Learning Library

← Back to Library

AI News: Claude, Walmart Agents, OpenAI Ads

Key Points

  • Anthropic unveiled Claude Sonnet 4.5, a model that excels at building/editing Excel sheets, creating PowerPoint decks, and coding, but its performance hinges on clear, well‑crafted prompts.
  • Walmart has rolled out a “WB” super‑agent across more than 200 AI tools, achieving a 95% autofix rate on bugs and proving that large‑scale AI agent orchestration is already viable in enterprise environments.
  • OpenAI introduced ChatGPT Pulse and Sora, two new advertising‑focused products that expand where AI‑driven content can appear, a move marketers and product teams need to factor into budgeting and platform strategy.

Full Transcript

# AI News: Claude, Walmart Agents, OpenAI Ads **Source:** [https://www.youtube.com/watch?v=DWb4SqILvvM](https://www.youtube.com/watch?v=DWb4SqILvvM) **Duration:** 00:10:36 ## Summary - Anthropic unveiled Claude Sonnet 4.5, a model that excels at building/editing Excel sheets, creating PowerPoint decks, and coding, but its performance hinges on clear, well‑crafted prompts. - Walmart has rolled out a “WB” super‑agent across more than 200 AI tools, achieving a 95% autofix rate on bugs and proving that large‑scale AI agent orchestration is already viable in enterprise environments. - OpenAI introduced ChatGPT Pulse and Sora, two new advertising‑focused products that expand where AI‑driven content can appear, a move marketers and product teams need to factor into budgeting and platform strategy. ## Sections - [00:00:00](https://www.youtube.com/watch?v=DWb4SqILvvM&t=0s) **Anthropic Releases Powerful Claude Sonnet** - The segment spotlights Anthropic's Claude Sonnet 4.5 as a versatile AI excelling in spreadsheet, presentation, and coding tasks—requiring clear prompts for optimal autonomous development—while also noting Walmart's new AI deployment. - [00:04:15](https://www.youtube.com/watch?v=DWb4SqILvvM&t=255s) **OpenAI's Sora Launch & AWS Agent Core** - The speaker notes OpenAI’s upcoming consumer release of Sora 2 as a sign of more 2026 products, and highlights AWS’s new open‑source Agent Core MCP server that aims to make AWS the default platform for building production‑grade AI agents, pressuring rivals. - [00:08:22](https://www.youtube.com/watch?v=DWb4SqILvvM&t=502s) **Salesforce's Secure AI Coding Push** - Salesforce is launching an AI‑powered, security‑focused low‑code co‑pilot called Agent Force Vibes to leverage its enterprise SaaS relationships and challenge rivals like Lovable and Bolt. ## Full Transcript
0:00I spend dozens of hours covering the 0:01news so you don't have to. And I'm going 0:03to give you the news that matters to you 0:05for AI in just 10 minutes. Okay. Number 0:07one, Enthropic launches Claude Sonnet 0:104.5. Why should you care? You should 0:12care because this is the first model 0:15that is extremely good at both building 0:17and editing Excel files and also at 0:20building PowerPoint. It also has a very 0:23strong coding DNA. This model famously 0:27built Slack over again in 30 hours 0:31straight coding. One task just went and 0:33did it. Now I have seen in my testing so 0:36far, not 30-hour run times but longer 0:39run times than I have seen before. This 0:41model likes to take its time, think 0:44through problems and solve them 0:45correctly. And it has a strong check 0:48your work ethos. So why you should care 0:51is it is driving us toward that 0:53autonomous development future. But it 0:56will only work if you set it against the 0:59right problem space. If you architect it 1:01within a framework that works if you 1:03think about the task you're giving it. 1:05In some of my testing outside the coding 1:07environment, I found similar constraints 1:10and opportunities. If I give it a good 1:12prompt for PowerPoint, it's going to 1:14produce a great tech. If I give it a 1:16lousy prompt for PowerPoint, I'm going 1:18to get something difficult to build and 1:21unreadable. So, this matters for you 1:23because it prioritizes your intent and 1:26it prioritizes your ability to define 1:29work. If you can define the work, this 1:32thing is superpowered. Story number two, 1:35Walmart has deployed the WB, which is a 1:38terrible name, super agent across 200 1:41plus AI tools. And you care because it 1:44shows that AI agents are operative in 1:47the biggest companies in the world right 1:49now and they deliver value. They're 1:51getting a 95% autofix rate on bugs and 1:55they're doing it across a complex 1:57ecosystem