Episode Categories:
Resources:
Join the discussion on LinkedIn: Got something to say? Let us know on LinkedIn and network with other AI leaders
Upcoming Episodes: Check out the upcoming Everyday AI Livestream lineup
Connect with Jordan Wilson: LinkedIn Profile
Start Here Series in our Inner Circle Community: Join for free access
The Latest in AI for Business: Codex Goal Mode, Gemini 3.5 Flash, and Productivity Breakthroughs
Staying effective in a rapidly evolving AI sector takes focus—especially as new features and tools launch weekly across platforms like Google Gemini, Codex from OpenAI, and Microsoft’s productivity suite. This summary outlines the latest AI updates specifically relevant for business leaders, tech decision-makers, and operational strategists, with a keen focus on measurable impacts and use-case clarity.
Google Gemini 3.5 Flash: Coding Benchmarks and Application Integration
Key SEO Keyword: Google Gemini 3.5 Flash
Google’s introduction of Gemini 3.5 Flash marks a shift in model hierarchy and practical utility. This model currently serves as Google’s flagship Gemini release and outpaces the previous Gemini 3.1 Pro in both coding tasks and “agentic” benchmarks, such as Terminal Bench 2 and MCP Atlas. It delivers four times the speed of its closest peers, raising the bar for companies leveraging AI for code-intensive workflows and long-horizon, multi-step agentic tasks.
The business-level impact is immediate: Gemini 3.5 Flash is now the default in the Gemini app and in Google Search’s AI mode, allowing broader exposure without the need for additional configuration. Notably, developers operating in Google AI Studio or Vertex, and those accessing through APIs, will notice significant speed enhancements.
While the “Flash” designation no longer equates to the lowest cost—losing the 20x cost-breakthrough of earlier versions—the speed and coding accuracy make it a strong alternative for companies prioritizing performance over lowest price.
ChatGPT for PowerPoint: Native Integration and Automated Reasoning
Key SEO Keyword: ChatGPT PowerPoint Integration
OpenAI now offers a direct ChatGPT integration within Microsoft PowerPoint, streamlining slide creation, editing, and review. This tool enables users to pull in source material from enterprise connectors—such as Gmail, Outlook, or SharePoint—transforming emails, documents, and spreadsheets directly into slides without manual copy-paste.
A significant business value is the addition of a reasoning feature that critiques slide narrative, flags weak storytelling, and anticipates probable audience questions. The outputs remain fully editable within PowerPoint, preserving brand guidelines and enforcing template uniformity—critical for regulated industries and large enterprises with established content standards.
This add-in is currently available across nearly all ChatGPT paid plans, with immediate applications for knowledge workers and enterprise teams looking to reduce production time on internal and external presentations.
Gemini Omni Flash: True Multi-Modal Video Input and Editing
Key SEO Keyword: Gemini Omni Flash Video AI
Gemini Omni Flash, introduced as Google’s “anything-to-anything” model, provides capabilities that extend beyond text-to-video generation. This technology supports the upload and conversational editing of video—enabling business users to adjust filmed content (such as altering background scenes or objects) without professional editing skills or intensive manual effort.
Distinct from other video AI tools that emphasize quality, Gemini Omni Flash’s primary strength lies in supporting video-in/video-out pipelines. This is especially valuable for marketing departments and creative teams involved in frequent campaign adjustments or localization requirements, who can now modify and iterate video content at reduced turnarounds and lower staffing costs.
Access is currently available to Google AI Plus, Pro, and Ultra users, with integration into YouTube Shorts in the pipeline.
Cursor Composer 2.5: Economical Coding Model for Long-Horizon Tasks
Key SEO Keyword: Cursor Composer 2.5 Coding
Cursor’s Composer 2.5 positions itself as a viable, cost-effective alternative in software development workflows. Built atop Moonshot AI’s Kimi K25 checkpoints, Composer 2.5 incorporates 25 times more synthetic training data than its predecessor. In benchmark tests—such as Terminal Bench and SWE Bench—it’s on par with leading models like Opus 4.7, but at roughly 10% of the operational cost ($0.50 input/$2.50 output per million tokens versus $5/$25 for Opus).
For engineering leads, this enables thorough code generation and debugging sessions with significant reductions in API expenditure. Organizations that have seen API budgets balloon due to high token consumption in other LLMs now have a practical, high-performance alternative for both background workloads and batch agentic tasks, especially if they are heavy Cursor IDE users.
Alexa Plus: AI-Generated Podcast Content on Demand
Key SEO Keyword: Alexa Plus AI Podcast
Alexa Plus’s recent expansion introduces on-demand, AI-generated podcast episodes mimicking the popular NotebookLM “deep dive” feature. Business users can request any topic, and Alexa generates a full audio episode using two synthetic voices that aim to provide comprehensive topical overviews.
Unlike NotebookLM, Alexa Plus does not require document uploads, instead sourcing data dynamically. This is positioned as a value-add for businesses wanting immediate topical briefings, though current quality and factual reliability may raise concerns for professional use cases. This feature is rolling out to Alexa Plus subscribers and Amazon Prime members.
Google Anti Gravity 2.0: New Agentic Coding Interfaces
Key SEO Keyword: Google Anti Gravity 2.0
Google’s Anti Gravity 2.0 launches as a suite of agent-first development tools, including desktop apps, CLI updates, and SDKs. However, business users should approach with measured expectations: despite heavy marketing, initial feedback has highlighted usability issues and confusion stemming from multiple overlapping products and deprecations (including the sunset of the open-source Gemini CLI for consumers in June).
Current implementations lag behind comparables like Codex, both in UI polish and agentic performance. Enterprise and engineering teams considering Anti Gravity 2.0 for production may benefit from a wait-and-see approach until stability and feature parity with competitors is established.
OpenAI Codex: Goal Mode, App Shots, and Remote Access for Enterprise Productivity
Key SEO Keyword: OpenAI Codex Goal Mode
OpenAI’s Codex suite is consolidating its leadership with practical, compound features tailored for everyday heavy users:
App Shots: By double-tapping command keys, users can instantly capture foreground application windows, including contextual text, making cross-app workflows seamless and reducing manual transcription across systems.
