Claude's New Updates Are Insane, But GrokBot Might Be Bigger
A 20-item weekly rundown of AI agent news: Claude Dashboards and Motion, GrokBot routing to Opus 5.5, ChatGPT's inline mini-apps, Nano Banana 2.1, and Adobe's open-source clones.
Posted
3 days ago
Duration
Format
Listicle
hype
Views
144.5K
1.4K likes
57 · 43
Big Idea
The argument in one line.
This week's real AI story is orchestration rather than raw capability: Claude and GrokBot both shipped systems that route, coordinate, and animate work across other tools and models instead of just getting individually smarter.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You already use Claude Code, Codex, or GrokBot regularly and want to know which new features actually matter this week.
You're a video editor or motion designer curious how far AI-generated motion graphics have come without touching a traditional timeline tool.
You track the AI agent space and want a fast, timestamped digest instead of reading twenty separate announcement threads yourself.
SKIP IF…
You want deep technical documentation on any single feature here; this is a rapid-fire overview, not a tutorial.
You have no existing familiarity with Claude, ChatGPT, GrokBot, or Google's AI products; the video assumes you already know all four.
TL;DR
The full version, fast.
This week's AI agent news centers on orchestration rather than raw model gains. Claude shipped two new artifact types, Dashboards and Motion, that turn connected-tool data into shareable animated narratives, alongside a broader rollout of Claude Projects, a coordinator that routes requests across local and cloud agent threads. GrokBot's biggest move is routing: it now auto-selects the best backend model for a task, including Claude Opus, instead of forcing everything through Grok's own models. OpenAI added mid-task steering to Codex and an inline mini-app generator called Intelligent UI to ChatGPT. Elsewhere, creators report editing nearly all their video with Opus 5.5's code-generated motion graphics, and open-source clones of Adobe's software suite are already circulating, built the same way.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Cold open teasing the three headline stories: Claude's biggest Claude Code feature yet, a new ChatGPT interface, and GrokBot adding Claude Opus.
00:19 – 02:30
02 · Claude Dashboards & Claude Motion
Anthropic's own demo: a year of bike-share ride data becomes a dashboard, then an animated narrative video via Claude Motion, commentable by teammates and exportable to design tools. Dashboards ship on all paid plans; Motion is Team/Enterprise only.
02:30 – 04:01
03 · AI Video Editing with Opus 5.5
Creators are reportedly doing 95-99% of their video editing with Claude Opus 5.5, generating motion graphics purely in code. A 10-second looping clip made this way hit 1.7 million views.
04:01 – 06:08
04 · Claude Projects Rolls Out
Claude Projects now broadly available: a coordinator thread routes plain-language requests to specialized threads, some local and some cloud, all manageable from a phone.
06:08 – 06:45
05 · Haiku 5.5 Released
Anthropic's newest small model, positioned as a token-conserving default for everyday tasks on the $20/month plan, reportedly ahead of OpenAI's comparable small model.
06:45 – 08:21
06 · 28 Days of Codex
OpenAI commits to 28 straight days of either a feature improvement or a full credit reset for Codex. Early wins: composer predictions, instant steering, meeting notes, and a new decisions API.
08:21 – 09:20
07 · ChatGPT Intelligent UI
A new ChatGPT feature spawns a small purpose-built app inline inside the chat, like a calendar or a home-affordability calculator, shareable with anyone in the conversation.
09:20 – 09:58
08 · OpenAI Math Breakthroughs
OpenAI published a GitHub repo of AI-assisted solutions to hard math problems, advanced enough to draw pushback from parts of the scientific community.
09:58 – 11:28
09 · GrokBot Now Uses All Models
Elon Musk announces GrokBot will route tasks to whichever backend model performs best, including Claude Opus 5.5, Midjourney, and Suno, not just Grok's own models.
11:28 – 12:49
10 · GrokBot Team & Primary Bots
Team bots connect to Slack and share plugins across a whole team; primary bots act as a 'chief of staff' that delegates new requests to specialized bots rather than doing the work itself.
12:49 – 13:41
11 · Nano Banana 2.1
Google's updated image model improves mask-based editing, subject consistency, and realism, and runs faster than GPT's image model, though it reportedly still trails slightly on quality.