inside their development 2:00environment now today deployed. And so 2:03this is showing us and reminding us that 2:05AI orchestration is here at scale 2:09already. And the federated agent model 2:11that they're using across multiple 2:13workflows is underlining that you can 2:16get really complex multi-workflow 2:19footprint set up for agents that do 2:21work. So why should you care? Because if 2:24you haven't believed that agents are 2:25coming, they are. And if you have 2:28believed that agents are coming, you can 2:30actually build them and deploy them at 2:31scale. It's just a function of having 2:35the right workflow, defining your intent 2:38properly, and then orchestrating it out. 2:40There's more detail on that coming. I'm 2:41going to do a deep dive on that. Story 2:43number three, OpenAI has launched 2:46ChatGpt Pulse and Sora. And I'm putting 2:49those stories together because they are 2:52intentionally together for OpenAI. Open 2:55AAI has launched two new advertising 2:57surfaces in the last week. Why? They are 3:00going after ads people. Why should you 3:03care? If you're in marketing, you care 3:04because of budgeting. If you are 3:06building for product, you care because 3:09you have to think about where your 3:11product is going to appear. Not just in 3:12chats, but now potentially as a product 3:15offering. Yes, I am not just assuming 3:18this will be a consumer ad platform. It 3:20is likely given the way we use AI, this 3:23will also become a B2B ad platform. This 3:26is going to be one of the biggest 3:28developments in marketing in the last 20 3:30years and we are going to start to see 3:32it and we saw the first product surfaces 3:35for it launch this week. So pay 3:38attention to Sora, pay attention to 3:40Pulse and where they're going. They 3:42point the future and they also, by the 3:44way, underline an important component of 3:47Chat GPT's strategy. They look at a 3:50model and they decide when it's ready 3:53and then they find a surface for it and 3:55kick it out. I would argue, they haven't 3:57said this, but I would argue that pulse 4:00is its own chat GPT model. The results I 4:02get from pulse overnight are quite high 4:05quality and they are different 4:08qualitatively than what I get from 4:10standard chat GPT5. 4:12Sora, of course, is its own model. Very 4:15famously Sora 2, right? They had it in 4:17Sora 1. and it wasn't quite ready. They 4:19didn't launch a consumer product. Now 4:20they feel they have it and they launch a 4:22consumer product. This suggests how 4:25OpenAI is going to approach future 4:27launches of their models in 2026. Look 4:30for more net new products out of that 4:32company. Next, AWS is launching the 4:36agent core MCP server. Why should you 4:39care? It's really about AWS playing 4:42catch-up. They are providing open-source 4:44infrastructure for building production 4:46ready AI agents with built-in runtime, 4:48gateway integration, identity 4:50management, memory, and all the rest of 4:51it. The server integrates with 40 plus 4:54different MCPaware clients including 4:56anthropic cloud code, cursor, etc. And 4:59it gives developers the option to build 5:01agents that can securely call external 5:03tools and also maintain context across 5:05sessions. You know why you should care? 5:07because Amazon is leaning into 5:10open-source to preserve their cloud 5:12revenue. And if you are on AWS, it's 5:15probably a great deal for you. If you 5:16are not on AWS, I don't know that it's a 5:18reason to switch, but it's something to 5:20keep an eye on because Microsoft and 5:22Google may be under pressure now to lean 5:24this hard into open-source agent 5:26infrastructure in their own way. They've 5:28done some, right? They have the A2A kit 5:30from uh from Google and I think there's 5:33the agent development kit from Google. 5:35So, there's some work being done here. 5:37AWS is pushing hard on becoming the 5:40place for developers to build because 5:42they need to preserve the cloud revenue 5:45and that is something that implies that 5:49Azure and Google are really gaining 5:51ground with AI native organizations and 5:53that has come through in recent 5:54quarterly earnings and that's something 5:56that is going to shift the balance of 5:57power and ultimately where developers 5:59and builders prefer to build because if 6:01you're not building with AI these days 6:03where are you all right next story 6:05Microsoft copilot Pilot is opening the 6:07door to other models. For the first 6:09time, we have co-pilot