Opus 5.5 Is JUST the Beginning. Things Are About to Get Crazier.
Five days into Claude Opus 5.5, one builder rebuilt Notion, MyMind, and Typefully himself, then laid out what OpenAI is rumored to answer with at Dev Day.
Claude Opus 5.5 is cheap and reliable enough that solo builders can replace bloated SaaS subscriptions with narrow, custom tools built in hours, and OpenAI is expected to counter at Dev Day with an always-on agent and a $500 tier.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
A solo founder or small-team operator paying for multiple SaaS tools (Notion, Typefully, bookmarking apps) who only uses a sliver of each one's features.
Someone already building internal software with Claude Code who wants a real example of an agent-native, self-hosted tool replacing a stack of subscriptions.
An AI-news follower who wants a grounded rundown of the OpenAI Dev Day rumors before the event: GPT-Cyber, an always-on agent, a $500 plan, ultra-fast mode.
SKIP IF…
You want a hands-on tutorial with exact prompts and code; this is a first-impressions reaction video, not a build-along.
You're looking for confirmed OpenAI announcements; everything about Dev Day here is pre-event rumor and speculation.
TL;DR
The full version, fast.
Riley Brown has used Claude Opus 5.5 in Claude Code for five days and calls it the first model fast, cheap, and trustworthy enough that he's ripping SaaS tools out of his business. He walks through an internal app he built in a few hours that replaces the narrow slices of Notion, MyMind, and Typefully he actually uses, including comment threads with drag-and-drop media, shareable links, and an agent-native skill that lets Claude Code write directly into it. He then previews OpenAI's Dev Day, expected to counter with a security-focused GPT-Cyber model, an always-on personal agent to rival Grok's Grokbot and Meta's Muse, a rumored $500-a-month plan, a Cerebras-powered ultra-fast mode, and possibly new hardware.
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.
Riley Brown opens saying Opus 5.5 blew his mind after five days of daily use, then previews the two-part structure: Opus 5.5 impressions first, OpenAI's Dev Day predictions second.
00:31 – 01:57
02 · Opus 5.5 First Impressions
Describes using Opus 5.5 in Claude Code without hitting a single limit, calls it fast/cheap/trustworthy, and credits the new Projects feature for making it easy to spin up the internal tool covered later.
01:57 – 05:14
03 · Replacing SaaS Tools
Argues most SaaS apps only get used for a narrow slice of their features, cites a viral tweet claiming most SaaS apps can be built in hours, and walks through why he's replacing Notion, MyMind, and Typefully, focusing on the specific commenting workflow he can't get anywhere else.
05:14 – 07:11
04 · Custom App Demo
Live demo of the Vercel-hosted app he built: comment threads with drag-and-drop image/video attachments, no file-size limit, shareable public links, and one-click zip export of all attachments.
07:11 – 11:19
05 · Agent-Native Workflows
Shows the app is agent-native: Claude Code writes into it directly through a project skill called 'native note.' Also has Claude turn the same designs into a Figma-style board and a feature-deck presentation, tying Claude Code's Projects, co-work, and design modes into one experience.
11:19 – 14:32
06 · OpenAI Dev Day Rumors
Sets up part two: OpenAI's Dev Day is hyped after a disappointing prior year. Covers the GPT-Cyber security model and vulnerability-patching product, and 'O'/'Aeon,' a rumored always-on personal agent meant to rival xAI's Grokbot and Meta's Muse.
14:32 – 19:31
07 · $500 Plan & Ultra-Fast Mode
Predicts a $500-a-month plan following the $20-to-$200 pricing trend, an ultra-fast inference mode via a Cerebras partnership (demoed 10x faster on Codex), a cheaper Astra Minor model, and improved managed agents.
19:31 – 21:26
08 · Stretch Predictions
Lower-confidence predictions: a next-gen GPT 6.1-6.5 'Astra' model, and a new OpenAI hardware device possibly influenced by Meta Connect's AI keychain and led by former Apple designer Jony Ive.
21:26 – 22:27
09 · Main Takeaways
Closing stance: bullish on infrastructure plays like Convex and Vercel and on builders making their own software, bearish on thin SaaS wrapper products; previews live Dev Day coverage.
Atomic Insights
Lines worth screenshotting.
Claude Opus 5.5 is fast, cheap, and reliable enough that one creator says he's now confident ripping SaaS tools out of his business to build replacements himself.
