so you can still do all
42. 1:22those things, but the big benefit was it
43. 1:24was just much better put together. It
44. 1:26had a real team behind it and it was a
45. 1:28first kind of personal agent that a lot
46. 1:30of people were using to get actual
47. 1:31productive meaningful work done. I used
48. 1:33it in Slack. My entire team was able to
49. 1:35interact with my Hermes agents. I had
50. 1:37Hermes agents that could that would do
51. 1:39in revenue management. I had Hermes
52. 1:41agents that were like turning off pool
53. 1:43heating,
54. 1:44um, working on guest communications. We
55. 1:46had so many different things that we
56. 1:48were running and still to this day
57. 1:49continue to run. So, overall it was a
58. 1:52great experience. Now, with all that
59. 1:54said, there were some real problems with
60. 1:56Hermes agent as well. And the main one
61. 1:58was it kind of started to get the open
62. 2:00claw issue, which was it was taking out
63. 2:03too much time tinkering with maintaining
64. 2:05it. It started to have updates where all
65. 2:07of a sudden things stopped working. I
66. 2:09mean, the amount of times
67. 2:11that I had an update or I would change
68. 2:13something about it and all of a sudden I
69. 2:14would get a DM from someone in my team,
70. 2:16"Hey, such and such agent isn't working
71. 2:18anymore." I mean, it was enough to drive
72. 2:19me mad. And don't get me wrong, I'm
73. 2:21still using them, but I am probably
74. 2:23going to start transitioning a lot of
75. 2:24the workload away from there unless
76. 2:26something changes really soon. The next
77. 2:28one, I use it on a daily basis. It is
78. 2:30called Grok Bot. A lot of people I'm
79. 2:31sure have heard about it. It is by
80. 2:33Elon's company, SpaceXSI. Yes, that is
81. 2:36the name of the company now. And what I
82. 2:37really love about it is you have a
83. 2:39separate bot for every single job. So,
84. 2:42this again, Hermes does a lot of that,
85. 2:44too. I want to be clear on that. But a
86. 2:45lot of the personal agents these days,
87. 2:47particularly, what they'll do is they'll
88. 2:49have one agent, you talk to it, and
89. 2:51sometimes it can spin off different
90. 2:53threads, sometimes it can spin off
91. 2:54different channels, but ultimately
92. 2:55everything happens through that one
93. 2:56thread. For me, I don't personally
94. 2:58always love that because I feel like
95. 3:00when you're talking about different
96. 3:01topics at once things get a little bit
97. 3:02convoluted and confusing.
98. 3:04But with Grok Bot, um
99. 3:06every single agent can have its own
100. 3:08role, its own job. So, I'll give you an
101. 3:10example. I have Grok Bots that only work
102. 3:12on my Meta Ads. And so, anytime I talk
103. 3:14about Meta Ads for one particular
104. 3:16company, I'm talking to that bot.
105. 3:18Or it could be that, um your Google Ads
106. 3:20manager. It could be a project manager.
107. 3:24Ultimately, you can define the roles
108. 3:26what you want, but for me, it does help
109. 3:28a lot to segment these bots for very,
110. 3:31very specific roles and use cases. And I
111. 3:33know when that topic comes up, boom, I'm
112. 3:36going to talk to that bot. I will say
113. 3:38this as well, Grok, if you did want to
114. 3:40talk to this one agent on one thread,
115. 3:43Grok Bot does have a chief of staff bot,
116. 3:45and it's the most proactive of all the
117. 3:47bots. Uh I think they actually
118. 3:48programmed it that way as well. And so,
119. 3:51it can then delegate work out to all the
120. 3:53other bots. So, if you did want to just
121. 3:55talk to one bot and have one {quote} bot
122. 3:57thread, you can still do that in Grok
123. 3:59Bot as well. Again, it's not my personal
124. 4:01way to uh favorite to work for them, but
125. 4:03it it is possible. And Grok Bot has some
126. 4:06other cool features. It has its own
127. 4:07mailbox. It has great integrations with
128. 4:09one password where like it will ask you
129. 4:11for an authorization when it needs
130. 4:12something, but in a way done that was
131. 4:14better than all the other bots.
