GPT-5.5 Drops, Walmart Fires OpenAI, Meta Cuts 14,000

Welcome to episode 6 of the AI Operators. I am so grateful to be here with you all. We are here because of the teams at Richpanel and AfterSell. He is Ben. The world is changing and you have a man who has bootstrapped his business to over $1 billion in sales, who is an AI expert. Here guiding you every step of the way. Ben, what's up, man? How was the last week?

Incredible. So many interesting things. ChatGPT 5.5 came out. I'm excited to talk about it, but you know, as always, I'm living the dream. Can't complain. How are you?

I am great. I am exhausted from celebrating Mahin, who leads this thing in the background. Her birthday bash over the weekend was wild. Mahin, happy birthday. But Ben, you are right. There is a lot to cover, so let's get into it. OpenAI just shipped GPT-5.5 and Workspace Agents within 48 hours, taking back the top of the leaderboard from Claude. Walmart killed their OpenAI partnership and built their own agent named Sparky, all inside ChatGPT. Meta has cut 14,000 jobs to fund AI, and Mark Zuckerberg's single very talented person era is about to come up in every conversation you have with your board this year. Bill Nguyen created an AI version of himself to run his life, and we're gonna go deep on it. Anthropic ran an experiment where nearly 70 agents negotiated over 190 deals with one another. We are going to dive deep on that. Plus, we are so lucky to be joined by Matt Bertulli on running Pela and Lomi, two brands, and how he's integrating AI across both companies and Arias Jury, who leads supply chain for Block Blue Light, is gonna walk us through his AI chief of staff that keeps him on point. And Ben, the brilliant Ben, is gonna break down model context protocol, the 3-letter tool that connects every AI workflow. Let's get into it.

Some brands are cutting their support tickets by almost 50% with AI. I've actually been recommending Richpanel to my community for the last 6 months, so I'm really happy they're the headline sponsor of AI Opera. Operators. What they're building is pretty different from most AI support tools. Most tools are basically just AI agents. You train them once and hope they behave. Ridge Panel built what they called a boss AI. You just tell the boss AI your values, principles, and desired outcomes. It handles the rest. The boss AI reviews thousands of conversations every week, scores them, and continuously improves them to get to your desired outcomes. Brands like Ridge cut ticket volume by 45% and Jones Road actually cut their support team size in half, and both of them were live in under 2 weeks. Ridgepanel even guarantees 30% ticket reduction in 60 days or your money back. If you want to see what that could look like for your support team, go to ridgepanel.com/demo.

Let's get into it. OpenAI's Counterpunch week, GPT-5.5 drops, and Workspace Agents ship. OpenAI dropped two frontier releases in two days, and once again, the top of the AI leaderboard flips. All right, we'll take you back to last Wednesday when an OpenAI API routing error did something that OpenAI would not have done on purpose. Developers on Codex opened the model picker and saw a menu that they were not yet supposed to see. GPT-5.5 alongside internal checkpoints codenamed Arcanine and Glacier Alpha. Somebody inside OpenAI panicked and the menu disappeared within minutes. But one day later, OpenAI made it official. GPT-5.5, codename Spud. Yes, the very Spud that we'd all been hearing about is now live. And the benchmarks say it took back the top of the leaderboard from Claude and Gemini, if you care about those things like Ben does. 82.7%. On TerminalBench 2.0, the highest score on record, ahead of Claude Opus 4.7 at 69.4% and ahead of Gemini 3.1 Pro at 68.5%. So this is truly state-of-the-art stuff. Of course, as always, there is one exception, and that is Anthropic's unreleased Claude Mythos preview, which still leads on 6 of 9 directly comparable tests. But some are saying that that model is just a bit too compute-intensive and a bit too expensive to release. All right. So the context here, no pun intended, underneath all of this, like, it really matters. OpenAI has been in internal code red for about 5 months since December 2025. That's Sam Altman's language, not ours, because Anthropic's Claude took the enterprise lead. We all see it. Let's be real. Everyone you know is using Claude. So OpenAI deprioritized projects like Sora and Shopping to focus on 5.5. This is their counterpunch, and it wasn't the only one. Last Wednesday, the same day as the Codex leak, OpenAI shipped Workspace Agents, a team-wide agent product that is replacing custom GPTs, which I love, for every business and every enterprise plan. All right. Two major drops in two days from a company that had not meaningfully shipped anything in months. Ben, what does this mean for the operators listening here?