Most SaaS subscriptions only get used for a narrow slice of their features, so a custom tool built around just that slice can replace them entirely.
A solo builder rebuilt the specific comment-and-attachment workflow he relies on in Notion, plus a bookmark manager and a script tracker, as one Vercel-hosted app in a few hours.
The app supports drag-and-drop image and video attachments in comments, shareable public links, and one-click zip export of all attached assets.
Claude Code can write directly into a custom-built app through a project-specific skill, turning a personal tool into something an AI agent can operate on its own.
OpenAI's last Dev Day was considered a disappointment with no major launches, raising the stakes for this year's event.
A rumored OpenAI model called GPT-Cyber is expected to ship with a product that automatically finds and patches security vulnerabilities in AI-built software.
OpenAI is rumored to be launching an always-on personal agent, internally called Aeon, with its own profile, DMs, and groups to compete with xAI's Grok and Meta's Muse.
Meta is treating its Muse agent as its most important product outside Instagram, and it will power the Meta Ray-Ban glasses.
AI subscription pricing has climbed from $20-a-month plans to $200-a-month plans in roughly two years, and a $500-a-month tier is rumored to be next.
OpenAI is rumored to be building an ultra-fast inference mode with Cerebras that reportedly ran Codex about 10 times faster than GPT 5.6 Sol in a demo.
Faster, more expensive models mean more prompts fired per hour, which is the actual argument for needing a $500-a-month plan rather than raw model capability.
Astra's viral momentum reportedly stalled because users ran out of weekly credits within a single day of heavy use.
Rumors point to Astra Minor, a cheaper version of OpenAI's viral Astra model, spotted under a model ID in Azure.
OpenAI is rumored to be improving managed agents so companies can package an agent with specific skills and plugins and deploy it as a reusable internal team.
OpenAI has reportedly been building a hardware device with former Apple designer Jony Ive, and Meta's new Tamagotchi-style AI keychain may be adding pressure to ship it.
The creator's bottom line: he's bullish on infrastructure plays like Convex and Vercel and on builders making their own software, and bearish on thin SaaS wrapper products.
Takeaway
Why One Builder Fired His SaaS Stack
AI Tooling Shift
When a model gets fast, cheap, and reliable enough to trust without checking, the SaaS subscriptions you only use 10% of become the easiest thing to replace yourself.
02Opus 5.5 First Impressions
Opus 5.5 in Claude Code held up for five days straight without hitting a usage limit, which changed how confidently the creator used it.
The new Claude Projects feature made it easy to multitask across chats, and it's what he used to spin up the internal tool covered later in the video.
His shorthand for evaluating a frontier model is a simple three-part test: is it fast, is it cheap, and does he trust its output without double-checking.
03Replacing SaaS Tools
Most SaaS tools only earn their subscription because of one or two features you actually rely on, everything else in the product goes unused.
Notion's dealbreaker feature for him was commenting directly on a script line and attaching images or video with no file-size limit, something Google Docs still can't do.
Feature bloat that helps power users can actively hurt onboarding, since new hires have to learn a sprawling tool before they can use the one feature that matters.
Treat 'you can build most SaaS apps in a few hours' as directionally true for your own narrow use case, not as a literal claim you can clone the whole product.
04Custom App Demo
He rebuilt the specific Notion commenting workflow, plus a MyMind-style bookmark collector, as one app and had it live on Vercel without even buying a domain.
Comments support drag-and-drop images and video with no size cap, a one-click shareable link for outside collaborators, and a zip export of every attached asset.
Getting from idea to a working, shareable internal tool took about 20 rounds of prompting back and forth, not a single one-shot generation.
05Agent-Native Workflows
The app is wired so Claude Code can write into it directly through a project-specific skill, letting the agent add new entries without a human touching the UI.
Asking Claude's desktop app to turn the same designs into a Figma-style board and a feature-deck presentation produced both artifacts in one pass, no separate tools needed.
Claude Code's new Projects tab now folds coding, a spreadsheet-style co-work mode, and a Figma-style design mode into one experience instead of three separate products.
06OpenAI Dev Day Rumors
Last year's Dev Day under-delivered, which is part of why expectations and hype are unusually high heading into this one.
GPT-Cyber is rumored to pair a security-focused model with a product that automatically finds and patches vulnerabilities in AI-generated software, a direct answer to 'vibe-coded apps aren't secure.'