13:41 – 14:07
12 · Google Argon Rumors
Unconfirmed social-media chatter points to a new Google model, called 'Argon,' rumored to beat Claude Opus 5.5 on several tasks.
14:07 – 15:34
13 · Prepare with Antigravity
Recommendation to download Google's Antigravity tool now, on the logic that a lab's own tools get the deepest early access to that lab's next model.
15:34 – 16:50
14 · AI Is Recreating Adobe Software
A developer released seven open-source Adobe clones (PhotoCraft, VectorCraft, FilmCraft, and others), built largely with Claude Opus 5.5, running without Adobe's license restrictions.
16:50 – 17:23
15 · Video Games Reverse-Engineered & Merged
AI-assisted reverse engineering is merging unrelated games: Call of Duty inside Minecraft, zombies mixed into Fortnite, NHL mixed with zombies.
17:23 – 18:09
16 · Play Console Games in Your Browser
A browser tool streams full console titles, including Black Ops Zombies, Halo, Skate 3, and Modern Warfare 2, in multiplayer, with no local install needed.
18:09 – 19:33
17 · New Personal Agent Products
New personal-agent products launch or leak: Hark/Hark Pro, a wearable-bracelet agent from Zach Yadegari, and a $200M-funded competitor to Instinct, all framed as agents replacing single-purpose apps.
19:33 – 20:35
18 · Hermes Agent Raises $90M
Hermes, an open, model-agnostic alternative to GrokBot, raises $90 million and prepares an iOS app, letting users switch between underlying models.
20:35 – 20:50
19 · Outro
Close-out pointing back to agentnative.inc for every link referenced, plus the full weekly updates archive.
Atomic Insights
Lines worth screenshotting.
GrokBot now auto-routes hard tasks to whichever backend model performs best, including Claude Opus, Midjourney, and Suno, instead of forcing every request through Grok's own models.
Claude's new Dashboard and Motion artifacts let a team turn connected-tool data into a shareable, animated narrative without leaving the chat interface.
Claude Projects acts as a coordinator that routes requests to dozens of separate agent threads, some running locally on a machine and some in the cloud, controllable from a phone either way.
Creators report using Claude Opus 5.5 for 95-99% of their video editing, generating motion graphics purely from code instead of timeline software like Premiere or After Effects.
OpenAI's new steering feature lets a user redirect Codex mid-task immediately, instead of waiting for the current tool call to finish before the correction lands.
ChatGPT's Intelligent UI generates disposable, shareable mini-apps inline inside a chat response, like a calendar or an affordability calculator, instead of a static text answer.
A GrokBot primary bot behaves like a chief of staff: its first instinct on a new task is to delegate the work to a more specialized bot rather than do it directly.
An open-source developer cloned seven Adobe products, including Photoshop and Illustrator equivalents, built largely with Claude Opus 5.5 and running without Adobe's license restrictions.
AI-assisted reverse engineering is merging previously separate video games, producing mashups like Call of Duty running inside Minecraft or zombies mixed into Fortnite.
A browser tool now streams full PC games, including Black Ops Zombies, Halo, and Modern Warfare 2, in multiplayer, with no local install required.
Haiku 5.5 is positioned as a cost-saving default for everyday Claude tasks, reportedly outperforming OpenAI's comparable small model.
OpenAI published a GitHub repo of AI-generated math proofs difficult enough that parts of the scientific community have pushed back on how they were published.
Nano Banana 2.1 runs faster than Google's prior image models but, by the creator's own account, still trails GPT's image model slightly on raw output quality.
Several new personal-agent hardware and platform products launched or were rumored this week, signaling agents are increasingly positioned to replace standalone apps.
Takeaway
This week's AI news is about orchestration, not raw power.
WHAT TO LEARN
The most important AI shifts this week aren't new models, they're new systems for routing, coordinating, and animating work across models and tools people already use.
02Claude Dashboards & Claude Motion
Claude can now turn a year of connected-tool data into a polished dashboard, then convert that same dashboard into an animated video narrative without separate design software.