So first I want to talk about the cost number because it's really easy to misread and there's been some confusion online. So OpenAI says GPT-5.5 is half the cost of Opus 4.7 and Gemini 3.1 Pro for the same intelligence. That's true if you're shopping these three against each other. But if you're already on OpenAI upgrading from 5.4 to 5.5, the actual cost is actually 20% more per task. The API price doubled. However, the model uses fewer tokens to finish the job, so it balances out at around 20% extra. Now, Workspace Agents, that's where the real shift is. You can now build an agent, share it across the org, and run it in ChatGPT or Slack, OpenAI shipped it with two really cool demos that show what these things actually do. Like one is an agent that watches your customer support tickets and also Slack channels and spots when multiple customers are complaining about the same product issue and turns it into a ticket for your product team. Another runs your month-end close. So it's actually pulling transactions, drafting journal entries, and reconciles the books, which I thought is incredible. This exists only on business and enterprise account, right? So individual plus users at $20 don't have access to this yet. What makes these agents actually work is computer use, right? 5.5 navigates software on its own, clicks buttons, it moves between apps, completes multi-step work without human in the loop. If that sounds familiar, Anthropic shipped the same idea in January with Claude Cowork, Craig's favorite. Which is you give Claude a folder on your desktop, it opens apps, fills spreadsheets, drives your browser basically. The difference is the entry point. Cowork runs locally on your machine for individual knowledge work. Workspace Agents runs in the cloud and is built for teams to share. So it's the same capability, slightly different use cases. So if your team builds custom GPTs, stop. They're being deprecated. Build Workspace Agents if you're on Business or Enterprise, and don't rip out Claude or Gemini Agents over a benchmark headline. Run one existing workflow on 5.5 for a week, measure quality against cost, then decide. That's at least how I would do it. One flag, Workspace Agent is free until May 6th, then credit-based pricing OpenAI has not detailed yet. So free until a future date from OpenAI is rarely generous. Uh, every agent you build in the preview is feedback data for them. So it's fine to use, just go in with your eyes open.

All right. So obviously Ben is getting giddy here and you shared a lot, right? Like, you know, agents are here, costs are going up, GPTs are done. There's a lot of noise in what was just discussed, but there's also a lot of merit. So I think two things that OpenAI has done here that will change the game, in my opinion, for enterprises is as follows. The idea that these kind of like agents and skills are moving off device and into the cloud for everybody to share and access together, that's a big deal and is going to be the future of this. And the other thing is the security settings, right? The fact that these can only be built at the administrative level is also a really big deal. You're not going to have individuals going off and creating their own things. But look, the reality is, is there's a bigger question here that is hanging over everyone who's listening. And that's, well, dang, I just switched to Claude. Now I've got to go back to ChatGPT. What's, what's happening here? The bigger question is like, how married do you want to be to each of these platforms? Because they keep one-upping each other every like week. It feels like one is getting better than the other. And for operators here, that causes like very serious FOMO, right? So this week ChatGPT takes the, the, the lead. Next week it's going to be Mythos and all that stuff, right? So. Every 6 months, you're staring at a tech stack with the company that you bet on, wondering if you picked the right horse, right? And I will tell you, having worked with like 90 companies on this stuff now, like you cannot make a bad decision. Just pick one, treat that as the gospel at the org-wide level and ride the wave. If you're a Claude shop, be Claude. If you're a ChatGPT shop, be ChatGPT. If you're Gemini, that's your one. And I can't believe I'm saying this, but like, I'm sure that Copilot at some point is just going to get good one day too. Like Microsoft doesn't stay out of the game. For that long, right? So just remember, whatever ChatGPT does today, Claude is gonna be able to do in the future. So yes, Ben is very excited and geeking out about the ability to sort of manage all this stuff in different ways. He is a token maxer. If that is you, go with God and make it happen. But the question for you is not which model is best this week. It is the one that your team can build muscle memory around fast. That is the most important and only thing that matters. We will go to this in a different segment, but people just need to be using the tools the same way and operating off of the same context and knowledge base. That is the whole game. It is not one tool for this and one tool for that. It is, are we leveraging this in the same way? That is what you need to be focused on as a CEO, not what crazy use cases people are kind of bragging about on X. Just get your team building the scaffolding and the shared context so they're operating off of the same tool in the same way. Ben, any closing thoughts?

No, I largely agree with you. I think it's a personal choice at this point, Claude or ChatGPT. I think the gains for each are, you know, minimal differences. However, if you are working through very complex workflows, I think it's still worth testing between the two and see where you get the better results and then implement that way.

No doubt. I often say, you know, ChatGPT is the moon and Claude might be Mars. Like, just going to the moon is okay. That's pretty good, right? Like, you don't have to go to Venus and Saturn all the time. So, all right, onto the next. Walmart fired OpenAI just 5 months after both CEOs stood on stage together announcing the future of AI shopping, instant checkout, buy Walmart products without ever leaving ChatGPT. That's the whole vision. Walmart quietly killed it. Conversion rates were 3 times lower than on walmart.com. Product data was scraped, not synced, so inventory details were not accurate. Delivery times could not be verified. The thing just did not work. So Walmart flipped the entire model. Instead of letting OpenAI own the transaction, they're now embedding their own Sparky chatbot directly inside ChatGPT and Gemini. That user experience really matters. The user sees Walmart's agent inside ChatGPT, not ChatGPT helping them buy things from Walmart. And the results of the flip are actually like pretty legitimately striking. Sparky users have average order values 35% higher than non-Sparky shoppers. I can't help but think of Sparky the dog in South Park as I go through all of this. So forgive me for smiling. When deployed inside ChatGPT, customers convert at roughly 70% of the rate that they do on Walmart. .com directly. Not bad. And orders of magnitude better than the instant checkout was doing. This is the new brand-owned agent playbook, and it is the template for every large retailer watching. Ben, you're the expert. What is going on here?