132. 4:16Has a ton of connectors obviously right
133. 4:18now, too. And two of the coolest
134. 4:19features is Grok Bot is also going to be
135. 4:21able to call on your behalf. So, whether
136. 4:23you want to make a reservation at a
137. 4:24restaurant, or in my case, if I have to
138. 4:26call, let's say I have to change my
139. 4:29internet plan. I don't want to be on
140. 4:30hold for 45 minutes. My Grok Bot can go
141. 4:32ahead and do that and can dial me in
142. 4:34when somebody actually answers the
143. 4:35phone.
144. 4:36And then the last one, and this might be
145. 4:38my favorite one. I drive a Tesla Model
146. 4:40Y. My Grok Bot can directly connect to
147. 4:43my Tesla. The same Grok Bot that's in my
148. 4:45app, and I can then delegate work to it.
149. 4:47I can ask about certain emails. I can
150. 4:49ask about some work that workflows that
151. 4:50I've going on. It's the same Grok Bot.
152. 4:52So, if you were doing your full
153. 4:54self-driving, which again, if you don't
154. 4:55have a Tesla and you don't have full
155. 4:56self-driving, highly recommend it. Big
156. 4:58fan. You can just go ahead and do that
157. 5:00full self-driving and still be able to
158. 5:02work fully while you're in the car,
159. 5:03which is a really really cool feature.
160. 5:05Now, the next one is an interesting one.
161. 5:07So, Meta was really behind on the AI
162. 5:10race for a long time, and then all of a
163. 5:13sudden they came out with Muse uh by
164. 5:15Meta. And when it came out, I wasn't
165. 5:17actually that enthused about it. It
166. 5:19didn't seem like it had anything that
167. 5:21Grok Bot or Hermes couldn't do. It
168. 5:23didn't really seem like Meta's models
169. 5:25were even good enough to try it. But
170. 5:27then when I actually did try it, I was
171. 5:29pleasantly surprised. The coolest thing
172. 5:31about it is it was at least at the time
173. 5:34the fastest agent by a mile. I mean, it
174. 5:36was lightning fast. I don't know what
175. 5:38Meta did. I don't know how much compute
176. 5:39they have to do that, but it was
177. 5:41lightning fast, and it got things done
178. 5:43really fast, too. And the second thing
179. 5:45is Muse has a really really cool UI and
180. 5:47UX experience. Like even like the avatar
181. 5:50for your bot is like this cute little
182. 5:52furry creature. I don't know if you know
183. 5:55what you would call it. For me, I named
184. 5:56my bot Alfred after The Dark Knight, and
185. 5:58so I gave him like the Alfred butler
186. 6:00look, and you can make him look like
187. 6:02that. Not awfully obviously not Michael
188. 6:04Caine,
189. 6:05but he can look like a British butler
190. 6:07and actually like be taking actions and
191. 6:09everything. It's a lot of fun to look
192. 6:10at. Um and again, highly recommend you
193. 6:12trying it just for that reason alone. It
194. 6:14also just feels really easy to use for a
195. 6:16beginner. Like Muse is the first agent
196. 6:19that I had my wife use. She doesn't know
197. 6:21anything about AI. She barely has ever
198. 6:23owned a computer. And a lot of times she
199. 6:25would ask me throughout the day random
200. 6:27questions that like realistically a
201. 6:29Google search could solve. And so
202. 6:31eventually I installed the Muse agent
203. 6:32for her. She absolutely loves it now.