Goal Mode: Now available via desktop, this autonomous operation mode allows Codex to pursue complex, multi-step objectives independently over hours or days. For operational teams, this means sophisticated process automation with minimum ongoing supervision—critical for scaling high-value tasks.
Remote Computer Use: Codex can now execute tasks on locked computers, further improving unattended automation potential for distributed or after-hours operations.
Shareable Plugins: Enterprise teams can deploy and cross-share custom Codex plugins, directly addressing long-standing collaboration pain points in large organizations using AI assistants.
Mac users currently enjoy the fastest deployment and updates; Windows releases lag due to developmental complexities. For organizations focused on productivity gains, repeatability, and true software delegation, Codex’s newest suite of features offers measurable reductions in friction and increases in process speed.
Conclusion: Navigating AI Upgrades for Strategic Value
With leading models releasing updates weekly, AI decision-makers must filter for usability, deployment readiness, and enterprise fit. Codex’s integrated, accessible toolset stands out for those prioritizing cross-team productivity and automated autonomy. PowerPoint’s ChatGPT integration transforms document workflows, while Google’s Gemini 3.5 Flash presents a substantial upgrade for code-first operations. Caution is advised with bleeding-edge releases like Anti Gravity 2.0 and early-stage Alexa Plus features, which may not meet business-grade reliability standards.
Evaluating these updates within the lens of real-world application—cost efficiency, task automation, and operational resilience—keeps organizations positioned to capitalize on AI’s incremental but tangible performance improvements.
Topics Covered in This Episode:
- Gemini 3.5 Flash Coding & Speed Benchmarks
- Gemini 3.5 Flash in Google Search & Apps
- Gemini 3.5 Pro Release Timeline & Rumors
- ChatGPT PowerPoint Integration Features
- ChatGPT Contextual Slide Generation & Review
- Gemini Omni Flash Multimodal Video Editing
- Omni Flash vs Seeddance Video AI Comparison
- Cursor Composer 2.5 Coding Model Release
- Composer 2.5 vs Opus 4.7 Price & Performance
- Amazon Alexa Plus Podcast Generation Feature
- Alexa vs NotebookLM Audio Overview Comparison
- Google's Antigravity 2.0 Coding Platform Launch
- Gemini CLI Deprecation & Community Response
- OpenAI Codex Goal Mode in Desktop App
- Codex App Shots & Remote Computer Control
- Codex Shareable Plugins for Teams & Enterprise
Episode Transcript
Jordan Wilson [00:00:17]:
If you're a Google Gemini user, here's the good news. You have dozens of new AI features and upgrades to tackle available now, including Gemini 3.5 Flash. But here's the bad news. In order to find them all or make sense of them, you're gonna need, like, a road map, a translator, and also dozens of hours. That's not all that's new in AI this week. If you're an Opus user inside of Cursor, there's a new model you might wanna take a look at. PowerPoint's got a lot easier with ChatGPT. And speaking of chat g p t, it's bigger, stronger brother codex, again, released a handful of AI updates that I don't understand how they're this good and this far ahead of the pack.
Jordan Wilson [00:01:10]:
So if you don't have hours every single day to keep up with all of these new AI releases, because that's how much time it takes, because that's all I do, well, this is the show for you. On Fridays, we bring you our new AI Friday features where we give you the lowdown on some of the most important and impactful AI tools that you can use today and LLM upgrades that you don't have to wait around for, not coming soon. These are ones you can use today to start making your life a little bit better. Alright. Let's get into it. If you're new here, welcome. My name is Jordan. This is Everyday AI.
Jordan Wilson [00:01:46]:
It's a daily livestream podcast and free daily newsletter helping everyday business leaders like you and me not just keep up, but how we can use all of this new AI technology to get ahead to grow our company and career. It starts here with the unedited, unscripted livestream podcast. But to be the smartest person in AI, our website is your cheat code, youreverydayai.com. Go sign up for the free daily newsletter. We're gonna be recapping the seven AI features that we're going over in today's show as well as a whole lot more. So make sure you go check that out. And, hey. Live stream audience, good to see you.
Jordan Wilson [00:02:19]:
Kimberly joining from New York City. Jose joining from Santiago. Joe is codex vibing in the mountains. I'm codex vibing in Chicago. Sounds a little bit funner. So thanks for everyone for joining. If you do have any questions as we go along, I'll try to get to them, but let's get straight into it. There's a lot to cover this week, and let's start with Google 3.5 Flash.
Jordan Wilson [00:02:43]:
So, yeah, there was a lot. We didn't get a Gemini 3.5 Pro. I think that there's a reason Google did this. I'll tell you here in a second. So here's what it is, what's new, who has access, all that good stuff. So, Gemini three point flash is Google's new flagship model. Yeah. That's right.
Jordan Wilson [00:03:04]:
Their flash is the flagship model for now in the Gemini three point family, and it was announced at Google's IO. We're gonna be, recapping, probably three or four of the, the Google IO AI updates. The rest will, be sharing about already shared about in our newsletter. But, it's already it is shipped. It is available now. And the crazy thing is, well, it beats Gemini 3.1 pro on coding and agentic benchmarks while also running about four times faster than comparable Frontier models. So right now, it is the default model in the Gemini app. And also, so even if you're not a Google Gemini user, you're gonna take advantage of this new Gemini 3.5 flash.
Jordan Wilson [00:03:47]:
That's because it is the new default mode in AI mode inside Google search worldwide, and it is going to be powering Google's new Gemini Spark. We're not gonna be going over that today because hardly no one has access anyways. Alright. So who has access to this now? Well, like I said, it's live. Not just in the Gemini app, but also in AI mode and search if you use Google's AI studio or Vertex, any API, and anti gravity two, which we're gonna be going over today. Alright. So it is, an internal, sorry. The Gemini 3.5 Pro, is expected next month.