So Walmart looks at a conversion rate that's 3x worse on ChatGPT than on their own website. And they're basically saying like, wait, what? Why are we okay with this? Like, why would we let a generic checkout from OpenAI represent us? When we've spent decades figuring out how to sell stuff online. So like you said, they flipped it, right? They built their own agent, Sparky, and they just put it inside ChatGPT instead. And that's what people mean by own the agent, rent the distribution. So OpenAI is the traffic source. Walmart is the storefront. It's the catalog, the checkout, the customer relationship, They're paying to show up where the demand is, but every part of the actual transaction is on their own stack. And honestly, for huge brands, this, I think this might be where things head. If you're, you know, if you've got the data, the logistics, and the brand to justify building your own agent and plugging into wherever your customers are could end up being the playbook. It's a bit too early to tell though. For our brands at scale, I'm not building a Sparky next quarter, right? What I'm doing is making sure our product feed, inventory, shipping windows are clean enough that whichever agent represents us doesn't embarrass us the same way OpenAI embarrassed Walmart, right? Like wrong items in carts, stale inventory, wrong shipping windows. That's really how instant checkout at OpenAI failed. And it's also how other small brands can have terrible conversion rates in agent-led shopping without ever knowing why. So what's interesting here is what happened next, right? So OpenAI got the message. They're moving instant checkout into a new apps architecture where merchants control the checkout. That's not a coincidence, right? That's OpenAI watching their biggest retail partner walk away and giving merchants a version of what Walmart wanted. So for now, the transaction belongs to the brand. The platform is just distribution.

It is exactly what you said. OpenAI is the traffic source, right? 800 million weekly active users. That is where your customers are. And we've seen what they've tried to do here a thousand times before. They tried to own the entire transaction. It didn't work. Meta has been chasing exactly that for like a decade, right? Instagram Checkout, the Facebook Store, every flavor of in-feed commerce. It's no different than what OpenAI tried to do here. They want to own the point of purchase. And yet Meta lost every single round to Amazon and Shopify. It seems candidly that TikTok Shop is really the only one that, that has kind of figured this out. So Ben, to your point, like when OpenAI's instant checkout in ChatGPT sort of dies, and Walmart walks away, like, that's not new. It is the exact same fight and just a different flavor of what we've seen. OpenAI wants to own the point of purchase. It just didn't work. But there is a wrinkle here, right? When you've got 800 million active weekly users, that is incredible intent data. They know what your customer is motivated by. They know what they're shopping for, candidly, before the customer even does, right? It is intent data that is so rich it's not Facebook ads, it's closer to like Google search in, in 2008, right? And that is a prize worth fighting for that you have to play in that sandbox, Ben. It is exactly what you said. You have to make sure that your product inventory data is accurate to the minute and that teams are doing everything that they can to show up well in LLMs. I think all of this LLM search, LLM shopping, LLM visibility, that warrants a full-time hire, whether it's part of your R&D budget or whatever, you need someone whose job is just making sure that the brand shows up clean in every LLM on every platform alongside all the relevant technical details. And you ride the wave of agentic shopping. One last thought here. I am so interested, Ben. I'm, I'm curious for your take if, if you'll give it, of Shopify's role here, right? Like the front door of the internet in some ways is changing. Will every single brand have their own sparky over time that is driving conversion in ChatGPT and in Claude before going to, you know, the website? Is this kind of a new version of apps? Any kind of closing thoughts on like the structure and Shopify's role in this?

You know, what Shopify is trying to do is like, how do we connect to that data source, right? So ChatGPT, OpenAI tried to do checkout, they failed. That's not their strong suit. They're building LLMs and AI experiences. So if Shopify can connect to that data source, connect to that traffic source of ChatGPT, which I think they're trying to do, I think that could work really well for them.

I know we've got the tickers here of the Polymarkets and other things like that. I'm not a gambling person, but I think I might put some money on OpenAI buying Shopify one day. I just kind of have a hunch. All right, on to the next. Meta has decided it is going to cut 8,000 jobs as the introduction of the single very talented person era is here. Okay. So on Thursday of last week, Meta told its staff that it is laying off about 8,000 people or 10% of the company. And on top of that, even worse, it is killing about 6,000 open roles that they had been planning to hire for. So that is 14,000 jobs. Layoffs will hit on May 20th. So here's what makes this one worth talking about. Just like Block before it, Meta is not struggling. They just posted a record quarter, almost $60 billion in revenue in the quarter, you know, up up 24% year over year. They are highly profitable. So why cut 14,000 jobs? It is because Mark Zuckerberg is doubling Meta's AI spending this year alone. Last year they spent $72 billion on AI infrastructure. This year they will be spending close to $135 billion with a B. So these layoffs are just how he is funding that increase. Less people, more AI. And he said the quiet part out loud on their most recent January earnings call, right? He said, look, we are starting to see that projects that used to require big teams can now be accomplished by a single very talented person. You hear about this all the time, roles compressing. It is no different than Jack Dorsey's announcement about the Block layoffs. I'm going to read it verbatim because it is the same. We at Block are not making this decision because we are in trouble. Our business is strong, gross profit continues to grow, and profitability is improving, but something has changed. We are already seeing that AI paired with smaller and flatter teams is enabling a new way of working. Which fundamentally changes what it means to build and run a company. And that is accelerating rapidly. This is not just Meta. This is not just Block and the founder of Twitter. In the same week, Microsoft offered buyouts to 7% of its workforce. Amazon cut 30,000 jobs since October. Nike, my favorite brand, is laying off 1,000 people. That is about 92,000 tech workers that have been laid off over the first 4 months of 2026 alone. Ben. You are an operator, you lead a team of 150 people. What is going on?