204. 6:34She calls it I think Colin or something
205. 6:36like that. And a lot of the random
206. 6:38questions that she would either have to
207. 6:39Google before or ask me, especially when
208. 6:41I'm working, that can get really
209. 6:43difficult sometimes, um she ended up
210. 6:45just asking the Muse agent now. And it's
211. 6:46done a lot of really cool things for
212. 6:48her. I mean, we just recently planned
213. 6:50our daughter's birthday party, so it
214. 6:51sent out all the RSVPs. Um it gave
215. 6:54recommendations of where we should host
216. 6:55the party. It also like flagged ahead of
217. 6:58time that it was supposed to rain that
218. 6:59day, so our original outdoor plan we
219. 7:01ended up moving indoors. And Muse
220. 7:03proactively told her that. And it also
221. 7:05has like a really cool newsfeed. So like
222. 7:08everybody obviously is against
223. 7:09doomscrolling, but you can curate a
224. 7:11newsfeed for your Muse that you can just
225. 7:13go ahead and look at it. It's
226. 7:14aesthetically really nice to uh pleasing
227. 7:15to look at. It has a goals tab. It even
228. 7:18has an ideas tab. So if you don't even
229. 7:19know what to do with your Muse, it has
230. 7:21really fun ideas of things that you
231. 7:22could do with it. And it can't be
232. 7:24minimized that Meta has a ton of
233. 7:26distribution. It has a ton of data. So
234. 7:28out of all the agents, it feels the most
235. 7:31I don't know if human is the word when
236. 7:34speaking to it, but it feels the most
237. 7:35easy to talk to. Like you're just
238. 7:37talking to a friend or something like
239. 7:38that. Now, my litmus test for that is
240. 7:40every single personal agent I've ever
241. 7:42had, there's this line in The Dark
242. 7:44Knight where um
243. 7:46Bruce Wayne says to Alfred, "People are
244. 7:48dying, Alfred. What would you have me
245. 7:49do?" And then Alfred says, "Endure,
246. 7:51Master Wayne." And I've used that to
247. 7:53every single one of my agents. And
248. 7:55usually they would say, "Oh my god, you
249. 7:56should call 911 if somebody is dying."
250. 7:58But Muse was the first agent that
251. 8:00literally said the line back, "Endure,
252. 8:02Master Wayne." And I was like, "Wow,
253. 8:05this is AGI." No, it's not actually AGI,
254. 8:08but I thought it was a cool touch that
255. 8:09it had that kind of awareness. Meta is
256. 8:11also going to be releasing a lot of
257. 8:12physical devices, which are going to
258. 8:13have awesome integrations with Muse.
259. 8:15Obviously, the the Meta Ray-Bans, they
260. 8:17have like the Muse charm that they're
261. 8:19releasing. I'm sure there's going to be
262. 8:21a lot of other stuff, and it's going to
263. 8:22have direct integrations with Muse. So,
264. 8:24this will be your technically your 24/7
265. 8:26always-on personal assistant. I'm sure
266. 8:28the other ones are going to be doing
267. 8:29similar things. Obviously, Grok bot did
268. 8:30it with Tesla. But again, Meta has a
269. 8:33distribution, has the compute, so I
270. 8:35would not count them out of them winning
271. 8:36potentially this personal agent AI race.
272. 8:39Now, the biggest downside, obviously, is
273. 8:41that Meta's models are not that great
274. 8:43yet, although they have improved a lot.
275. 8:44So, with the amount of compute that they
276. 8:46have, I'm sure they're going to catch up
277. 8:48or get very close to catching up with
278. 8:50OpenAI and Anthropic. Next one is Dots.
279. 8:54What can I say about Dots? I mean,
280. 8:57to me, and I hate to say this cuz
281. 8:58overall, I love OpenAI's models, I
282. 9:00thought Dots was a colossal
283. 9:02disappointment. And the reason why, they
284. 9:04hyped it up during Dev Days so much. I
285. 9:07think it was Tibo that said they had
286. 9:08like 6 months worth that they were 6
287. 9:11months ahead in terms of development.
288. 9:12So, I thought that they were really
289. 9:14going to release something unique and
290. 9:15really special. And then they released
291. 9:17Dots, and when it came out,
292. 9:19it wasn't really anything special. Like,
293. 9:21everything that it did, Grok bot could
294. 9:23do it, Muse could do it.
295. 9:25It was super buggy. Its computer use
296. 9:27didn't work really well.
297. 9:29And and the worst part was it it
298. 9:31required to be part of the Codex app,
299. 9:32and I was originally the biggest fan of
300. 9:34the Codex app, but over time, it got
301. 9:36really confusing and bloated and really
302. 9:38hogs the RAM of your computer. So, the
303. 9:40fact that you had to be in Codex to use
304. 9:42Dot was kind of a real negative and
305. 9:45downside for me. Again, it is powered by
306. 9:47the GPT Astra model, which are right now
307. 9:49are OpenAI's strongest models. Although,
308. 9:51with the new release of Tera 6.1, I'm
309. 9:53pretty sure they're going to be pivoting
310. 9:55all the Dot agents to that. It can spin
311. 9:57up different Codex threads when you give
312. 9:59it things to do, and then you can go
313. 10:00ahead and look at those Codex threads.