Jordan Wilson [00:04:22]:
So Google did say that next month is going to be the, Gemini 3.5 Pro release. Presumably, what we think is gonna happen. Right? And this is how it's kind of happened in the past around these timed big events. Usually, an OpenAI or a, anthropic or someone is gonna come out with a model usually in response and, you know, so, kind of to dethrone Google here. I don't know if that's going to happen. Right. I don't know if we're gonna see an open, you know, GPT five six. We've seen rumblings of that.
Jordan Wilson [00:04:53]:
If we're gonna get, you know, SONNET four seven or, you know, eventually an OPUS four eight or OPUS five, whatever it is. You know? So depending on the benchmark you look at, yeah, right now, a flash model, Gemini 3.1 flash. So, this is supposed to be the smaller, lighter, faster, cheaper version, although it's not really cheap anymore. But in some benchmarks, it's the best model in the world. Although, all around, is it? No. It's not. But, that's what we might see out of Gemini 3.5 pro, which is going to be coming later. So here is what Google says on their kind of marketing.
Jordan Wilson [00:05:29]:
So they say Gemini 3.5 Flash delivers intelligence that rivals large flagship models on multiple dimensions at the speeds you have come to expect from the Flash series. It is our strongest agentic encoding model yet, outperforming Gemini 3.1 pro on challenging coding and agentic benchmarks like terminal bench two, GDP BOW a a, and MCP Atlas, and leading in multimodal understanding. So, Google did share kind of the benchmarks, how this compares to both, the old version of Flash as well as Gemini 3.1 Pro and, Opus four seven in GPT 5.5. And you'll see here at least on the screen that I'm sharing, there's actually a handful of benchmarks. Right? I know it's not just about benchmarks, but there's a handful of benchmarks that actually the Gemini 3.5 Flash is the best in the world. Right? So as an example, MCP Atlas, which is multi step workflows using, MCP's, toolathon. There's a couple others, that this new, quote, unquote, small fast model is the best in the world. That will obviously change once Google announces Gemini 3.5 pro next month.
Jordan Wilson [00:06:45]:
And if we do get a, you know, g p t five five, SONNET four seven, right, whatever we may get from the other labs, but pretty big news. So, here's why I think it's gonna be useful. Well, you get Frontier, Frontier level coding and long horizon, agent work at the flash tier, speed. The the price no longer. Right? So I I think if you went back to an older version, I think it's like Gemini, you know, two point o flash or 2.5 flash. Right? It was essentially 20 times cheaper. You don't have that anymore. So, there's actually some things that's kind of going up in price, on the flash side if you look at previous flash models.
Jordan Wilson [00:07:26]:
So, with Gemini 3.1 flash, it is getting much more expensive. So, you know, no longer when you see the flash moniker, at least out of Google Gemini, do you assume, oh, this is a cheaper model. I know a lot of companies had really been using the flash series for that reason because it gave you a pretty high intelligence level and a very fast, you know, output. So you still have the speed of the flash, but you don't quite have the same price breakthrough at least if you're comparing it to previous generations of flash. So I still think it's gonna be, you know, valuable for developers and engineering teams who are running anything on the agentic side. You know, operators that are just running, you know, deep research, multistep workflows, things like that. So it's it's still a very capable model. I think probably the best, you know, overall, you know, I guess, outcome for everyday people is if you use Google's, AI mode, it's getting much better than it was before, because this was on the previous version of Flash.
Jordan Wilson [00:08:29]:
Alright. You you know, here's someone from, YouTube, funny name here, the permanent underclass twenty six, said the updates from Google hasn't made me want to, to try it over codex. Yeah. For me, I can get out of codex right now. And it is one of those things. When you talk about, you know, speeding, speed and price. Right? One of the big things with speed as well if you're just sitting there waiting. But now with codecs, when you can schedule things around the clock, some of the new updates now, I think certain things on the speed side are less important for me personally and I think for a lot of other power users as well.
Jordan Wilson [00:09:05]:
Alright. Let's go next. So next, we have PowerPoint. Just got a little bit better. That's because there is a new chat GPT integration. So here's what's new. This is a new native add in that puts chat GPT directly inside Microsoft PowerPoint to help create, edit, and review docs from natural language props. So you can pull source material from any of the connected services you have, which is the really cool thing.
Jordan Wilson [00:09:32]:
Right? So you get to use a lot of the power, of Chad GPT's kind of harness and tooling. It's not just using the model. Right? You can use all of your connected services, like, you know, Gmail or your Outlook or your SharePoint. Right? So you can pull in all of that, information, which is huge. So then, you know, slides can be built from your existing emails, documents, spreadsheets, right, without copying and pasting or having to manually go look that thing up. So it also includes a reasoning feature, that critiques drafts, flags, weak storytelling, and anticipates audience questions, not just slide generation. So this one, you know, just came out yesterday. A lot of these are fresh hot out the oven.
Jordan Wilson [00:10:14]:
So I haven't had the chance to try this one yet, but I'm actually super excited for some of those reasons. Aside from just having, you know, a a new chat g b t integration, inside PowerPoint, which is really nice. Just that new reasoning feature that critiques and flags weak storytelling, anticipates audience questions is really cool. This is a great, you know, use case or example of how it's so important to use a reasoning model day to day. Right? For the most part, most models are either reasoning or hybrid off the shelf, but this just goes to show you, right, with this integration, it's not just gonna spit out a bunch of, you know, hopefully, legible and and good text, but it's gonna pull from your day to day. Right? This is context engineering inside of PowerPoint, powered by one of the best harnesses around inside of ChatGPT. But also it well, it can kind of take a step outside of the work that you're doing and critique it a little bit. So here is what OpenAI says about the new integration.
Jordan Wilson [00:11:14]:
So it says it turns conversations into presentations in minutes. So it can create and edit in plain language. So you can start from notes, docs, spreadsheets, prompts, or an existing deck. You can ask Cheggi PD to create slides, rewrite content, tighten the hierarchy, and, add a new section or publish a draft while keeping the outputs editable in PowerPoints. You can get a clearer story from your deck. You can ask what your presentation says, where the narrative is weak, what's missing, and what an executive audience might ask. So you can use Chat GPT to summarize, restructure, and turn dense slides into clearer takeaways. And then you can follow along so you then you can trust the work.