Look, I mean, you said what's actually happening. Zuckerberg needs more money to fund the AI race, right? He went from $72 billion in CapEx last year to $135 billion this year. That money has to come from somewhere, right? And he found it by cutting 14,000 roles out of next year's plan. This continues the trend that we've been seeing and discussing from every other big tech company. Microsoft, Amazon, Block, Oracle, they're all signaling to Wall Street they can grow faster with fewer people. But I'd be careful applying this to smaller DTC brands. A lot of these big tech companies were overstaffed before AI even entered the picture, right? Meta had 87,000 employees in 2022. Amazon ballooned during COVID The cuts aren't just about an AI productivity story. They're partly just a return to a sane org chart. So if you're running a tight, efficient $20 million to $50 million e-com brand, you probably don't have 30% of your team doing redundant work. So the headline AI is replacing humans doesn't translate to every size company in my opinion. The takeaway for me is more about being thoughtful with the next hire rather than running a layoff playbook.

Ben, I couldn't agree with you more. And candidly, the reason I wanted to talk about this story is if we had talked maybe a month ago, I would have said AI is coming for absolutely everything. You see it here in these announcements, but you are right. I think that this is simply a return to sanity. And candidly, I am quite confident that AI will be a net job creator. The jobs will look different, but Mahin will show a tweet from Anthony Pompliano with some recent data, right? Like, Uh, you know, the number of college grads who are getting hired is increasing. Unemployment for college-age 20 to 24-year-olds is decreasing. So ultimately, yes, the types of jobs and the skillset that is going to be required to be successful in those will change, but that is not insurmountable. I think, Ben, you are absolutely right. For operators listening today, particularly for smaller D2C brands, this is not how can we lay off a bunch of people. It is quite the opposite. It is how can we do more? With the team that we have. And on top of that, as you are thinking about hiring, it's just being really strategic around the roles that you hire for now that the technology is here. Is the new role that you have, is it just a workflow that can be automated? If so, try that first before expanding headcount. The brands that I am working with who are doing this in the best way are just being strategic around the next new hires that they have now that AI is here. Okay, Ben, I am so stoked. To talk about this story. So the team at Semaphore wrote a profile on a guy named Bill Wen. He is a Silicon Valley founder who sold his company to Apple back in the day. He now runs a really cool voice startup called Olive, but the story is not about his startup and his work. It's about what he's been doing at home on his own time with AI. So Wen is paying tens of thousands of dollars a month to run multiple AI models at once. For him. He has created an AI version of himself. He has hooked AI into his life via the API directly, running Claude, ChatGPT, and Gemini all side by side, all day and all night. He has pointed that compute at one thing and one thing only, a personal version of himself. Or if you look at this just slightly differently, his own Samantha from the movie Her. So he gave the AI his Slack messages, his calendar, his call logs, his location history, his journals, even his computer changelog. He gave it his voice, and in exchange, the AI now runs his life. It writes his agenda in the morning. It emails people for him on his behalf. It books in-person meetings for him with people that he didn't know that he was talking to or had ever met. It listens in on conversations with his kids and tells him how to parent better. It buys things for him on his behalf. The line that made this pop for me was what he said. I didn't ask it to help me. I asked it to be me. And he is clear-eyed about why he is spending all of this money. He says in a year or two, and I agree, tokens will be cheap and everyone will be doing this. He's just paying to see what the future looks like a little bit early and a little sooner than the rest of us. Ben, what in the absolute heck is going on here?

Look, for me, the story is a glimpse of where everything is heading, right? So whether you're running a business or just running your life, the operating system is going to be AI agents handling the mundane administrative work, right? Learning your voice, your preferences, your context. So you spend your time taking projects from 80 to 100 or 120% instead of starting from zero every time. So something to think about for operators is how do I build that context layer for my business, right? My customer data, my finance data, my ad accounts, my Slack history, SOPs, brand voice. Right now it's scattered across 15 different tools. The work for me, at least this year, is connecting those data sources into one or several agents that can run the routine stuff, pulling reports, drafting responses, flagging anomalies, you know, prepping the deck. What you get back when that's working isn't just time, it's the top of your day, right? So instead of starting at zero on a P&L review or a creative brief, you're starting at 80%, right? Your job becomes the part that actually requires you, and now you have more time solving harder problems, being more creative, being more strategic. That's how I think about it.