An always-on personal agent, reportedly code-named Aeon, is rumored to get its own profile, DMs, and groups to compete directly with xAI's Grokbot and Meta's Muse.
07$500 Plan & Ultra-Fast Mode
AI subscription ceilings have climbed from $20 to $200 a month in a few years, so a rumored $500-a-month tier is a continuation of a trend, not an outlier.
A partnership with chipmaker Cerebras is rumored to power an 'ultra-fast mode,' demoed running Codex roughly 10 times faster than GPT 5.6 Sol.
Faster models mean more prompts fired per working hour, which is the real driver behind needing a more expensive plan, not just raw per-token pricing.
Astra's viral use cases reportedly dried up because heavy users burn through a full week of credits in a single day, creating rumored demand for a cheaper Astra Minor.
08Stretch Predictions
A next-generation 'Astra' model in the GPT 6.1-6.5 range is a lower-confidence stretch prediction for Dev Day, not treated as a near-certain rumor.
A new OpenAI hardware device is considered plausible given the Jony Ive partnership, though it's framed as a coin-flip on timing rather than a Dev Day lock.
Meta's new Tamagotchi-style AI keychain from Meta Connect is read as competitive pressure that could push OpenAI to reveal its own hardware sooner.
09Main Takeaways
His closing stance: bullish on infrastructure and platform companies like Convex and Vercel that let anyone build software, and on builders making their own tools with AI.
He's bearish on thin SaaS wrapper products, arguing they get squeezed once building the underlying tool yourself becomes this easy.
Glossary
Terms worth knowing.
Claude Code
Anthropic's terminal-based coding agent that can read and write directly to a codebase or, as shown here, to a custom-built app.
Claude Projects
A workspace feature in Claude that groups related chats, files, and generated apps together, making it easier to multitask across separate builds.
Agent-native
Software built so an AI agent can read and write to it directly through an API or skill, not only through a human-facing interface.
Vibe coding
Building working software mostly by prompting an AI model, without manually writing most of the underlying code yourself.
GPT-Cyber
A rumored OpenAI model reportedly focused on cybersecurity, including a product that automatically finds and patches vulnerabilities in AI-generated software.
Astra
OpenAI's current flagship coding-focused model, described as powerful but expensive and quick to exhaust a user's weekly usage credits.
Cerebras
A chipmaker known for very fast AI inference hardware, rumored to be the partner behind an OpenAI 'ultra-fast mode.'
Managed agents
A rumored platform feature letting a company package an AI agent with specific skills and plugins and deploy it as a reusable internal tool.
Resources
Things they pointed at.
00:44toolClaude Code
00:00toolClaude Opus 5.5
03:00toolNotion
03:00toolTypefully
06:17toolMyMind
05:26toolVercel
21:30toolConvex
15:33toolCerebras
12:31productMuse (Meta)
12:31productGrokbot (xAI)
Quotables
Lines you could clip.
00:44
“The only way I can describe Cloud Opus 5.5 is that with it, I glide.”
tight, visceral one-liner on model quality→ TikTok hook↗ Tweet quote
01:58
“Opus 5.5 is the first time I felt confident enough to start ripping a ton of SaaS tools out of my business and building them myself for my team.”
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
metaphor
Guys, I can't believe how good Opus 5 .5 is. I have been using it for five days straight and it is simply blowing my mind. I have yet to hit a single limit and it feels free to use.
So we're going to talk about Opus 5 .5 first. But I also want to talk about OpenAI because they never let a launch like Opus 5 .5 go unanswered. Dev Day is in like 40 hours and they're expected to release a ton of new stuff.
So that's what we're going to talk about second. Okay, so we have a lot to cover. First thing I want to talk about are my thoughts on Opus 5 .5.
And in case you guys don't know, if you go to your cloud... desktop app, and you look at the model selector, you will see a new model, Opus 5 .5. I'm using Opus 5 .5 in Cloud Code.
And after five days of using this new model, the only way I can describe Cloud Opus 5 .5 is that with it, I glide. So not only does Opus 5 .5 feel free, and it is just insanely cheap to use, at least for now in the Cloud app. It's also insanely fast.
And so the way that I've been using Opus 5 .5 is in their brand new projects feature. It makes multitasking really easy. And this new feature is really cool.
And so this projects feature is exactly how I created my internal tool that we'll get to later. That is a mixture between Notion and MyMind and Typefully. Tracks all my brand deals and Google Drive.