Jordan Wilson [00:11:53]:
So OpenAI says that chat g v t works inside a PowerPoint, reads your deck structure, and helps preserve editable slide contents, review changes, check important claims and numbers, and keep control, that you share. Alright. So to install it, you are gonna wanna look in the upper right hand corner, the add in, the add ins, kind of toggle there and then search for ChatGPT, and then you will need to connect, your ChatGPT account. So this one is actually, pretty cool because I know that there's still a lot of people that maybe either don't have the experience or the expertise yet and kind of vibing slides. Right? So notebook out loud with nano banana is great for vibing slides. Right? The new GPT images too, great for vibing slides. People don't know this. I shared this on one of our shows.
Jordan Wilson [00:12:41]:
Right. You can create a codex skill that can kind of stitch all of those together so you don't have to do it, like, slide by slide. But here's, I think, you know well, first, who has access? So it's available now for a lot of paid chat GPT plans. So right now, it's available in beta for chat GPT business enterprise, EDU teachers, k 12, free go plus pro. So, essentially, just about every single chat GBT account. It does also obviously require a Microsoft account, so you have to have access to Microsoft PowerPoint, obviously. But here's why it's useful. So, I mean, it removes the copy and paste from other tools.
Jordan Wilson [00:13:20]:
Right? Kind of the, the human duct tape that I've been talking about for, so long. It keeps the slide slides editable in the need of PowerPoint structure, right, rather than just producing a flat export. Because, yeah, you might be looking at this and being like, wait, the new GPT images too, is absolutely amazing. Right? Why would I, you know, not just use that? Well, that gives you a flat JPEG. Right? And, obviously, in PowerPoint, there's gonna be some, you know, some things that you have to use in your, you know, your preapproved company template. Right? A lot of times you just get these decks. You have to use them. Right? There's all these different things.
Jordan Wilson [00:13:54]:
So, you know, just right there, you know, being able to adhere and stick to your company's brand guidelines, but being able to kind of take a step outside of it. Right? It's almost like having, you know, a junior consultant come in and they're an outside, you you know, version of yourself because they're, you know, poking and prodding, but also, being able to use some of the best features as, inside of ChatGPT as well. So, I mean, who's gonna find this valuable? Any any knowledge workers. Right? So if you are still kind of you know, I know a lot of people are, quote, unquote, forced to use PowerPoint. I know some people love PowerPoint by default. Me, I don't know why I'm still stuck on Canva. Right? I'm I'm PowerPoint is great, but I know a lot of, you you know, enterprise companies. I would say the majority of enterprise companies, when you're creating a deck, you still have to do it in PowerPoint.
Jordan Wilson [00:14:42]:
Right? Even if you wanted to, you know, vibe something up in one of these other tools. Right? So any knowledge worker, that is living inside of PowerPoint, but also that has data scattered across any of those other chat TPT connectors, this is gonna be a huge time saver. Right? This is why I think this Friday show is actually super helpful and important because chances are hardly no one saw this. Right? Because OpenAI yesterday shipped so much on the codecs end. This probably slipped under the radar. Right? Like, be be be honest, livestream audience. Did anyone know about this? Right? Did anyone know actually that came out for PowerPoint? Maybe Nisiani did. Right? Since she works at Microsoft.
Jordan Wilson [00:15:22]:
But I don't I don't know. I didn't see anyone talk about this. Right? I obviously have a lot of agents crawling everything twenty four seven. So when I saw this pop up yesterday, I'm like, Why is no one talking about this? Alright. Something people are talking about our next AI news story. A new well, technically, not a video model from Google, more of an anything to anything model as they're calling it. It is the new Gemini Omni flash. So, this is Google's, let me kinda give give everyone the visual there, for our livestream audience.
Jordan Wilson [00:15:56]:
This is Google's new multimodal, what they're calling the world model that they announced at IO. So it generates and edits high quality video from any combination of text, image, audio, and video inputs. So this combines Gemini's reasoning with DeepMind's nano banana VO and Genie lineage, with claims of better physics, gravity, and kinetic motion simulation, versus prior video models. So this also supports conversational editing of existing footage. Right? So you can upload a video and say, you know, change what's happening in this scene or change the background to make it, Chicago instead of New York. Right? So pretty cool. So here's who has access, and I'm gonna get to my thoughts on it here in a second. So Gemini Omni Flash, has already rolled out to paid users.
Jordan Wilson [00:16:50]:
So if you have a Google AI plus, Pro, or Ultra plan, it is available inside of the Gemini app as well as inside of Google Flow, and it is coming to YouTube Shorts, at some point, although we don't have a firm date on when. Alright. So here's a couple important pieces to think about. So Google well, they said this very directly. This is positioned as a family of models. So, obviously, the Flash name attached to it lets you know that there is going to be a bigger version, whether that's called OmniPro. We'll see. The other thing, I think people are kind of overlooking, the true value of this.
Jordan Wilson [00:17:31]:
People are looking at this as just a model that outputs video, which is not really true, but that's all really you're seeing online is people are just demoing. Right? Like, oh, here. I put in this text prompt and got this video. But, again, this is not a text input video output model. This is an anything to well, anything. Eventually, it'll be an anything to anything, but right now, it's kind of an anything to video model. And that's actually important and, an an important designation to talk about because a lot of people are right now comparing Omni to Seeddance. Alright.
Jordan Wilson [00:18:06]:
So Seeddance right now is the best AI, video tool out there. So from a quality perspective, Google Omni, Omni Flash at least, is not the same quality as Seeddance two. Right, a very popular, Chinese AI video model. But I think that's people are missing the point because this is video input, video output. Not only that, but, you know, eventually, my guess is it's gonna be able to output, world models. It's gonna be able to output, I don't know, Google Street View, you know, in you know, output CAD drawings. Right? So, I think if you just look at this as a text to video output and comparing it to CDANS and saying, oh, this new Google Omni flash stinks. I think you're missing the point.