Ben, I could not agree with you more, and candidly, we're gonna need to start to disagree a little bit. Yes, this story is deeply inspirational, but not because some guy handed his life over to AI. It's because he built the right infrastructure and scaffolding for the AI to have all of the relevant context that it needs. So in my day to day, I'm building brand brains for companies sort of left and right, but it's the exact same thing. It's building the same infrastructure, the strategic vision, the goals, the voice of the customer, and it's connecting to outside datasets, Meta, Klaviyo, et cetera, to be recursive and learn in real time. That's all this is. It's creating the right infrastructure for operators to then make sure that the AI has all of the information that it needs to be successful on your behalf. So as you listen to this story, just start to think about everything that Ben just said, which is how can we document all of the right standard operating procedures, the way that we do business in a series of markdown files across the organization so that the AI has everything it needs to be effective on our behalf. Okay, Ben, a story that you have been stoked to talk about. AI agents just closed nearly 190 deals for for one another, and the losers did not know about it. All right, let's take you back a couple months. In December, Anthropic ran a 1-week quiet experiment inside their San Francisco office, and they just published a write-up about it. They called it Project Deal. They picked 69 of their employees and gave each one of them a $100 budget and had Claude interview each person about what they wanted to sell from their closet and what they wanted to buy. I need to use this as I sell all my homebrewing equipment as we move to New York. The interview that Claude made captured every person's negotiation style, and then the humans just kind of stepped away. What actually ran for the next week was a Slack channel with 69 Claude agents, each one set up to represent its person, posting listings, haggling, and closing deals with one another. No humans were involved once the project kicked off, and by the end of the week, those agents had struck close to 190 deals across more than 500 listed items totaling over $4,000 in transactions. The employees physically traded the actual goods, a snowboard, a folding bike, a bag of 19 ping pong balls. I don't know why an agent would want that, but here is where things get interesting. Anthropic ran 4 parallel versions of the marketplace without telling its participants. In 2 of them, every agent was a Claude Opus 4.5, their best model at the time. In the other two, half of the agents were swapped for Claude Haiku, a much smaller, much weaker model. Same instructions, same prompts, just a different brain doing the negotiating. Opus agents, of course, won. Ben, no human was in the loop as always. What is going on here?

Science fiction. Uh, I, I love this story. There's two big takeaways for me. So agent-to-agent commerce actually works, right? 186 deals, $4,000 in real money. 46% of human participants said they'd pay for a service like that. Mind you, these are Anthropic employees, so maybe a little biased, but it's a little bit of a proof of concept, even though it's still pretty small and nascent. The other finding is that model quality is doing way more work in negotiations than the prompt is, right? So aggressive instructions barely moved the needle here. Whichever side had Opus walked away with more money than the ones with Haiku. right? 2.68. So $2.68 extra per item on average. In one example, I think it was $65 versus $38 on the same broken folding bike, which is crazy. Here's the part that sticks with me though. After the experiment, Anthropic asked everyone how fair their deals felt. The Haiku users and the Opus users rated their experience basically the same score. 4 out of 5 across the board, meaning the people whose agents got out-negotiated had no idea a smarter model on the other side had just taken them, right? So they walked away thinking it was fair. So to me, it's crazy to think about a future in which, you know, an agent representing you in a transaction or a negotiation of some kind, you know, the one showing up with the better model wins and the loser doesn't even know they lost. It's a little dystopian.

Yes, the model quality is exciting. The agent-to-agent commerce is incredible. I unfortunately am not yet smart enough to understand what much of that means, but I do know that these tools are currently incredibly powerful for negotiation on an individual level. Make sure that as operators, we are recording the conversations with our customers, with our colleagues, with advisors, and you can bring all of that rich context into the LLM of your choice and then say, hey, what really mattered to this person? What are they optimizing for? And then you can call in context from sales books. Chris Voss, Never Split the Difference. Robert Cialdini, Influence. Dale Carnegie, you know, How to Win Friends and Influence People. And you've got all of that expertise supporting you as you're negotiating and going back and forth with individuals, negotiating on their terms, understanding what matters most to them and positioning things in the right way. Ben, I have interacted with an AI version of you countless times as I have partnered with you on bringing you onto the pod. How does that make you feel?

Totally fine. I do that too. I, I talk to an AI version of myself as well.

Wonderful. Well, I hope the AI version that you created of me is witty and handsome and cool. Most brands have deployed AI across acquisition and creative, but the confirmation page, the highest intent moment in commerce, remains a dead end. Rockt Aftersell is the design system for the full post-purchase flow, unifying cart, checkout, post-purchase, and confirmation into one intelligent system. Rockt Thanks brings real-time AI decisioning to the confirmation page, powered all by Rockt Brain, which analyzes 1.95 trillion data points across 7.5 billion transactions annually to determine the next best action. The result, instead of generic offers, customers see timely, relevant partner offers from over 500 brands like Hulu, HelloFresh, and Venmo, generating pure profit per order at scale with zero ad spend and zero customer data shared. Brands like Sephora, Nordstrom, True Classic, and Jones Road have generated over $1.5 billion with a B in pure profit this way. Activate Rock Thanks at aftersell.com/operators. And unlock AfterSell's full optimization suite, cart checkout, and post-purchase, all completely free. We are so lucky to be joined by Matt. You know him as the host of The Operators Podcast. We know him as an amazing advisor and guide as we build this pod. Matt, what's up, man? Welcome to the show.