The animated version supports inline comments from teammates, so a data story gets revised collaboratively before it's presented, not just handed off as a finished export.
Dashboards are available on every paid plan, but the Motion video artifact is currently restricted to Team and Enterprise plans.
03AI Video Editing with Opus 5.5
A growing number of professional editors now do 95-99% of their motion graphics work by describing the result to an AI model in code rather than hand-animating in a timeline tool.
A 10-second, code-generated motion graphic clip pulled 1.7 million views, suggesting AI-native visual styles are resonating independent of production budget.
This shift moves the bottleneck in video production from software skill to direction and taste, since the AI now handles the technical execution.
04Claude Projects Rolls Out
A coordinator-style AI project automatically routes a plain-language request to the correct specialized thread, so the user never has to pick which agent handles a task.
Some project threads run locally on a machine and others run in the cloud, and both types can be checked and steered from a phone even when the computer is off.
After about a month of use, one creator's single project had accumulated 35 separate threads, each spun up just by talking to the coordinator rather than being manually created.
05Haiku 5.5 Released
Smaller, cheaper models are improving fast enough that routing routine tasks to them, instead of a flagship model, now makes sense for most day-to-day work.
On a fixed monthly plan, defaulting to a small model for simple tasks is a direct way to conserve usage limits for when a harder task actually needs the flagship model.
0628 Days of Codex
A coding agent running through consecutive days of either a shipped improvement or a full usage-credit reset means the tool can change meaningfully day to day during that window.
A 'steering' feature that redirects a running agent task immediately, instead of waiting behind its current tool call, avoids wasting tokens on work that's about to be discarded.
A decisions API shipped as a direct answer to a competitor's decision-tracking tool, and the same agent's backend was also optimized to run roughly 50% faster than at its initial announcement.
07ChatGPT Intelligent UI
An inline UI generator can produce a small, disposable app, like a calendar or a calculator, directly inside a chat reply instead of returning a block of text.
Because a generated mini-app lives inside a shareable conversation, another person can open the same link and use the same tool, not just read about it.
This continues a shift where standalone apps for narrow one-off tasks become unnecessary, since an agent can build the exact interface a request needs on the spot.
08OpenAI Math Breakthroughs
Publishing AI-assisted solutions to hard, previously open math problems signals frontier models are now being pointed at unsolved research math, not just applied tasks.
Results advanced enough to draw pushback from parts of the scientific community over how they were published are themselves a signal of how fast this capability moved.
09GrokBot Now Uses All Models
An agent platform that no longer locks every request to its own models, but automatically routes hard requests to whichever backend performs best, is a meaningful upgrade over staying locked to one vendor.
Cross-model routing like this is positioned as what makes a general personal-agent platform strongest right now, ahead of competitors that stay locked to a single model family.
10GrokBot Team & Primary Bots
A 'team bot' that connects to shared tools like Slack exposes its plugins and skills to an entire team, not just the person who created it.
A 'primary bot' that behaves like a chief of staff defaults to delegating a new task to a more specialized bot rather than attempting it directly, mirroring the coordinator-and-threads pattern used elsewhere.
11Nano Banana 2.1
An image model that improves mask-based editing, subject consistency, and realism while running faster than a comparable model, even if it still trails slightly on raw quality, shows the real differentiator maturing toward speed and control rather than output quality alone.
12Google Argon Rumors
Unconfirmed reports about a competing model outperforming the current flagship are worth tracking but should be weighted as social-media speculation, not an official claim, until confirmed.
13Prepare with Antigravity
A frontier lab's own first-party tools tend to get the earliest and deepest access to that lab's next model, so learning the tool ahead of a rumored release is a low-cost way to be ready for it.
14AI Is Recreating Adobe Software
Independent developers releasing open-source clones of major creative software, built largely with a single AI coding model and running without the original vendor's license restrictions, shows how far AI-assisted software cloning has come.
The same automation that let large companies cut staff and lean on AI is now letting individual developers use AI to rebuild the software those companies sell, which is a notable inversion worth watching.
15Video Games Reverse-Engineered & Merged
AI-assisted reverse engineering is now capable enough to merge previously unrelated games into working mashups, not just generate new cosmetic content on top of an existing engine.