Jordan Wilson [00:18:49]:
But I do think one thing that it is instantly the best at in the world is video in video out. And I think there's great examples of that that Google has shared already online that show if you whether you're using something for production or if you're just, you know, kind of playing around, to be able to upload a video, without any real technical knowledge and change anything. Right? Change the shirt I'm wearing. Change the person in this video, but keep everything else the same. You know? Like I said, if you I actually talked with a, a a client recently. They had a really good workflow on this using, I think it was a combination of seed dance and runway, but it was a lot of human hours. Right? And the quality was, oh, you know, it was good, but it wasn't okay. Right? They did an expensive video shoot.
Jordan Wilson [00:19:32]:
They had to say you know, they had to change some things on the back end. Right? The the the team was like, hey. This doesn't really fit our vision even though this is how we shot it. So they had to use a lot of these tools and a lot of, you know, human hours to, use these AI tools to change some things. But now it's much easier to do that piece with the new Omni model. So here is, kind of what Google says about it. They said we're introducing Gemini Omni, where Gemini's ability to reason meets the ability to create. Omni is our new model that can create anything from any input starting with video.
Jordan Wilson [00:20:05]:
With Omni, you can combine images, audio, video, and text as input and generate high quality videos grounded in Gemini's real world knowledge. You can also easily edit your videos through conversation. They said today, we're rolling out the first model in the Omni family, Gemini Omni Flash to the Gemini app, Google Flow, and YouTube Shorts. Like I said, YouTube Shorts is not out just yet. In time, we will support, output modalities like image and audio. And then they said, you know, one of the biggest things, like I said, I'll just leave with this and then we'll move on. But, the biggest thing is being able to edit your video. Right? Any any, video that you uploaded to be able to edit it conversationally and to transform the world around you.
Jordan Wilson [00:20:48]:
All right. So what do you guys think? And let me know if the Spotify comments as well. I like, I'm always at a loss. It's it's like, I don't know if the, you know, if the average, you know, knowledge worker is really interested in these kind of things. If, you know, marketing departments, creative, you know, people, like, should we be covering this more on the show? Should we stick to more on the, you know, agentic harnessing, you know, text input output getting work artifacts done. Personally, I'm very interested in this side. Right? In a previous light, I did a lot of photo and video. Right? I have, I don't know, five at least five DSLR, you know, high grade Canon cameras somewhere in a closet.
Jordan Wilson [00:21:33]:
Right? I used to really be into photo and video. So for me personally, I love this kind of stuff. But, you know, ultimately, when it comes to growing your business, let me know if this is something that you really care about or not. Alright. The next thing, maybe you care about this if you use cursor. Alright. So we don't cover cursor a ton on the show, but I think this one is actually pretty important for what it represents. So here's our Nox AI feature this Friday, Cursor Composer 2.5.
Jordan Wilson [00:22:03]:
So if you don't know anything, let me give you some background first. You know, Cursor is essentially an agentic IDE. Right? You can bring in a bunch of different models and, you know, you can buy code something. Right? That's a very simplified way to say what cursor is. Right? But think of, like, you know, codex or Claude code, but you can use any model. So So it's built by Cursor, but Cursor also has their own in house models. But until this new update, I don't think people were really using composer. But with composer 2.5, I think that could change.
Jordan Wilson [00:22:35]:
So, this is Cursor's in house coding model that they just released. It is built on top of Moonshot AI's, Kimi k two five checkpoints, but with 25 times more synthetic training data, compared to Composer two. So kind of the sweet spot is they're targeting long horizon agent decoding, multi file edits, terminal sessions, iterative debugging, and sub agent recovery. So here's some of the reasons why this is really important because for the first time, you know, we have a model from the coding and agentic side that is at least in the frontier, column. Right? For the most part, it's always just been, you know, g p the latest version of GPT, the latest version of Opus, and the latest version of Gemini Pro. And then there's usually a pretty big drop off with everyone else. So, composer is not ahead of any of those three models, but it's at least in the same league. It's in the same territory now, which is big, especially when we look at the price.
Jordan Wilson [00:23:39]:
Because on things like Terminal Bench, right, and SWE Bench, two of the more important kind of, agentic coding, benchmarks, It's literally neck and neck with Opus, 4.7, on those. It's way behind. Right? Terminal bench, g b d 5.5 is in another category. But Swee bench, multilingual at least, it's it's on par with, you you know, the big players, and this is really important. So this is live now. So who has access? It's live now for all, paid plans inside of Cursor, and they also doubled, included usage for subscribers for a week after launch. Here's why it's important. Alright? So you are getting a near, and I think a lot of people using Cursor are still heavy, Opus users, right, via, you know, anthropics model.
Jordan Wilson [00:24:35]:
One of the reasons is I mean, I feel fine saying this because the data backs it up. The harness of of of Claude is really bad. Right? When you compare it across, there's there's actually harness benches. Claude's is the worst. It's it's one of the actual worst harnesses or pieces of software to use for their own models, which is why I think a lot of people have used cursor and kind of a sweet spot. But the price for Opus is well, compared to this, it's now doesn't make sense. Right? So now with the new composer 2.5, it is 50¢ per 1,000,000 token inputs and $2.50 per 1,000,000 token output. So let's compare that.
Jordan Wilson [00:25:15]:
That's about a tenth of the price because on the Opus four seven side, that 50¢ input becomes $5 and that $2.50, output becomes $25. So it is literally a tenth of the price, but about, about, you know, depending on what metric you're looking at, it's about 90 ish percent of the capabilities. Right? Just a general broad overview if you're looking at benchmarks across the board, maybe 80 to 90%. So you're getting 80 to 90% of Opus capabilities at 10% of the cost. And this is one of the things not to go off on a side tangent. Right? So much on Anthropic's business model is people using their models via the API. Right? Their own, you know, tooling and rate limits and and harnessing is is fairly bad. So that's why I'm always, like, from a mode perspective, when you see things like this and when you see these Chinese models come out that are great at coding, but even something like cursor and composer 2.5, pretty big.