It is so good to finally be here. Uh, I've been a listener. And a watcher. Now I get to like hang out and chat. This is great.

Well, for those who don't necessarily know you quite yet, give us a little bit of background on who you are, what you work on, and then we'll get into the AI of it all.

Yeah, I guess most people watching this, or some people would know me from the other shows, right? So we have the Operators Pod, we have a Titan Series, but I've built a lot of brands over many years. I've been in this e-commerce game for 20 years. I currently own and run a holdco that has two brands. One's called Gila Case, one's called Lomi. And just overall, I'm, I guess like you could describe me as like consumer obsessed. I've grown up in it. It's all I know. Like it's my family's entire history is just like making and selling things. I love things. I like what they stand for in culture. And I somehow took a detour as a software developer when I was very young. And now this whole AI thing has like, just renewed life in me, and I'm very hopeful and excited about everything right now.

One of the reasons I'm so excited by this conversation is because both you and Ben are engineers by trade who then took on leadership roles, CEO roles for the companies that, that you founded. So let's get into it. As an engineer who's leading two brands, you know, you've talked about the insane productivity that's been unlocked by these AI tools. Bring us into the room at Pela, maybe walk us through one use case in particular that's been transformational for, for the team?

Sure. Okay. So the, I'll give you a little bit of context first for the people listening. So Pela is, we manufacture compostable phone cases and a bunch of other accessories like AirPod cases, wallets, like lots of stuff, right? All around a material that is ours. It's proprietary. We built it over many years. We also own our own manufacturing. That is a very important point. So For us in this industry, we have no inventory. Everything we sell is on demand, which means every product that you see, all 16,000 of them on our site, don't exist yet, right? They only exist when somebody buys them. Now, if you can imagine what that's like from an asset creation, asset management, merchandising, ad catalogs, it is a monster worklift, right? So like taking photos, making videos, getting those things into Meta and Google and everywhere else that all those assets have to go. So what we've built is we call it Signal. We, we basically built a set of tools, um, largely with Claude code. And, you know, there are some, I would call them like light agentic versions of this, but what it allows our team to do is generate product photography generate product videos, get those things into all the platforms we need them to in, in sort of like in a human in the loop way. So the AI goes out, makes all the stuff, brings it back to the team. The creative team can then look at things and creative direct instead of creative produce. The way that we viewed this was our team guys, like you ever talk to designers or engineers, they don't like making a hundred 50 ads a day. They don't like taking the same photograph 15,000 times. It is the equivalent of digital factory work. And we looked at all of this technology and just said like, this is an uninspiring job for somebody with your talents. Can we automate away all the uninspiring parts and then just put you guys into a place where you're art directing things? And that is work you love to do and you're brilliant at it. So that is like, you want a measurable, like, I think we've saved like 6,000 or 7,000 man hours of work a year. Right now with this thing.

It's unbelievable. And that's candidly the holy grail that I know so many folks are working towards on the AI for content. There's a couple of things, um, that I want to build on. I love what you said about sort of like, yeah, this is digital factory work. And candidly, we want to empower people to, you said creative direct and not creative produce. And that is what, when used properly, this technology allows for, candidly, right? It's focusing on higher level, more strategic work where some of the routine grunt work can be removed and candidly done better, right? So ultimately, as we kind of think about how you're staffing the org, you run two brands, there's potentially a lot of redundancies overlap. How are you thinking about org structure with the introduction of these tools alongside both, both companies?

Yeah, I mean, I think I, I've sort of taken to this term, the great flattening. Which is that I think companies, ours included, are just gonna get more and more flat and more organized around small teams. Um, and I think a lot of this, like, so for us, like running two brands, I mean, Ben, you know this, like running multiple brands historically was like unadvisable. I think if you go back far enough in the other pod, I would literally be the one saying, please don't do this. This is like a special kind of pain and suffering that you're signing up for. You're laughing 'cause you know it's true. I think the thing that, what we're seeing now, so like, think org structure, I believe actual structure is flattening and I think what's gonna happen is we're, we're gonna start to organize more around workflows and outcomes. And people are gonna be far more generalist. And I think that what's enabling that is obviously AI, cuz it's handling a lot of the context switching that used to be very difficult for us to do as a team. Right? So now I can have like one email person that can very easily handle 2 brands, could probably handle 8 because the context switching can be offloaded. That mental load is not hers anymore. It's, it's the machines. Uh, and then she again is more art directing and she's more taste and judgment and what should we do and a lot less like pushing the buttons and how do I do it.