This is possible because AI can now reconstruct a game's underlying logic well enough to port or combine it, which is a different and harder capability than content generation.
16Play Console Games in Your Browser
A browser-based tool streaming full console titles in multiplayer with no local install required is the same underlying reverse-engineering capability applied to distribution instead of remixing.
This kind of application reportedly got the gaming community genuinely excited about AI rather than defensive, which is notable given how much resistance AI has faced elsewhere in creative industries.
17New Personal Agent Products
Multiple new personal-agent products launching or leaking in the same week, including hardware like a wearable bracelet, signals investors and builders both expect agents to replace single-purpose apps.
A competitor raising roughly $200 million specifically to challenge an existing agent product signals investors expect this category to keep consolidating around a small number of winners.
18Hermes Agent Raises $90M
An open, model-agnostic agent platform that lets a user switch between underlying models, rather than being locked to one, may end up being the cheapest way to access frontier-model agent capability as open models keep closing the gap with proprietary ones.
Glossary
Terms worth knowing.
Claude Dashboards
An Anthropic artifact type that pulls data from a user's connected tools and lays it out as a shareable, interactive dashboard, available on all paid Claude plans.
Claude Motion
An artifact type, limited to Team and Enterprise plans, that turns a Claude Dashboard into an animated video narrative that teammates can comment on and revise together.
Claude Projects (coordinator/threads)
A project structure where one coordinator agent routes incoming requests to specialized sub-agent threads, some running locally and some in the cloud, all manageable from a single chat.
Steering
Redirecting a running AI coding agent mid-task with a new instruction; in OpenAI's Codex this now applies immediately instead of waiting for the agent's current tool call to finish.
Intelligent UI
A ChatGPT feature that generates a small, purpose-built interface, such as a calendar or calculator, inline inside a chat reply instead of returning plain text.
Team bot vs. primary bot (GrokBot)
A team bot is shared with a group and connected to tools like Slack so a whole team can use its plugins; a primary bot acts as an orchestrator that delegates incoming tasks to other specialized bots.
Antigravity
A Google development tool positioned as the one most likely to get early, deep access to Google's rumored upcoming model, referred to in the video as 'Argon.'
17:23toolRadiant Optimizer (browser game streaming)
19:33productHermes Agent
Quotables
Lines you could clip.
03:50
“It doesn't even need to use like Premiere or DaVinci or even After Effects for that matter, it can literally create videos with just JavaScript, like pure code.”
concrete, surprising claim about AI replacing an entire editing stack→ TikTok hook↗ Tweet quote
15:21
“This model, I swear, took psychedelics because it's so visually creative.”
vivid, quotable hyperbole about Opus 5.5's output→ IG reel cold open↗ Tweet quote
10:57
“It'll always choose the best model when you're using Grok bot.”
tight one-line summary of the week's biggest routing announcement→ newsletter pull-quote↗ Tweet quote
The Script
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
17px
Huge updates this week in the world of AI agents. Cloud is finally releasing their best feature they've ever shipped inside Cloud Code, which I've been talking about for a while. OpenAI is releasing a new interface into the ChatGPT app, and Grokbot is adding Cloud Opus.
directly into their platform. The first update this week in the world of AI agents is Motion and Dashboards, a new feature released by Anthropic into Claude. Let's take a look at that.
This is a short animation about empty bike docks. It started as a dashboard. Today I want to show you two new artifact types, Claude Dashboards and Claude Motion.
Pretend that I run the operations team at a bike share. Every weekday, the docks on the hill are empty by 8 .15. we need to get to the bottom of it.
We let Claude pull a year of ride data from our data warehouse and lay it out as a nice dashboard. As we look at the dashboard, a story takes shape, and we'll need to tell this story to the whole team at our next all -hands meeting to get them on board. So we asked Claude to turn this dashboard into an animation using Claude Motion.
Claude turns it into an animated narrative.
Claude usually does a pretty good job right away. But I can continue to tweak it and steer it with comments if I want to, right there on the video. I can send it to colleagues and they can comment too.
We can build the story together. And if the team wants to take it further, we can export it straight into their design tools. Today they announced two more types of artifacts.
