Is anyone actually using AI agents?

This week, Mike Belsito identifies the three AI stories product managers need to pay attention to.

This week's stories:
— Salesforce making Slack an MCP client sets a new expectation for enterprise AI — the interface layer is wherever your team already works, not a separate dashboard you have to visit
— OpenAI's up-to-90% price cuts mean it is time to audit your roadmap graveyard for features killed because the AI cost didn't pencil out
— When AI inference becomes cheap enough to be infrastructure, the moat shifts to data, distribution, and customer relationships — not the model
— Enterprise AI agent adoption is further ahead than the public narrative suggests — if you sell to large companies, assume your customers are already in Agentforce conversations with Salesforce
— The gap between AI agent capability and mainstream consumer adoption is a UX and distribution problem, not a technology problem — and it is the defining product question for anyone building consumer products right now

Referenced
— Salesforce Agentforce: https://www.salesforce.com/agentforce/
— Slack: https://slack.com
— Dreamforce: https://www.salesforce.com/dreamforce/
— OpenAI pricing: https://openai.com/api/pricing/
— Josh Miller on X: https://x.com/joshm
— Browser Company / Dia: https://thebrowser.company/
— Arc Browser: https://arc.net

Transcript

SPEAKER_00 0:00

I'm Mike Belcito, and this is Now Shipping, the weekly AI news show for product people. Every week I bring you three AI news stories that matter to you, the people actually building products for a living. Not the hype, not the noise, just the stuff you need to keep up with. Brought to you by the team at Mind the Product, this is Now Shipping. All right, story number one is all about Salesforce and Slack. And I'll start this off with a question. Where is it that you and your team go to actually get stuff done? Like, where do you go for the coordination, the handoffs, the communication? For me, it's Slack. Like that happens to be where my team lives throughout the day. And that's true for a lot of teams. Now, Slack, of course, is owned by Salesforce, and they just spent the last 18 months building a massive AI agent platform called Agent Force. And it's working. Salesforce reported $1.4 billion in annual recurring revenue from Agent Force. And that's 114% year-over-year growth. And a lot of that growth is coming from large enterprise customers. But they've been wrestling with a problem that anybody who's built agent-powered features would probably relate to. And that's if your AI agents live in one place and your team lives in another place, agents don't actually work. I mean, they technically work, but when you think about the fact that they're a tool that you have to go visit and they're not happening in the places that you're living in all day, it probably isn't helping you the way that you were intending for it to help you, right? So Salesforce announced that their Slackbot is now a fully functional MCP client. And by making Slack an MCP client, Salesforce is bringing the agents to you specifically to the place where you're already living and working with your team every single day. Now, what does that actually look like in practice? Well, it could be that a sales rep, maybe they're in Slack and they pull up a deal on the Slack bot and pull the latest account activity from Salesforce's CRM. They summarize the last three customer interactions and they draft a follow-up email all in that same thread. The bot actually orchestrates across Agent Force, across all of Salesforce's 6,000 different app exchange integrations, the 2,600 Slack marketplace apps, and it does it all in one thread. Now, ahead of Dreamforce, which of course is Salesforce's big annual conference that's coming up in September, they've been previewing this news. And the theme of Dreamforce sort of tells you everything, which is the agentic enterprise. They're saying the enterprise is reorganized around agents doing real work. Now, for product people, I think one important takeaway is that the interface layer of enterprise AI is now starting to be wherever people work, right? It's not a separate AI dashboard or some new app. So if you're building for enterprise and you haven't thought hard about how your product fits into Slack or Teams or the places where people are actually working today, it might be something you want to start thinking about, probably like right now. I think products and companies that end up living within this layer, they're gonna have a huge advantage. And also, I'd say if you sell to the enterprise, some percentage of your customers are already in agent force conversations with Salesforce. Some of them have already signed, some of them are already living this reality now. So the assumption that, you know, oh, our customers, yeah, they're big enterprises, they're not ready for AI agents quite yet. That's probably a lot less true now than it was even six months ago. So it's worth asking your sales teams, your customer success teams, like, what are you hearing out there? What companies are actually starting to adopt agentic work within specifically enterprises? I think it's more than everybody's letting on. And I think we're gonna continue to see more and more companies consider MCP as just table stakes for how others can access the features of their product. I don't think it's it's one of these like pilot program sort of things now. I think everybody is just sort of expecting it. So that is story number one. On to story number two. Okay, story number two on July 30th, OpenAI cut the price of its lower cost GPT 5.6 models by up to 90%. So look, if your product is making 10,000 AI calls a day and each call processes an average of 5,000 tokens, that's 50 million tokens a day. At $2 per million, that's $100 every day in AI cost. It's over $35,000 per year. Now, what does that look like at 20 cents per million? That's $10 a day. That's less than $4,000 a year, close to a 90% reduction. Now, at that price point, entire use cases that didn't make economic sense before, they start to make sense. High volume extraction pipelines, running AI over large historical data sets, products or features where you had to be selective about when they fired off an AI call because of the cost. Now, maybe you don't have to make that trade-off anymore. OpenAI also announced separately that it's giving roughly 100,000 researchers free access to its frontier models through 2027. That is a market development move. I think it's like get a lot of builders deeply embedded into the platform at no cost, and probably they'll keep building on that platform. So, what does this all mean for you? I think a few things. First, go back and look at the features that you killed in prioritization