Ben, I'll just ask one follow-up before I pass it off to you. I mean, that is the case, right? We hear about this great flattening. We heard from Cody last week, you know, we've, we've heard quotes from Mark Zuckerberg, you know, Jack Dorsey. Like, I'm very curious because I come from, you know, Crocs and Hey Dude. I was exposed to the integration, obviously multiple brands. It's, it's challenging. So when you have this great flattening, Matt, how is this going to play out? How long is it going to take? What is the transformation going to look like? You're changing the organization as the technology itself is evolving. How are you thinking about this from a tactical level over the, the structural years to, to come?

Dude, I, okay, so I mean, that's a really good question. Uh, 6 months ago I thought this was gonna be like the fastest set of change management transitions I've ever had to deal with. And now I'm like, oh no, it's the same as everything else that we've ever done. This is gonna take time because there's a lot of people. We got a diffusion problem, we got a training problem. We have like a trust problem. Like there's a lot of like, I don't know if you guys have heard this term, but like every company has, has an immune system and then society has an immune system. So like we all have antibodies and like these organizations, I think this is gonna take a lot longer than any of us would actually like it to. Uh, at the same time, I think that it, it happens quickly in smaller chunks inside of the organization. So like I've, what I have witnessed at least is that instead of these like big giant wins, it's been a series of like 1 to 5% wins. And the thing I described earlier, Signal was like, is many, many months of building and adding things on and refining. So it was never, it didn't start out as like, hey, I got this grand idea, this big thing that we're gonna build and we're gonna get it right. Like, I'm not gonna one-shot this thing. So it sort of stacked up over time and I, I don't have anything else to pull on right now that would say it's just not going to continue that way. I think it's just going to be this series of 1 to 5% wins that, you know, I'm a compound interest fan. I'm hoping that's what it compounds to is like something big at the end of the day. And I mean, we're already seeing it organizationally, like people are just doing more than they ever have before and they're having more fun than they ever have before. And I think that that's broadly a win.

I'll, I'll close with this and then Ben, it's your turn. I love this framing of small 1 to 5% wins that compounds. I'm fortunate to work with CEOs all the time and candidly, particularly amongst larger, amongst larger organizations, I actually hear the opposite, which is, hey, I don't buy into this. We tried something last year. It didn't work. We're just going to kind of sit it out. Right. And the technology is so different a year ago than now. So this idea of leaning in, riding the wave, small wins that compound, and your other point too. From a diffusion of antibodies and a culture, right? My friend Jamie is the chief people officer at Groons, and I was speaking with her, and we'll have to kind of bring her on the pod, but her comment was, we actually don't have an issue, but that's because we screen for bias to action curiosity at the top. The culture of the organization is such that this is embraced, and I actually think it happens there, candidly.

I agree.

I totally agree. Yeah. I, you know, and I think the other thing that's really fascinating to watch is, uh, as a software engineer, and I'm Ben, I'm sure you, You look at this too, like, I can't believe that we're in a world where technology is like composable and disposable. You know, like I used to make these decisions and I'd commit to something and it would be like a 1 to 3 year commitment. And I'm like, now I'm like, man, I don't know. It's moving so fast that whatever I make today is probably gonna be tossed away in a month.

I think for a lot of founders and, and operators, it's deceiving because the promise of AI is so grand. And then you come in and you try to do everything with it. And you often don't get the result that you want. So attacking it the way you're describing, chunk by chunk, 5%, 10% increases at a time, I think is the right way. That's how you build software too, right? We start with an MVP and then you expand on that. Great. Matt, I want to ask you about the day-to-day stuff and the stuff that you delegate to AI or to people that are using AI and then What are the things that you wouldn't delegate to AI or to a person using AI? What are the things that you enjoy doing yourself?

I mean, broadly, I am known for having weird amounts of like, I call it dead time. So like I've shown people my calendar and they're like, what are these 4 and 6 hour chunks of time that you are like, just all it says is dead. Right. And, uh, it is literally just blocks of time that I, I hold for myself to be bored, uh, to go for a walk, to like hang out with my friends, sit out back if it's nice. Like, I don't know what I'm going to do at that time. And that is my thinking time. That is like, that's where all of my best creative work comes from. I don't think I could, I don't see a world in which I outsource that to AI. I think for me, AI has been far more of like a like a thinking Sherpa, you know, like the analogy being like, I, I'll go climb the mountain, but it's awesome if I've got a Sherpa with me who can also help me with all the things that I just, maybe I'm not great at. I'm probably not as fit as the Sherpa. Uh, I don't know the mountain as well. My knowledge is definitely not as broad or deep in certain things, but AI has just been this incredible tool for me to do that level of, like, to do the level of thinking I want. Like, my job is largely judgment, right? Like I allocate capital, resources, like time, all that stuff, trying to figure out what to do next. And that has been the most valuable thing is like just having a partner that has infinite capacity, is always getting better, and helps me with the work that I love to do and that I'm probably most uniquely suited to do inside of my company.

I, I love the Sherpa analogy. It's kind of carrying your mental load. Uh, that's, that's a really good one.

So can I jump in for one second with that?

Yeah, go ahead.