Jordan Wilson [00:26:15]:
So this is, you know, gonna be useful for people that are running, you know, long sessions that you need to stay coherent. You know, companies that are maybe having to cut back on, you know, anthropic token budgets. Right? Companies a lot of companies that are using anthropic are hitting their, you know, budget allocation for the year already or are already having to see themselves cut down on token spending. Right? The token maxing of, you know, quarter one is now showing up on the quarter two budget sheets. So I think that's gonna be a a key factor here. So engineering teams that are running heavy background, or batch agentic workloads are gonna like this. Any cost sensitive solo developers or small teams that have kind of priced themselves out of, you know, Opus four seven or, you know, GPT five five high, you know, or if you're just obviously a Cursor Pro subscriber. So, pretty pretty actually big news.
Jordan Wilson [00:27:07]:
Like I said, I know we don't cover Cursor a ton, but I think why it's a big deal is for anyone that's actually using anthropic, Opus four seven and aren't liking the the the bugginess and the uptime, right, in their desktop app or their CLI, or if you're just seeing all these codex features and you're like, wait. I want that, or you're just pricing yourself out. It's actually a pretty pretty big, pretty big update. Alright. Gotta get a drink of coffee there. Alright. Our next one, Amazon Alexa. Oh my gosh.
Jordan Wilson [00:27:42]:
There's an update. Are we actually gonna talk about Alexa here on the show? Yes. We are. So it's an interesting one. They're just kind of taking notebook l m's most popular original feature that made it go viral, the deep dive where it's two AI voices, having a great, podcast discussion, and they're trying to incorporate that into Alexa. So this is a new Alexa plus feature that was just launched, and it generates full audio podcast episodes from AI hosts on demand for any topic that the user requests. So here's specifically what it is. So it's two AI generated voices that cohost in a conversational back and forth similar to the, notebook l m audio, overviews, but they're not prompted from documents.
Jordan Wilson [00:28:31]:
So users can adjust the length, tone, and focus after Alexa Plus proposes an outline, then you can finish the episode and it's delivered as a notification to your Echo Show devices and then saved in your Alexa app. So here's the, you know, the downside is, well, you kinda you gotta have Alexa Plus. So it is already starting to roll out, in The US, and it's included if you already have Alexa Plus at no cost. So if you do have Alexa Plus, FYI, if you are an Amazon Prime member, if not, it is a monthly fee. So, yeah, if you're already paying, for, you know, Amazon Prime, which I think a lot of people are, like, I don't know how the world existed before you could, I don't know, get, like, Nespresso delivered to your door in, like, an hour. But a lot of people probably don't have Alexa Plus or they didn't know that they could get it via Amazon Prime. So it might be worth checking out, to see if you want this kind of notebook l m flavor on the go. So here is, what Amazon said about it.
Jordan Wilson [00:29:40]:
So, they said creating custom audio content is now effortless. No documents to upload. No prep work needed. Just tell Alexa what you're curious about, and it does the rest in minutes. Alexa will pull together the relevant information, give you an overview of what it plans to cover, and let you adjust the length and direction conversationally before generating anything. Once you're happy with the plan, Alexa creates a recording with AI generated host voices. When your episode is ready, you'll get a notification on your Echo Show device and the Alexa app. Just tap to start listening.
Jordan Wilson [00:30:15]:
You can also find it in the music and more section or tune in through the Alexa app whether you're on the go. Alright. Here's one thing that I want to call out. Alright. It's not Tuesday, but I have a small little hot take. So clearly this is Amazon calling out Google in their notebook l m. But the thing I don't think Amazon understands here, and we'll see because personally, I don't see this working out very well. Number one, Alexa Plus is is hot garbage.
Jordan Wilson [00:30:46]:
Right? Alexa Plus, still garbage. Siri, still not usable. Maybe that changes, you know, in a couple of months, when and if, oh, this is funny. Sorry. I just had a a codex automation pop up on my, on my screen here. I should probably turn that off when I'm doing my live stream. Right. But, you you know, Amazon here really going after and calling out Google and notebook l m by name by saying, you don't need to upload any documents.
Jordan Wilson [00:31:13]:
Right? Which here here's the thing. Alexa Plus is not good. Right? It it it hallucinates. It it's it's terrible. It doesn't understand what you're talking about. So for me, personally, at least for my use cases that I use audio overviews for and I use them all the time, I want to provide those documents. Right? I don't want Alexa plus to, you know let's just say, I want, hey. Give me the latest AI news updates this week.
Jordan Wilson [00:31:39]:
Right? If you try to replace me out there, don't you do it, at least not with Alexa, plus podcast because I'm guessing it's gonna be abysmally bad. It's gonna be laughable. It's gonna pull things from months ago. Right? So at least for me, Amazon here is trying to, you know, make a point here where I don't know. The point I think and why notebook LM audio overviews are so freaking good is because they do work on your documents, and they ground it only in your documents. So I think at least it's worth talking about here because it's interesting that Amazon is, you know, literally intentionally calling out. Right? You don't have to upload documents. Everyone knows this is a a straight up copy of notebook l m's audio overviews.
Jordan Wilson [00:32:18]:
Personally, I don't see this coming out well. Like, I don't know. Some of these big companies, I think you should probably talk about this and say this out loud, and I don't know. Maybe ask someone outside of your your own company, right, if this is a good idea or bad idea because I think they're trying to make a splash with this to get people to use. Alexa plus, they probably thought it would, you know, maybe go viral. Like, the audio overviews did. I don't think it's gonna turn out well. Alright.
Jordan Wilson [00:32:43]:
Speaking of that, since apparently, I'm giving my opinion on all the AI feature updates, I don't think this one's gonna turn out well, at least not in the short term for Google. So the next AI update is Google's AI, anti gravity two point o. Alright. So, a lot of, you you know, IDE and CLI terminology. Right? Someone told me on LinkedIn the other day, they're like, Jordan, you use too many buzzwords. Explain things. Alright. So, IDEs are essentially these, you know, agentic coding interfaces.