The first one is a dashboard, and this is available to all plans, I believe. And then the next one is motion. And so a dashboard is you can take all of the data from all of your plugins, right?
In cloud, you can connect all. Of the tools that you use, whether you're a notion, Slack, Excel, anything, wherever your company's data is or your own personal data or your own personal finance, you can create these dashboards super easily. And now it's a type of artifact that you can easily share, make public, et cetera.
And the next thing that they announced, which is only available in the team and enterprise plans, unfortunately, you can create these motion graphic videos, as I've talked about in previous. videos. Claude is incredible at creating motion graphic videos with code and you can create those as a new artifact and you can even comment on them as a team.
So teams are going to be able to work together on motion graphic videos directly inside Claude. So the second update in the world of AI agents is not an official announcement. It's a realization by me, which is video editing is completely changing forever.
And after talking to a ton of different creators about their editing workflows, people are starting to use Claude specifically Opus 5 .5 for basically. 95 to 99 % of their video editing. And so this video right here got 1 .7 million views.
It's only 10 seconds, but look at the quality of this motion design by Opus 5 .5.
And it loops perfectly. Here's another example, same guy.
And so the point here is that Opus 5 .5 is just getting really good at creating videos with code. It doesn't even need to use like Premiere or DaVinci or even After Effects for that matter, it can literally create videos with just JavaScript, like pure code.
And the type of videos I'm starting to see is absolutely insane. And I'll be making a dedicated video on this in a few days. And so the third update of the week comes from Anthropic again, and that is Claude projects is rolling out more broadly.
So for the past, I think a month and a half, I've been talking about Claude. projects or Claude code projects. And so they just posted a four minute walkthrough to get started.
So Claude projects is a new type of project that is up here at the top and you should have access now if you're on the paid plan and all Claude projects is, is if you go to projects, if you click on any of the projects, here's the one that I use to, um, basically maintain my website. And so this is the main. orchestrator, or I think they call it a coordinator and then it has many different threads.
So I've been using this for about a month and you see we've have, we've created over 35 threads in this project and all of these are created just by talking to the coordinator. So if I were to say, Hey, can we make the text in the weekly updates better and just Make it look better.
This isn't a real update I want to do, but the point is, is this will automatically redirect this request to a different agent. Look, it said sent to thread on it. I've asked weekly updates calendar, which is the name of a thread.
And now it is working on it in this chat right here, and I can full screen this. And now we're in a thread within a project and all I do. on my phone as I talk to the coordinator.
What's really cool about this is some of these threads are local and some of them are running on my computer. As you can see here, This one is a local thread.
These are two local threads. All of these other ones are running in the cloud. So even if your computer's off, you can run these from your phone.
And if your computer's on, you can control your local threads from your phone within this project. And it's really cool. You just open up your project.
Say your request and it will route it to the right thread. And the final update in the Claude section is that they released Haiku 5 .5 at the beginning of this week. If you come down anywhere inside Claude, let's open up a new chat inside Claude code.
If you go to the model selector, you will see a new model, Haiku 5 .5. This is the first Haiku model that they've released in a long time, and it is apparently better than Luna, which is the small model on OpenAI, and it's pretty good. But if you're on the $20 per month plan, this will probably make sense for a lot of tasks that you do, just to conserve those tokens because it is getting better.
These smaller models at Frontier Labs are getting... So moving on to OpenAI, the fifth update this week is that we are entering the 28 days of Codex. Let me explain.
Thibaut from OpenAI said over the next 28 days, each day will either ship one thing that is a clear improvement and relevant for most Codex slash work users or ship a full reset, meaning you will get way more credits. Let the improvements begin. So far, we've gotten...
Codex predictions that they call composer predictions, where it basically predicts what you're going to type next. Somewhat useful. They've also added an improved steering to be instant.
And so all that means I could type something into Codex here and I could run this prompt and let's say I actually meant a different app. Wait, I actually met the agent native landing page. This steering is basically all steering is, is basically.
entering a prompt while the AI is working and now steering is immediate. It used to wait a little bit for the next tool call and it could basically waste a ton of tokens before it actually read your steer but now the steering is immediate. They also released their brand new meeting notes feature and they also released the decisions API which is a direct competitor to Jev which is the brand new AI decisions model, which I've made a full video on a couple videos back.