My entire team uses this software and I created this entire app in just a couple of hours using Opus 5 .5. This model is fast. It's very good.
And I just fully trust this model. And so this is kind of the trifecta right here for Opus 5 .5. It's very fast.
It's very cheap and I fully trust it. I have not given it one single feature request where it failed and it's incredibly efficient. And that's why I said this the other day.
I said Opus 5 .5 is the first time I felt confident enough to start ripping a ton of SaaS tools out of my business and building them myself for my team. Yes, I'm not going to vibe code an entire app like Notion, but what I am going to do is I'm going to vibe code one that's better for me and my small team. And what I'm realizing is every single app that I'm paying a ton of money for has a certain amount of features that's represented by the whole circle.
And only a small percentage of those features are features that I actually use. And so when people... like Nizzy, say something like this.
You can genuinely build most SaaS apps in a few hours. When people say something like this, I spiritually agree. While technically I can't rebuild Notion in a couple of hours, I can rebuild it functionally for my own business.
Let me explain what I mean. For the past two years, I've been using these three tools. And I've only been using a very narrow set of their features.
And specifically in Notion, let me show you the only reason I use Notion. In Notion, I like to write my script. So here's a script.
for a video that I've written and I love to be able to comment directly on a portion of the video and then upload videos or images to this, right? I can take an image right here and I can drag it into the comment and say, show this when I say, and then my editor can actually see where I commented this and I can upload an image, a video or anything here and there's no file size restrictions.
If Google Docs have the same feature, I would have switched a long time ago. So what I realized is notion has one single feature which has kept me there.
It has prevented me from switching to a different application. But what I've also realized as I build out my company and build up my team and bring on more people, they don't use Notion. And so I have to teach them Notion.
And since Notion has like a million features and it's pretty intimidating to use when you first start out, it's really hard to just teach everyone who joins my company how to use Notion. And so. all of these features are doing it a disservice.
It is becoming bloated. And so while I can't technically build out Notion in the sense that I can't build out all of the different features that Notion has, like link databases, and you can create these custom dashboards, and you can create a page within a page within a page within a page within a dropdown. It is just an insane piece of software at this point.
The only reason I'm using Notion is a very... tiny set of features. And so what I'm deciding to do is I'm just going to take those slivers of features from all the apps that I use, and I'm going to turn it into a single app that has all of those features.
And before today, I've always thought about doing this, but I'm like, oh, it would be annoying to maintain. And you know, there's going to be bugs. There isn't.
I did like 20 different prompts back and forth. I've tested it with multiple video editor agencies. There are share links.
Let me show you the app that I'm using, which is just hosted on Vercel. It doesn't even have an actual domain yet. I can just go to this.
This is this video that we're working on right now. Remember that comments feature that I liked? Well, look, there you go.
I can actually just, let's go ahead and copy this image. Look at this. I can very easily just drag images.
into the comment i can do the same thing with videos i could very easily upload a video to a comment and i added this on in one shot right i'm going to upload this mp4 this is a video being uploaded there you go look at this i can play it and then when i'm done with my video and i want to send it to my editor i just click this copy link button and look we'll go to an incognito window and we'll paste it in and this share link works.
And then my editor can read my script before he edits my video. He can come to this comment and he can just download all the assets that it needs. And guess what?
He can export all of the attachments as a zip file. Look at how cool that is. And I'm just, I'm flying now.
I added a lot more than that too, right? There's an app called MyMind, which collects all of my Twitter bookmarks and they make everything really hard to export. Well, my version of MyMind that I created in 10 minutes doesn't.
Every single day, this gets updated with all my bookmarks and I can save them. And this is kind of what jogs my memory and allows me to like come up with video ideas. I can see all of the different things that I save in my bookmarks and add them to MyMind.
I can also upload images or video. Additionally, every single comment, right? Every time I comment an asset on my script, it just shows up in this library and then it uses the linked line that I comment on it and it makes everything searchable.
So as you can see here, I said subscriptions in the line that I commented on and I can just type in subscriptions and there you go. Additionally, This app is fully agent native, meaning I can go to Cloud Code and I can say, hey, I have a new idea for a video, open AI dev day, coming up, add it to native note.
That's what it's called, right? Now we have a skill called native note and Cloud Code can write to that database directly. And you can see here, it's spun up a new thread and look, I can click on this link.
And it takes me to the script that it added. Here's a little script template. And you can see it was actually added to the list, right?