because AI cost was the blocker. Like, literally, right now, go back and do that. I mean, some of the features that you probably cut were cut for the right reasons, but maybe cost was one of those reasons. Maybe the only reason you couldn't do it was because it was just too expensive. Now the math has changed, so it's time to revisit that. And second thing is if frontier quality AI is nearly free, all right, maybe not free, but cheap enough, right? Like what is the moat? I mean, the moat in your product is probably not the AI anymore. It probably is like your data and your distribution, your customer relationships. AI is just becoming the infrastructure. It's sort of like cloud storage or compute. It's basically now table stakes. So the faster your team internalizes this, the better off you'll be. Now, I think we will start to see some pricing pressure on Anthropic and Google and really everyone else for similar models. Now, we saw that kind of thing happen when the pricing for Sonnet 5 changed. I think we'll continue to see it again. And that's good news for our budgets for sure. So hopefully we'll continue to see that cost get driven lower and lower. All right, one more story to go. This last story is a little bit different from the stories we've been doing. This is actually about a post from Josh Miller that's been making the rounds on X. So quick background Josh Miller, he's the co-founder of the browser company. Now, the browser company made the Arc browser. Maybe you've used it or at least have heard of it. And now they're building Dia, a new kind of AI first browser. Um, but the point is, this isn't somebody just like random, somebody from the sidelines. This is somebody who's actually staked his company on a bet of where AI agents are going. And his post on X goes something like this. I'm going to paraphrase it, but basically he asks the question isn't it kind of crazy that basically nobody outside of tech is using AI agents? Now, think about that. It's really true, right? I mean, you have software engineers, you have the early adopters, you know, those of us in product. Um, you have people at companies where their company is sort of force feeding it to them. But if you think about your college friends in the group chat, you know, people that aren't working in tech at all, you you probably remember with that group, you know, the feeling you got when people started first using Instagram or TikTok or maybe Uber, like word started going in those circles about it, and very quickly everybody sort of latched on. I don't think we're there yet with AI agents. Anyway, that's what Josh was saying. Josh is saying we're not there yet. And he's asking, when will that moment actually come for agents? Now, I will say that group has probably already experienced the chat GPT moment. That wasn't just for the early adopters, that really made it to everybody, including your college group chat, right? But it's an interesting question. He lays out this sort of contradiction because the models are remarkable. I mean, where we're at right now compared to where we were a year ago, it's crazy. The infrastructure's there. I mean, now every major tech company seemingly has their own agent platform. I mean, we just talked about them earlier in this episode. Hundreds of, you know, YC startups, they're all building vertical agents. And yet, again, most people out of tech, the people who spend all day on their iPhones and get paid to work in browser tabs, they don't really care yet. They're using ChatGPT, they're using other chat-based platforms. Um, but he says, uh, Josh says that the engagement data for those platforms like ChatGPT and Claude, it shows that most people are using them like a glorified Google Meets Grammarly, which is useful. It probably helps a lot, but they're not using it in the way of like, hey, I gave it a goal and it went and accomplished those things. So why doesn't the general public care about AI agents yet? And why might the fact that they don't care, why should that make you care? Now, I think the gap between agents and consumers is mostly a user experience problem. Think about what made Chat GPT go viral. Just a blank text box, one input, one output. Very easily people could just type something and the AI responded. I mean, it felt like magic, right? The interface eliminated every possible barrier to entry. Agents ask something much different of people. You have to articulate a goal and delegate the execution and then trust something to complete a multi-step task on your behalf without you watching. And you have to do that all in a place where it doesn't quite feel accessible yet to just sort of like the general public. Now, when the user experience removes all of that, things start to change. I mean, I just think of myself with open claw at first. It took me a lot of work to get to the point where open claw could just feel magical, right? I had to go through all sorts of technical hurdles. Definitely my sort of like non-tech friends, it's not something they would have even tried to do, right? I think Claude Cowork took a step in the right direction. It's easy to fire off agents without even really thinking of it. It sort of feels like, you know, just chatting with the AI, but it might not be there completely yet. I think the other reason why we might not have so much adoption on the consumer side is nobody's really forcing it, right? Enterprise has IT departments to deploy things, it has well-defined workflows to plug agents into. Companies can force it on their employees, and it's hard at first, but then people get used to it and then uh they can't imagine a life without agents. But nobody's telling your parents to set up an AI agent to manage their schedule. Um, unless you are. Maybe you're encouraging your parents to do that. If you are, great, but um, in all honesty, nobody's forcing them to do it. They have to discover it themselves and put in the work to get to that point. So distribution's a real issue. Now, is the consumer agent breakout adoption coming? That's the trillion dollar question. I think yes. I in fact, I would bet hard that it will be coming. But for product people that are building consumer products right now, this is a really important question to think about. And you'll have to ask yourself whether you would make that same bet. All right, that is going to wrap up this week's episode of Now Shipping. If you found this valuable, please subscribe, tell a friend, and if you have any suggestions, please leave a comment below. I promise I will read every single comment you leave. I will see you right here next week with three more AI stories that matter to you as a product person. Once again, my name is Mike Belsito, and from the team at Mind the Product, this is now shipping.