It has around 200 ,000 views if you want to check it out. And then on day one, they optimized it to be 50 % faster from when it was on dev day. So it's getting much faster.
For the next update, OpenAI today released a new feature called Intelligent UI. As you can see here, I just said inside just the chat GBT app, I said, use Intelligent UI, make a calendar. And then it just made a calendar in line.
I can do another one. Hey, I'm thinking about buying a house in Brooklyn. I want to buy an apartment.
Please, I want to figure out what I can afford. Please use Intelligent UI and help me make a quick graphic for this or UI. And so it can just spawn the perfect UI for whatever it is that you're asking in line, which is pretty damn cool.
And as you can see here, it will just make a little mini app. in line that automatically adjusts. And I've been talking about this for a while about how apps are dying a slow death because you can just spawn an app like this in line and you can share this conversation with someone else and they could use it as well.
So that's the premise behind Intelligent UI. The next one is a little bit over my head, and that is OpenAI published a GitHub repo with a ton of math manuscripts. Apparently AI is really smart now.
And it solved a ton of math problems that go way over my head that you can find on github .com slash open AI slash math. And they solve the ton of really hard problems. And many people are saying that it's mind blowing.
And they solve so such hard problems that a lot of the scientific community are revolting against this. And they want this not to be allowed. And so that's how insane this is.
So if you're smart enough to comprehend this, check it out. Moving on to SpaceX, the eighth. Update of the week, and arguably the biggest update of the week, is that Grokbot now uses all models.
By the way, on agentnative .inc, this is in the description. You can actually find, if you go to weekly updates and you select this week, you can see all of the links that we're referencing in this video. And you can come down here, click on it, and we can find Elon Musk's tweet right here.
He said, important note regarding Grokbot. Going forward, SpaceX will use the best... backend model for any given task, including Cloud Opus 5 .5, Mid Journey, Suno, and other leading APIs.
What this means is that Grokbot won't just be using Grok. I've been talking about how Grokbot is the best UI so far for creating workflows with agents and talking to agents. And now if the task seems pretty hard and the AI thinks that Grok 4 .7 won't do that well.
It'll route it to Opus 5 .5. It'll always choose the best model when you're using Grok bot. It honestly makes Grokbot arguably the best personal agent platform right now between Grokbot, Muse, and DOTS.
By the way, on agentnative .inc, this is in the description. You can actually find, if you go to weekly updates and you select this week, you can see all of the links that we're referencing in this video. And you can come down here, click on it, and we can find Elon Musk's...
Tweet right here. And then the ninth update is just a little summary of my latest video or two videos ago where I talked about how on Grok bot they have a new type of bot called a team bot. You can see that your bot is a team bot in the side panel.
It'll say team and you can actually add this to Slack. If you open up its side panel right here, you can see that it is connected to Slack and now I can add plugins and skills and my whole team can use. this team bot.
The next one is a primary bot. You can see that it's a primary bot, but this little star, this is like a chief of staff. Remember Claude projects from earlier, how you have a coordinator and then a bunch of like sub agents or threads.
That's kind of how this works. And so I want you to watch what I, uh, what the first instinct of this team leader bot is when I ask it to do something. Hey, I actually want to work a little bit on making the dark mode more aesthetic on native note.
Can you please make some changes on that? and make a PR. And notice here, it's not going to start working on it.
Its first instinct is going to be to delegate it. I'm asking native note dev to polish dark mode. You can see it messaged this native note and you can see that the native note dev is working.
These are all agents and this is kind of the main agent. This agent delegates and you can see that it's this type of agent because it has a little star. Okay, moving on to Google.
So the 10th biggest update this week is Nano Banana. So we're back on the Agent Native site. If we click on this right here, we can see the Nano Banana tweet.
Nano Banana 2 .1 is here. This upgraded version outperforms our previous models across the board with notable... Leaps in visual design mask based editing subject consistency and more natural looking imagery I get a little bored of image generation because it just feels solved at this point based on people I've talked to I do know that it is faster than GPT 2 .1 or whatever the latest GPT model is I do know that the best model is still the GPT model, it is still slightly better than the nano banana, but this one is a little bit faster.