My agent can fully use the same software that I can use. So I want you to think about the implications here. With no problems whatsoever, I went to Claude and I said, here are all of the features that I use for Notion, MyMind, Typefully, Google Drive, and more.
And I was like, I want an app that does all those things that looks really good. And with no problems, and I mean no issues on my end, it created this application. Think about the implications of that.
And then I also need to give Claude some credit for their desktop app. I also said this. I said, please, can you turn the dark mode designs into a Figma style board for me to look at?
And then also in the style of the site, make a presentation on the features we've added and show the components of what we added in a presentation. And then it spun up a new chat and I can click on it right here. It spun up a new chat called dark mode designs and feature deck.
And then it created both of them. And look at this. I can click on this.
And here is a Figma -like board with all of the designs for this application. And I could very easily, I could say, hey, for the scripts list, I want to explore other possibilities. please make three other options.
And then AI will just create other options. And then I can just say like, add them to the board. And while that's loading, not only does it create this Figma style board, we also can create presentations.
And so this is a slide deck that goes through all the different components. So like what we built in five days, every video in one list, it writes blocks like in Notion, everything about a video is one. tab away.
And look at this, right? We can just see all of the features that it added and we can post from X. And so on top of Opus 5 .5 getting better, I've noticed that Claude has done a really good job of taking everything that we liked about Claude code.
along with co -work, which is creating spreadsheets, documents, and presentations, and then taking Cloud Design and kind of turning it into one experience, right? We're using Cloud Code's new projects feature right here. It's really easy to spin off new chats.
And this is design mode. This is the spreadsheets feature. And I was using that with Cloud Co -work a lot.
These used to be separate features. Now it's kind of all in one experience inside the Cloud Code tab. Really cool.
Oh, by the way, it is done. and we can see the other options. This one's not that great.
This one is classic Kanabon board. I do like this one, actually. We have this like dashboard above all of the scripts, which I don't mind as well.
And this whole experience is just like incredibly fast, which is amazing. So that's what I have been able to do. I have been absolutely blown away by how good Opus 5 .5 is, and I've been using it so much.
And everyone I've talked to, has basically said that this model is their favorite model to use. Maybe Astra is better than it at a good amount of things, but it's slower and way more expensive.
So right now it is well known that Opus 5 .5 is overall the best model in the world. And with that, we know that OpenAI is going to respond and they are going to respond with Dev Day and they are going to try and reclaim the lead. So that brings me to the second thing that I want to talk about today, which is OpenAI's Dev Day.
And I've never seen an event hyped this much by OpenAI. Last year's Dev Day was kind of a disappointment. They didn't really launch anything that cool last year.
And if you're skeptical of this Dev Day, I don't blame you. However, I've talked to a lot of people about this. This event is going to be insane.
And so I want to talk about the things that are likely to be announced and then some stretches about what they might announce. based on some rumors that are circulating around. And so the first rumor about Dev Day that's coming up is this new thing called GPT -Cyber.
And so GPT -Cyber is a model that is actually built for security work. And more specifically, It's actually rumored to come with a product that will help companies find and patch security holes.
And so there are a lot of criticisms about people who use AI to build internal software or build apps and sell it to other people. And one of those is that it's not secure. Well, AI is about to get a lot better at security.
And there's going to be products that automatically check for security vulnerabilities. And so it's about to get better thanks to OpenAI. Maybe again, these are all rumors.
Okay. Now I want to go to probably the most fun one. And I think this is the one that I'm most confident about.
And this is a new always on agent called O. So over the past few months, there's been really successful personal agents by the names of Grokbot was pretty successful. And one that was more successful was Muse.
So Grokbot was obviously by space. X AI and Muse was by Meta and Muse is now probably the most important product outside of Instagram to Facebook or Meta right now. Mark Zuckerberg is going all in on Muse.
Muse is also going to power their Meta Ray -Bans glasses and it is a very important product to Meta, especially because I know that Meta is very scared of what OpenAI is going to do with their personal agent as well. And I believe people are saying that it's called Aeon is it's like code name internally at OpenAI. And this guy says OpenAI Aeon leak.
Aeon is going to be more than just a chatbot. It could have its own profile, DMs, and groups. It's designed to compete directly with Grokbot and Muse.
And it could launch as early as Thursday. It also could launch at Dev Day. And that's what it's looking at.