Jordan Wilson [00:33:14]:
Right? And a a CLI command line interfaces when you're coding or using something like Claude code or codex via the terminal. Right? So your your CLI is that's kind of how you're coding in the terminal. Right? And then you have these desktop applications that allow you to code, but it's a little more user friendly. That's kind of your IDEs. So that's what anti gravity is. Think of it as well. It's, Google's new version of codex. Little confusing, but let's talk about what's new and, yeah, it wasn't all good.
Jordan Wilson [00:33:42]:
So this was also announced at Google IO, and it's an expansion of the original anti gravity IDE, and Google's, kind of, first foray, I guess, at least if they're putting a lot of momentum behind into a, full agent first development platform. So, yeah, here's where it starts to get confusing and why you need a map and a translator and yeah. It's it's confusing. So there's five services. So, there's now a standalone desktop app, which I tried. There's still the anti gravity CLI, which is code named a g y, which replaced the Gemini CLI. Right? There's also an anti gravity SDK. There's the managed agents in the Gemini API, and also Google's AI Studio, can use the anti gravity harness.
Jordan Wilson [00:34:29]:
So it's also a harness now. And yeah. And there's still the Gemini desktop app, so it's different. So it's confusing. So let's at least talk about who has access. So, the desktop app is now available on Mac OS, Windows, and Linux. There is also a new pricing tier, and they even adjusted the, temporarily tripled the limits inside of anti gravity. I think that's just a short term thing, though.
Jordan Wilson [00:34:55]:
Because a lot of people were not happy when it rolled out with the limits and just overall with the product. So the other thing that people are very upset about is the Gemini CLI is the is being deprecated for consumer users. So that's, I think, is the biggest thing that a lot of people were upset about because I believe the Gemini CLI was open source, and a lot of people loved it. So, essentially, Google was allowing this great command line, you know, coding tool. And I do know originally when they announced it, it was like free tokens originally. Right? So now they're sunsetting it in June, and this is part of the replacement. So, let's talk about why it's useful. Well, if you are I'm trying to be honest and, also factual here because for me, I don't I you know, I tried it.
Jordan Wilson [00:35:46]:
I don't find it useful. I don't think anti gravity is very good right now, which is interesting to me because Google seemed to push it way more than they should have. If I was Google, I probably would have announced this on the down low. I wouldn't have put a lot of, you know, given this a lot of, you know, shine at the keynote. But they did, and they're continuing to push it afterwards, which is interesting because it's not good. Right? A lot of what Google released is really good. Right? Gemini three Gemini three flash, if you're using it inside the Gemini app, right, not paying for the usage, really good. Gemini Omni, really good, really exciting.
Jordan Wilson [00:36:19]:
A lot of other things that they announced that maybe aren't available yet to anyone, really good. Anti gravity two point o, really not good. Right, especially when you compare it to codex. And why I do have to bring this up is because Google has gotten a lot of flack and probably rightfully deserved. Because in the actual announcement video, I kid you not, in the actual announcement video from Google, it was a sixteen minute video. I watched it, about the two or three minute mark. I don't know if no one edited it or watched it, but literally on the person that's demoing the new anti gravity, there is a desktop folder that says codex, alright, or a folder on their computer that says codex. And why does that matter? Well, visually, this looks like codex.
Jordan Wilson [00:37:03]:
It looks the exact same. Right? When I opened it, I I obviously had codecs open on my computer, and I opened and installed anti gravity, you you know, the hour after it came out. And I'm like, wait. Why is this not launching? And where did all my codecs chats go? Like, I literally thought that because it looks identical to codex. Right? And, actually, the codex team called this out, you know, tongue in cheek. But, yeah. Again, I'm not sure. I think Google could have gotten this right.
Jordan Wilson [00:37:31]:
Maybe they felt the need to get out a an unpolished version, which I commend companies for putting something out that's maybe not quite perfect. But if you're gonna do that, don't highlight it as much as they did. You know? Just say, hey. This is a a beta product. You know? It's it's working, you you know, work in progress, but go go use it. So alright. Let's wrap up here because our last one, I think, is the big one. Alright? Oh, yeah.
Jordan Wilson [00:37:56]:
Sorry. Joe said quiet. Don't say a l e x a's name. Yeah. Sorry. I've already done it. So, let's see. Alexa.
Jordan Wilson [00:38:04]:
Alright. We'll see. I just screwed up, probably for a lot of people. Sorry. I, yeah, I I think we need a code name if I'm covering, you know, a l e x a or s I r a, to to not trigger that for you. Alright. But, yes, Google obviously won the week when it came to total releases. But like I said, you almost have to have a degree in daily AI like me to understand what Google released because a lot of the things I wanted to cover this week, but I couldn't because some things are only available to workspace members.
Jordan Wilson [00:38:38]:
Right? Some things are are sorry. It's only available, well, some things all are only available to workspace members. Some things are only available if you're a free Gmail user. Some things are a paid personal Gmail user. Right? You know, you have to be an early tester. Some are things are in beta. It was too hard, to navigate what's actually available, and that's what this show is, which I think if you wanna talk about the most consequential single announcement this week, I think it's codex. I know you're probably tired of me, guys, like saying codex, codex, codex, but let me repeat what the most important thing for you to hear right now is.
Jordan Wilson [00:39:11]:
Ready? Codex, codex, codex. The team at OpenAI continues to embarrass the entire AI industry with how good and useful codex is and how far it is. Right? I said about two months ago that Anthropic was winning 2026, but I think that tide has shifted. And I didn't think anyone would catch them, maybe not until the end of the year. Certainly not in the second quarter, but I think OpenAI has retaken the lead in 2026, especially after a somewhat lackluster at least response, to Google IO. I think a lot of people bought Google IO. We We were gonna get the Gemini two point or 3.5 pro. We're gonna get a more cohesive and easier to understand way to use all of Gemini's products, but it was just fragmented.