314. 10:02It also has a feature where you can call
315. 10:04and talk to the agent. For me
316. 10:05personally, I don't understand why they
317. 10:07made it like a phone call. It absolutely
318. 10:08makes no sense to me. Like, you should
319. 10:10just have one button where you
320. 10:11immediately you can talk to your dot.
321. 10:12That would make a lot more sense. But,
322. 10:14again, you can talk to it. GPT's voice
323. 10:17mode is really good. Um so, there are
324. 10:18benefits to that. It's been at least
325. 10:20over a week since I've last used it, and
326. 10:22I know they're iterating a lot, so I'm
327. 10:23sure improvements will be made along the
328. 10:24way. But, again, there is something
329. 10:26about first mover's advantage and
330. 10:27getting used to what you're already
331. 10:29working with, and I think
332. 10:30I think OpenAI kind of missed the mark
333. 10:32with dots.
334. 10:33Now, the last one is kind of
335. 10:34interesting. It's Claude Duet, and it's
336. 10:36Anthropic's version of the personal
337. 10:37agent. Now,
338. 10:38I'm going to start off by saying in some
339. 10:41degrees, I do have personal agent
340. 10:42fatigue. Like, I've tried a lot of
341. 10:44different ones, and then finally
342. 10:45Anthropic released theirs. But, by the
343. 10:47time they released it, myself, and I'm
344. 10:48sure a lot of other people were already
345. 10:50kind of wary about trying another agent,
346. 10:52and they're already really kind of set
347. 10:53in their workflows. That being said,
348. 10:56there are real positives to Claude Duet.
349. 10:57For starters, it is powered by the best
350. 10:59model in the world right now, which is
351. 11:00Opus 5.5. I mean, Opus 5.5 is leaps and
352. 11:04bounds above everything else right now.
353. 11:06It has incredible usage. It's a really
354. 11:07really good model. And it's pow- And
355. 11:09Claude Duet is powered by it, so that
356. 11:10alone gives it already an advantage over
357. 11:12every single other personal agent. Now,
358. 11:14the issue that I have with it was kind
359. 11:16of the same issue that I had with Muse
360. 11:18is that it's just one thread. And again,
361. 11:20they do something cool where like, you
362. 11:21have one thread, but then it will spin
363. 11:22off different uh channels like in Slack.
364. 11:25So, anytime you talk about your
365. 11:26finances, it will go to the finance
366. 11:27channel. Anytime you talk about um a
367. 11:30certain project, it'll go to that
368. 11:31project. Anytime you talk about video
369. 11:33editing, it will talk it will spin up a
370. 11:34channel and keep it in that video
371. 11:36editing channel. And that is cool, and
372. 11:38it's obviously uh mirrored and tailored
373. 11:41off the Claude code projects feature,
374. 11:42which recently came out.
375. 11:44That being said, again, it's not my cup
376. 11:46of tea. I'd rather have virtual agents
377. 11:47that have names, that have really
378. 11:49clearly defined roles, and not one agent
379. 11:52that I'm talking to where I'm starting
380. 11:53to get confused what I was talking
381. 11:54about. Like, if you have five, six
382. 11:56different topics you're going on about,
383. 11:57like, you get really confused and mixed
384. 11:59up with with with what you're doing at
385. 12:00that point. Now, obviously, Opus 5.5, it
386. 12:03creates incredible decks, incredible
387. 12:05dashboards. Again, it's proactive. It
388. 12:07does a lot of things great.