working with 6:12clawed models and we have the idea of a 6:15multi- aent enterprise strategy inside 6:17the co-pilot ecosystem. This is not 6:20super surprising since Microsoft has 6:23already done this with Azure and Azure 6:26has been a place where you can get 6:27multiple models in a secure sandbox. But 6:30they are going a step further with 6:31co-pilot. What Sachin Nadella is 6:34realizing is that he needs people to 6:38care about co-pilot and the AI Microsoft 6:41productivity experience. And if that 6:43means he can't have his own proprietary 6:47AI models and he needs to bring in other 6:49people's models to make that work, he 6:51will do it because he needs to hold on 6:54to his attention and distribution 6:56footprint for co-pilot products. That is 6:58the next generation of office. And if I 7:00were Satcha, I would also be worried and 7:03I would also be releasing now because of 7:05the power of Sonnet 4.5, which was the 7:07first story I shared in this little 7:09update. Sonnet is attacking work 7:11primitives that Microsoft has dominated 7:14for decades. They're going after Excel. 7:15They're going after PowerPoint. They're 7:17going after Docs and they're doing it 7:19well. Microsoft would rather join them, 7:21right? like Microsoft would rather bring 7:22them into the fold, work with them on 7:24co-pilot and make sure they get the 7:27distribution advantage secured for 7:29Microsoft because since the days when we 7:31shipped CDs on software, Microsoft has 7:34built off this distribution advantage in 7:36the office. Everybody runs Windows. This 7:39is why it doesn't really matter what 7:41Slack does. It matters what Teams does 7:44because Teams has the distribution. So 7:46you should care because this is the 7:48first case where a major enterprise 7:51platform is offering competing AI models 7:54inside their same interface they use for 7:57selling their own software. Right? Like 8:00this is this is much more integrated 8:03than we have seen when you have back-end 8:05developer availability of multiple 8:07models. This is no you can choose as the 8:10end customer what model you want. We are 8:13starting to see enough competition where 8:15vendor lockin for AI may just not be 8:18tenable any and that is going to matter. 8:20Competitive pressure is going to 8:22intensify. We are going to start to see 8:24IT departments able to negotiate better 8:27terms because of moves like this. Story 8:30number six. Salesforce is shipping agent 8:35force vibes which I don't know that I 8:37would have called it that but here we 8:38are. They are looking to bring natural 8:40language coding to enterprise 8:43environments and they're leaning on the 8:45salesforciness of it all. They're saying 8:47we'll have built-in security governance 8:49and compliance controls. In many ways, 8:52this is exactly like the Teams and Slack 8:55conversation, except ironically, in this 8:58case, Salesforce is the one with the 9:00distribution advantage versus tools like 9:03Lovable or Bolt, which don't have the 9:05long-term SAS enterprise deals. 9:08Salesforce does. And so, they can go to 9:10a CTO and they can say, "Oh, you've 9:12heard about Lovable. I love Vibe Coding, 9:14too. Why don't you use Agent Force 9:17Vibes?" And the platform includes the 9:19special autonomous AI agent. It connects 9:21to Salesforce orgs. It can talk to your 9:22data securely. It can actually enable 9:25you to build securely. 9:27This is Salesforce's play to build a 9:31vibe coding co-pilot for lack of a 9:34better term. What they want is to give 9:37enterprise developers a way to expand 9:40their footprint and to work with product 9:43managers and marketers and others in 9:44their orgs in ways that are integrated 9:47into their systems instead of having all 9:49of this shadow IT. I get why they're 9:51doing it. I get why security first is 9:53attractive, but what you should watch 9:55for is the actual usage and conversation 9:59around this tool. It is easy for this to 10:02get sold to CTO's. It is not easy to 10:06persuade product managers, marketers, CS 10:10leaders who are already vibe coding 10:12using shadow it like lovable to move to 10:15a tool like this if it's not actually 10:19good. And that's really the trick. You 10:20can sell the software, but you can't 10:22make people use it. And so what I'm 10:24going to be watching for is whether we 10:27get an enterprisebuilt vibe coding tool 10:29that is actually good. I think the jury 10:31is still out. And that's the news that 10:33mattered this week.