Jordan Wilson [00:39:56]:
Right? Some good products, some confusing, but all over the place. But codex continues and opening eye continues to put more and more useful features that are easy to use inside of one single platform. So here's what's new. Yeah. Because now OpenAI has moved to this new weekly Thursday drop. So a lot new. So one called app shots. So this is essentially you can press both command keys when you have codecs open, and it takes the foremost app window, as a kind of like a screenshot, and it brings it into codecs.
Jordan Wilson [00:40:31]:
But it's not just a screenshotting tool because you'd be like, okay. Well, I I can screenshot anything and drop it into codecs. It also brings the text in the context of that window into codecs. So it just makes day to day working with other platforms so much faster. One of the things that I think slows me, quote, unquote, slows me down from my codecs flow, even though I have a good screenshotting program. Right? I use clean shot, but it slows me down to, you know, do my little keyboard command, you know, shift command four, drag my mouse, and then drag the screenshot over even though it's a very efficient way to screenshot. Right? But then I still don't get the context and the text from that app. So that's a lot of at least what I do, and I know I have to explain this because you might just be like, okay.
Jordan Wilson [00:41:10]:
What's the big deal? It's actually a big deal. Right? This little thing called app shots. So that's new. The other thing, goal mode is finally in the desktop app. So goal mode is absolutely bonkers. So, codex was first, but they only brought this to the CLI, to the command line interface version of codex. Then, Infropic followed suit and brought co goal mode, to their CLI. So this is the first goal mode that is in the user friendly desktop interface, and it is in codex now.
Jordan Wilson [00:41:45]:
Huge. So this essentially, graduated out of the experimental phase and is now available, in the codex app. And it allows codex to drive autonomously toward a specific objective for hours or even days. Alright. So I will warn you. Alright? Goal mode requires well, you need a little bit of experience and you need a very specific prompt with use cases, examples, all of that. The last thing you wanna do is go give codex a very, wide and, general goal because it's gonna work to that. And, well, even though, OpenAI's, rate limits are the best in the business.
Jordan Wilson [00:42:23]:
Yeah. Google even downgraded their paid rate limits after IO. And tropics are laughable. You know, OpenAI's are the best in the business by far. But if you give codex a goal that's way too broad, it's gonna work and work and work to get it done. So if you're on the $20 a month plan, it's not gonna last as long as you might want. Alright. So the three big ones, app shots, number one.
Jordan Wilson [00:42:45]:
Two, goal mode. Number three, this one's actually big. You no longer have to carry your laptop halfway open. So they have a new remote computer use. Okay. So this is different than being able to control codecs via your phone inside of the chat g p t app. This allows when your computer is locked. Alright.
Jordan Wilson [00:43:04]:
So when the lock screen is up, I don't know how. How is the codex team shipping these things so quickly that just defy the laws of software. Right? So, yes, you can use computer use now while your computer if you you have a laptop, it it's closed, it's locked. Absolutely bonkers, and you can still control it, obviously, via the codecs mobile, section inside of the chat GPT app. So who has access? Well, everyone. Yeah. That's the other thing. Everyone.
Jordan Wilson [00:43:34]:
So go use codex. I already told you. And go go make sure you repost Wednesday's show. If you haven't already gotten the cookbook, people are loving it. Alright. We gave that away. If you go repost, Wednesday's show which was part two of our codec series. Alright.
Jordan Wilson [00:43:46]:
So, the codecs, obviously, a lot of these things are only available for Mac users. I know window users aren't gonna like that, but that's why, you know, it's very hard to develop things for Windows. FYI, it's much easier to develop things for Mac, which is why the cutting edge of AI lives inside of Mac for now. Alright. So here's why it's useful. So, I mean, app shots, it collapses one of the most annoying handoffs in the agentic workflow is just explaining or pasting what's in another app. I mean, literally, the way they have it set up by default, this is how much I I am loving what OpenAI is shipping. Just think, how do your hands sit on the keyboard? Right? For the most part, your your your two thumbs sit near the space bar.
Jordan Wilson [00:44:27]:
So to launch this on a Mac, you hit the command key twice, which is the key just to the left and the right of the space bar. So talk about the small details that really matter. This stuff is huge. So goal mode, that's gonna be extremely useful. This is your, you know, your Ralph loop. Right? This is your agent that's gonna run for, you know, hours or maybe even days. Right? That other vendors keep promising. It's here.
Jordan Wilson [00:44:50]:
Right? And then remote computer use, obviously, just closes the loop for genuine background autonomy. So you don't have to, like, leave your Mac, you know, your MacBook open and plugged in and, you know, turn off all the power settings overnight. My gosh. I forgot another one. This is how much opening I just shipped yesterday. And then they also, shipped shareable plugins across teams, which is a huge pain point, for enterprise teams. Right? I did a a a Claude, actually, Claude training for a huge enterprise client a couple of weeks ago. And one of their biggest pain points company wide was how do we share skills within Claude? And it actually is a big problem, that there's no built in way to do that.
Jordan Wilson [00:45:28]:
And OpenAI just fixed what I think is one of the biggest problems that no one really talks about in the enterprise, which is the plugins. Right? So plugins are skills plus app use, and now you can share those across teams if you have a team account inside of codex. My gosh. Do you guys see why now I'm just like codex, codex, codex. Google shipped a lot of great stuff. It's hard and confusing to use. None of it works together. Right? So, a lot to cover this week.
Jordan Wilson [00:45:57]:
Sorry. This one went a little long. There was a lot to cover. But I hope this one was helpful. If so, please, if you're not already subscribed to the podcast, please do that. That really helps both me keep the show going, and it helps other people find this, who don't just want, you know, a bunch of people reading a script and, you know, showing for some product. I try to give it to you straight. You know, I always have, and I continue to always will do that.
Jordan Wilson [00:46:23]:
So please subscribe to the podcast, and then go to our website, youreverydayai.com. Sign up for the free daily newsletter. Thanks for tuning in. We'll see you back next week. Actually, Tuesday. Alright? Monday, holiday. We'll see you Tuesday with the AI news and every day after that for more everyday AI. Thanks, y'all.