But if you have a Gemini account, this is useful. You'll probably get a ton of credits to use this feature there and you can edit images and it's really good. And the 12th update or the 11th update rather of this week is that there's more rumblings of a new model coming from Google called Argon.
And for those of you who don't know, I talked about this a little bit last week, Google's releasing a new model, which is Gemini for Argon. And apparently it's better than Opus 5 .5 at a lot of different things. Again, these are just rumors and this is what people are saying on Twitter, but keep an eye out for this.
And that takes me to the 12th one. If there was a tool that you wanted to use in preparation for this, I highly anticipate that anti -gravity is the tool where you're going to get the most usage out of argon. I highly recommend looking into using anti -gravity, at least understanding it or maybe getting it downloaded because just like with Codex or Claude, you get the most usage when you use their tools.
And if you want to test this model or push it to the limits, I would have anti -gravity downloaded. And if you have a Gemini business account, it shouldn't be hard to get some credits once this model releases. And I've already talked about 14 and 15 or 13 and 14, which is AI video editing.
And I've kind of already talked about it up here. Everyone I trust who is very skeptical about AI is diving in and they are saying like, oh my God, this is way better than I thought. It can basically create any type of animations.
It can do color corrections. It can do sound design. They're starting to learn how to use DaVinci Resolve and Premiere Pro.
And so I highly recommend just looking into AI powered video editing, especially Opus. 5 .5. This model, I swear, took psychedelics because it's so visually creative.
And a lot of creators who were honestly anti -AI about a year ago are starting to come around and using this technology. So for the next three updates, I want to discuss a little bit of a different topic. And that is, is software actually dead?
The first update within this is that people are completely recreating the Adobe products. And this person tweeted, AI is now being used to recreate Adobe software. A developer has released seven free open source apps as clean rune alternatives to Adobe's main tools.
And you can see PhotoCraft is Photoshop, VectorCraft, Illustrator. All of them have craft at the end. And I'll put this link in the Agent Native weekly update.
But you will be able to find the GitHub repos. And I've actually downloaded. I think film craft and I tested it and they're still adding features, but it is pretty good.
And I'm starting to see this more and more that people are just using AI to like create direct clones of all of these software applications. And many people are starting to find it ironic that all of these tech companies got rid of people at their company and replace them with AI. And now people are using AI to replace.
the tech companies themselves and this is a very interesting trend and then of course the 16th update is that video games are just being recreated right now and merged with each other and let me show you what i'm talking about people are successfully reverse engineering games and so here is a clip this is call of duty inside of minecraft and you can also play skate Here we have zombies mixed with Fortnite.
Here we have, I think this is NHL mixed with zombies. And then the 17th update is that you can now play all of these video games directly inside your browser. This guy, Radiant Optimizer, you can now play these games in your browser.
Your PC does not need the game installed. You can play Black Ops Zombies, Skate 3, Halo, and Modern Warfare 2. And you can even play it multiplayer.
So gamers are really excited about this. This is the first time I've seen the gaming community super excited about AI. Because AI was one of the main reasons why all these games are...
playable because they're using AI to basically reverse engineer the game. I don't fully know what that means, but apparently AI can just hammer away and figure out how to like get the game running on whatever machine. And you can even run PS5 games on your computer as well.
And then to close up our weekly update, we just have some more agent products coming out and we are going to continue to see more agent products. There was... Basically a new agent product that's like Muse, Grokbot, and Dots.
And it's called Hark or Hark Pro. And this is a personal agent. We also have a new one called Tab.
They raised like $200 million to compete with Instinct. And then there's another one by Zach Yadigari who is releasing this like bracelet that is your personal agent. And we're going to continue to see so many more of these personal agent platforms because like I've been saying for a while, agents are replacing software.
It's kind of the final frontier. You saw with chat GPT, it can create these generative UIs and all chat GPT is now is an agent. And so we're seeing this convergence around this product that looks a lot like dots, which is like this agent with a computer that can just spawn apps in line, right?