389. 12:08My two cents is I think all these
390. 12:11personal agents probably are going to
391. 12:12converging to a degree. So, to then go
392. 12:15and do a whole thing with Claude do it,
393. 12:17it's not super appealing to me
394. 12:18personally. So, with that said, we are
395. 12:20going to be doing a tiers list of where
396. 12:23I would place each agent. So, I would
397. 12:26say dots is going to be C tier. It felt
398. 12:30rushed, it felt bloated. It doesn't
399. 12:31really do anything that Grok Bot or Muse
400. 12:33couldn't already do. So, for that, it
401. 12:35got a C tier. Hopefully in the future,
402. 12:37maybe it'll rise up, maybe they'll
403. 12:38improve it. Right now, I think it's been
404. 12:39a bust overall for Open AI. Second one
405. 12:41is Hermes. Or or Hermes? How do people
406. 12:44pronounce that anyway? Uh I'm putting
407. 12:46that at C tier. It used to be A or S
408. 12:48tier for me, but the fact is, again, it
409. 12:50got bloated. There's a lot of tinkering
410. 12:52involved. There's a lot of maintenance
411. 12:53to do. Again, if you're trying to get
412. 12:55real meaningful work done in business or
413. 12:56anything I don't think Hermes is the way
414. 12:58to go outside of a couple of really
415. 13:00specific use cases, whether you're if
416. 13:02you're running local models or if again,
417. 13:04you run it run Slack agents, although
418. 13:06Grok Bot does that now, too. But
419. 13:07overall, I find it's just too much work.
420. 13:10I would say it's a C tier right now.
421. 13:13Claude do it.
422. 13:14Again, it's run by the best model in the
423. 13:16world. Um
424. 13:17there isn't really too much negative to
425. 13:18say about it other than the format isn't
426. 13:21for me. I don't like the one thread. If
427. 13:22you like that one thread kind of thing,
428. 13:24it's probably S tier for you. For me,
429. 13:26I'm going to put it at A tier. Uh Muse.
430. 13:29I'm going to put Muse on A tier as well.
431. 13:31It's fast, it's proactive. Uh it talks
432. 13:33to me the way I like. Uh there's a lot
433. 13:35of really good things about it. The UI
434. 13:36is probably on par with Grok Bot as
435. 13:38being the best UI for agents.
436. 13:41Um but ultimately, the models themselves
437. 13:43are not as strong as some of the other
438. 13:45models available right now, and because
439. 13:46of that, I ended up putting it in an A
440. 13:48tier. And again, it kind of had that
441. 13:49same thing of one thread only. So,
442. 13:52I really like Muse as a personal agent
443. 13:54for my day-to-day life, but for pure
444. 13:56work reasons, I generally don't like
445. 13:58Muse.
446. 14:00And then my favorite one is going to be
447. 14:01Grok Bot. Now, I made Twitter posts
448. 14:03about this like a month or 2 months ago
449. 14:05uh disparaging Grok Bot because the
450. 14:07usage was just so bad. I felt you
451. 14:09couldn't get any meaningful work done.
452. 14:11That has since improved a lot. I think
453. 14:13the usage issues kind of are still a
454. 14:15thing, but they've improved a lot. I can
455. 14:17get meaningful work done with it and
456. 14:18more than anything else, it just feels
457. 14:20incredibly reliable. It works outside
458. 14:22the box. I really like the team that's
459. 14:24behind it, too. And so for me, for pure
460. 14:26work reasons, I think most people should
461. 14:28probably be going with Grokpod. So there
462. 14:30you have it. Ultimately, it comes down
463. 14:31to how you like to talk to your agents.
464. 14:33I have a very clear way I like to talk
465. 14:35to mine. Yours may be different, so your
466. 14:37list may be looking a little bit
467. 14:38different. So feel free to experiment
468. 14:39with different ones and it ultimately
469. 14:41don't count out open source models like
470. 14:44or open source harnesses like Hermes
471. 14:46because I think as the models get
472. 14:48better, faster, smarter, I do think the
473. 14:50future is still going to be open source
474. 14:52like even something like an open claw. I
475. 14:55think these are going to be coming back
476. 14:56at some point when the models get
477. 14:58really, really good. So So ultimately,
478. 14:59don't count them out. Keep experimenting
479. 15:01with all of these different agents. Find
480. 15:03your best one and once you find one that
481. 15:05you like, just stick with that until
482. 15:06something clearly better comes out.
483. 15:09Don't sit there experimenting with every
484. 15:10single one of them if you don't need to.
485. 15:12All right. Hope to see you guys in the
486. 15:13next video. Bye.