And that's actually what people at OpenAI are kind of hinting on Twitter as well. So we are going to get... very likely, a Grokbot and Muse competitor for OpenAI.
And I'm very excited about this. If you've been watching my channel, you know I'm very excited about agent platforms. I want to be able to use AI agents as easily as possible, especially from my phone.
And I'll be covering this in great detail if it does indeed launch. Additionally, and some of you are not going to like this one. We are hearing rumors of a $500 per month plan.
I remember making a video on Instagram a few years ago, talking about the first ever $200 per month plan. This is back when the biggest monthly plan was $20. This is when vibe coding was $20 for cursor.
You had like unlimited tokens for $20 back then. And then I think it was Claude who released the first $200 per month plan very quickly. It was copied by everyone.
OpenAI, Cursor, Devon, I believe, everyone started getting these more expensive plans because these tokens are not cheap, right? These AI models are very expensive. And OpenAI has, I think, more models coming out on Dev Day, but also they have a new faster model coming out as well.
Because the next rumor about OpenAI is that they're going to be releasing ultra fast mode. Now, they have a partnership, a deep partnership, with a company called Cerebrus. And they're a company that's very similar to Grok with a Q.
And they've been teasing this new ultra -fast model, especially with GPT 5 .6 Sol. They showed a demo of using Codex. Using Codex with ultra fast and apparently was 10 times faster than GPT 5 .6 Sol was.
And all I'm thinking about during this is like, I cannot believe the implications. I cannot believe a world where we would have access to a model like Opus 5 .5, 10x the speed or even 5x the speed that we have it now. And that's what's coming very soon.
But if you imagine as these frontier models get more expensive and it gets much faster to use, that means per hour, you're just going to be entering more and more prompts. And that's one of the reasons why you're going to need a $500 per month plan. People are going to be spending more tokens using frontier models at 5x speed, especially if those GPUs are just like.
more expensive to use on a per prompt basis, and you're entering more prompts, and you're using the frontier model, we are going to need more expensive plans. And a lot of people, it's not going to be worth it, right? This is a very expensive monthly membership.
This is $6 ,000 per year just on AI tokens. And for most consumers, it doesn't really make sense. But for most businesses, it's an automatic.
It makes perfect sense. So they're going to be adding that. And so one of the effects of this is I just think Anthropic and Grok, SpaceX, they're going to copy it.
And I think $500 per month plan will become much more common. And the next two, I don't think are that cool. So I'll do them together.
And that is Astra Minor and Managed Agents. There's people that have been talking about how they were using Azure and they saw the model ID in Azure and it was called GBT6 Astra Minor. And this could actually just be a cheaper version of Astra that they're releasing on Dev Day.
And multiple people saw this. So we don't know for sure, but there's clearly a demand for Astra. Astra was one of the most viral models.
But if you notice on Twitter, all the viral use cases of Astra kind of stopped. Because everyone just ran out of credits. And now people are rationing because they've seen that if you use Astra for an entire day, you're going to be out of your weekly credits.
Where I've been using Opus 5 .5 all day, every day, and I haven't come close to my limits. So it's a very interesting situation going on. They will need a cheaper version of Astra to stay competitive or whatever model they're going to release.
Because their last two models, right? OpenAI released GBT, GBT6 Sol, and also... GPT -6 Luna.
And Luna was pretty cool because it's access to a pretty light model that's not expensive at all. And it's a lot faster. It's pretty good.
But GPT -6 Sol, it is, most people were very underwhelmed with it and think that it is much worse than Astra. And I haven't done too much testing just because I've been using Opus 5 .5. It's because it's just amazing.
So I'm excited to see how they respond. Because certainly they don't want the best model in the world after Dev Day to still be Opus 5 .5. And then they're going to be making improvements to their managed agents.
Again, these are all rumors. I'm hearing rumors that they're going to make improvements to managed agents, meaning it's going to be very easy for you to create like an agent wrapper. And you're going to be able to go to one place.
Create an agent that has specific skills and plugins, and you're going to be able to release these or use them internally within your company. You're going to be able to create your own managed agent team. Okay, so now let's move to the stretches, right?
The things that... Some people are talking about a little bit that we might see that could be kind of their, oh, one more thing moment at Dev Day. The first one is something like GBT 6 .1 Astra or something, or 6 .5, whatever they decide to call it.
Maybe they just come out with the next Astra model that's like super top of the line, best model in the world. So I'll put 6 .1 to 6 .5. I don't know if they will actually do this, but we'll see.