It can just create whatever app you might need. It'll just create it while talking. So it does make sense that we're going to see this mass influx.
to a new type of product, which is this personal agent platform and work. This is going to be the most competitive space in the world. And I find it unlikely that anyone's going to be able to compete with open AI meta and SpaceX, but.
I will continue to share. I think another thing to mention here is Hermes agent just raised 90 million. They're doing incredibly well.
They're like the open source, open version of like rock bot basically. And they're going to be releasing an iOS app. And I'll be talking about that more as the open models get better.
This may actually be the cheapest way to use agents or at least use an agent that allows you to switch to any model, which I think is really valuable. And so there's going to be a ton of new agent platforms coming out. If you.
Go to agentnative .inc and you go to landscape. You can see all of the different tools. So not only can you see all the different personal agents and the self -hosted personal agents, you can also see the different types of agent tools.
You have your vibe coding tools, your company agents that kind of operate in Slack. You can give your agent an email, a credit card, a phone number, iMessage. And you can come here if you want to learn about all of the best.
agent tools in the world. This is on agentnative .inc. This is my company.
I'm releasing a lot of fun things in the very near future. So anyway, that's an agent update. Thank you guys for watching.
All of the updates are up here in this weekly updates tab. If you click on this week, any link you may want to find, you can find it here. Thank you guys so much for watching.
I'll see you here in the next one.
The Hook
The bait, then the rug-pull.
The promise is simple: Claude just shipped its biggest Claude Code feature yet, OpenAI opened a new door into the ChatGPT app, and GrokBot started routing hard tasks straight to Claude Opus. What follows is a 20-item sprint through all of it, checked off a hand-drawn corkboard in real time.
Frameworks
Named ideas worth stealing.
04:01model
Claude Projects: coordinator + threads
One coordinator thread receives the request
Coordinator routes it to the matching specialized thread
Thread can run locally or in the cloud
User can steer any thread from desktop or phone
A single project holds one coordinator and many purpose-built threads; talking to the coordinator alone is enough to get routed to the right thread.
Steal forany multi-agent setup where a human shouldn't have to pick which agent handles a request
11:28concept
GrokBot bot types: team vs. primary
Team bot: shared across a group, connects to tools like Slack
Primary bot: orchestrator that delegates to specialized bots
GrokBot's two bot roles split group-wide utility (team bot) from task-routing (primary bot), mirroring the Claude Projects coordinator pattern.
Steal fordeciding whether a new bot should do the work itself or hand it off
CTA Breakdown
How they asked for the click.
VERBAL ASK
10:00link
“agentnative.inc — every link referenced in this video, filed under this week's weekly update”
Mentioned repeatedly across multiple chapters rather than pitched once; a recurring soft plug for the host's own aggregator site rather than a single hard sales CTA.
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
A week where xAI undercuts Anthropic and OpenAI on price, and GrokBot's named-agent structure hints at how every AI platform is about to fix the chat-versus-work problem.
A 22-minute weekly roundup covering the open-source model that finally passed the vibe check, Codex screen-to-skill recording, the SpaceX/Cursor acquisition, and two Claude updates.
A 26-minute breakdown of OpenAI’s Intelligence at Work event: Codex merges with ChatGPT, role-specific agent plugins become startup killers, and Sites turns vibe coding into hosted apps.
A two-hour, filmed-in-one-day walkthrough of everything Codex can do with GPT-6 Astra, from one-prompt iOS apps to an agent that reads your Slack and tells you what to work on.
00:19
00:32
01:10
01:26
01:41
01:57
02:12
02:28
02:44
02:59
03:30
03:46
04:17
04:33
04:49
05:04
05:20
05:35
05:51
06:22
06:38
06:54
07:07
07:25
07:56
08:12
08:43
08:59
09:14
09:45
10:17
10:32
10:48
11:04
11:19
11:35
12:06
12:22
12:37
12:53
13:09
13:24
14:11
14:42
14:58
15:14
15:29
15:45
16:00
16:16
16:32
17:19
17:50
18:21
18:37
18:52
19:08
19:39
19:55
20:10
20:42
25:21
22:18
26:18
26:45
21:08
1:54:47