This one is less likely. Now, there's another thing that they've been rumoring and talking about, which is a new hardware device. If you guys saw Meta Connect, Mark Zuckerberg announced this new Tamagotchi -style keychain thing that you could hold.
You could also put it on your neck like the friend necklace that you could talk to Muse on. Now, this might have put pressure on OpenAI because they are 100 % working on a new hardware device that will be personal superintelligence. And they even spent billions of dollars on Johnny Ive, who designed a lot of the Apple's famous products.
OpenAI clearly wants to do hardware. The question has always been when. Is Dev Day going to be when they release their hardware device?
We will see. If they are releasing it, they're doing a really good job keeping it under wraps and keeping it a secret. I've heard no actual leaks from people around this, but it is something that is possible.
So... I'm excited for hardware. I'm actually very excited for hardware.
In fact, I just ordered a 3d printer and I'm going to get a laser cutter very soon. And I'm going to start building some hardware and using AI and creating little agent native hardware devices just to tinker and have fun with it. Because I think that's when the most fun business ideas come about is when you just explore.
And so that's what I want to do. And yeah, anyway, the main takeaways of this video, I feel incredibly confident in building any software with Opus 5 .5. I can't even begin to think what this means for the world of SaaS.
It's going to be tough out there if everyone's just building their own software. Very bullish on companies like Convex and Vercel and companies selling software that agents can use when you build your own software. But I'm very bearish on wrappers, creating apps and selling them.
I'm very bullish on building whatever app you want or whatever app you need for your company, just building it with AI. AI is getting that good. So that's kind of the main takeaway for part one.
Part two, we have a really fun day coming up. I will be covering this in very great detail. Every new update, I'll be making a long video covering whatever gets released on Dev Day.
Make sure to subscribe for that. I'm excited for this week. I think it's going to be insane.
And I will see you guys here in the comments. I'll be responding to every comment. So...
Hit me up down below. Talk soon.
The Hook
The bait, then the rug-pull.
Riley Brown says five days with Claude Opus 5.5 changed how he thinks about his own SaaS stack: it's fast, cheap, and reliable enough that he's already ripped Notion, MyMind, and Typefully out of his workflow and rebuilt them himself. Then he turns to what OpenAI is rumored to answer with at Dev Day, from an always-on agent to a $500 monthly plan.
Frameworks
Named ideas worth stealing.
00:44list
Fast, Cheap, Trust
Fast
Cheap
Trust
The creator's shorthand test for whether a frontier model earns daily reliance: it has to be fast, cheap to run, and trustworthy enough not to double-check its output.
Steal forquick internal rubric for evaluating any new AI model or tool before switching to it
CTA Breakdown
How they asked for the click.
VERBAL ASK
22:12subscribe
“Make sure to subscribe for that.”
Soft, single-line ask at the very end, tied to a promise of live Dev Day coverage rather than a hard sales pitch.
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.
One structured prompt into Claude Code's new Fable 5.1 turns into a Trello clone that the creator, his team, and three different AI agents can all update on the same real-time board.
A 37-minute conversation between Riley Brown and Ras Mic (Michael Shimeles) on why the model provider building its own tools is changing the AI coding wars.
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.
A weekly AI-agent roundup: Opus 5.5 beats GPT-6 Sol and Luna on Anthropic's own scorecard, ChatGPT's voice mode gets full tool access, and Meta's Muse quietly becomes the biggest agent story of the year.
One creator spent a week and $1,500 in credits with early access to OpenAI's new model and came away with seven reasons he thinks the agent era just changed.
00:21
00:54
01:15
01:31
01:49
02:23
02:39
02:56
03:10
03:30
03:46
03:55
04:20
04:45
04:54
05:11
05:45
06:02
06:18
06:35
06:52
07:26
07:43
07:59
08:16
08:33
08:50
09:07
09:26
09:46
09:57
10:14
10:31
10:48
11:05
11:38
11:55
12:12
12:29
12:46
13:02
13:19
13:40
13:53
14:10
14:43
15:00
15:17
15:34
15:51
16:08
16:24
16:41
16:58
17:15
17:32
17:49
18:06
18:22
18:39
18:56
19:13
19:47
20:12
20:25
20:33
20:54
21:11
21:44
22:01
22:18
25:02
37:08
1:54:47
25:21
23:59
21:08