I, I think Matt, just kind of building on this, I love that idea of thinking time, but more than that, you know, as an operator, you are no longer thinking alone, right? When you've got your Sherpa with you, it's got infinite context to help you in two ways. The first is you're able to upload all of the business context, all of the conversations that you've had, the transcripts around the topics. That you're thinking of. So it's grounded there, your brand, your role, et cetera. But then you're also able to bring in the context of experts. You don't have to think alone. You've got Walt Disney, you've got Jim Collins, you've got Robert Cialdini and other brilliant thought leaders alongside you if you manage the context right. So again, it's just an accelerant for you in your thinking time through that process. I, I, I love that, that workflow.

Yeah. I think that I honestly, I am not great at pulling in outside context yet. I think that's like, that's actually on my list of like, how do I figure out, like, you know, Karpathy's, uh, personal markdown-based knowledge system that he has.

The wiki.

Yeah.

I need, I need that, like something like that for myself. And then I'm also thinking of like, how do you build that for teams and workflows and like, how does interop, like how do companies work? I think that Craig, what you're hitting on is like super, super important. Uh, I'm not great there yet. I, I think that that's like a big area of improvement for me in the next, like, you know, 30 days or 60.

Uh, circling back to what you said about decision-making, um, I, I often say some of my hardest weeks is when I need to make a single decision in a week, and that single decision requires 40 meetings, 12 dashboards, right? Uh, me take a couple walks outside, take a break, talk to my wife, talk to some advisors, sleep on it, right? And then I make that single decision. Now with AI, that it shortens it a bit, right? Because you have a thought partner, you can offload some of your thoughts, you get maybe a new perspective, a new way to think about it. But at the end, you know, the hard thinking is still done by us, but this is a bit of a shortcut. So that resonates a lot.

Yeah, you know, I don't know if you've gone this like, have you gone here mentally yet? But I think a lot about just like my own capacity. So because these tools allow you to shrink that time, are we now able to make more decisions? Can we actually take on more or are we genuinely the limiter? Like I get exhausted if I try to like actually stick more load onto myself. So what I've found really interesting, this is just, again, my own personal experience is Having all that free time has turned out to be an incredible superpower with AI because I'm not trying to fill that time with more work, with more decisions. I'm actually just saying this is a really high quality, high leverage amount of time, but got more leverage out of it because I've got this thing now that can take it and shrink the, shrink the action part or help me actually get to a conclusion faster. So I do think that, I wonder if we're actually going to just see like, people think that like, oh, there's going to be this massive increase in individual productivity. I actually don't know. I don't know if we can get more productive. Where is the limit to an individual's mental load? You know, uh, developers have been doing this for years, Ben. Like, there's only so much good code you can write in a day before it starts to come out pretty crappy.

I think productivity will increase, especially for high-level knowledge workers, because I do think that that Sherpa analogy that I love so much, it's taking a lot of that load off. So you have a little bit more capacity to make maybe a couple more decisions a day. That's how I see it at least. But no, it's a really, really great point. Switching gears. So you're, Matt, you're a repeat founder, two exits, multiple platform shifts. What does the AI era feel like compared to like e-commerce, say in 2010, 2012?

I mean, to me, this is more like 2002, you know, like the, the post-dot-com. We've weeded out all the crap. What's left is like, these seem like durable companies. There was an excitement back then, like I wanna say like the mid-2000s, like the 2005, '06, '07, '08, like in there. Even though there was a financial crisis, like if you were working in e-commerce at that time, like I was at NetSuite in 2007 watching sort of like frontline, watching companies move now, like we're all gonna embrace this e-commerce thing. And the energy was incredible. There was not a lot of answers. Nobody really knew how a lot of things worked. We were still talking about putting like, you know, security badges in checkout to tell people that their credit card wasn't gonna get stolen. So I, I, I— this feels like those years to me where there's a lot of like trust issues and we're going really fast. Nobody really knows what's happening. It's like a Faffo moment. That is the, the era that I am like, it feels most familiar to me as the old man in, in the room sometimes. Um, you know, or, or maybe like the mobile thing, like when mobile really showed up and you saw like Facebook had to like burn their company down to rebuild mobile. All of us did. That kind of also feels like that. It feels like that now. And I, I sort of like it. I think I really do agree with this YC take that we have to start looking at AI as like, how are we building this as like the operating system inside of businesses? I, I do think that's directionally correct because it feels— that's what those eras felt like. You know, if you were a brand or a retailer, it was like, how do you become a digital first organizations. So like all your revenue might not be there, but it is definitely the, the pointy part of the arrow.

100%. This is where it's all leading to, uh, a new operating system for businesses, for computers, for personal work. Completely agree with that. Um, Craig, you had a question.

I have a couple things. One, I, I mean, your idea of this being digital sort of tied to You know, security badges at checkout. I love that. I talk to CIOs all the time and the threat that people feel from a security standpoint. Meanwhile, Walmart, Target, other kind of Fortune 50 companies are, are leaning into this aggressively. I just, I'm going to steal that as an analogy, but what I want to dive into, Matt, maybe you can answer this, maybe you can't, you know, you are uniquely positioned in that you own your manufacturing and you've got, like you said, 60,000 SKUs. Are there any AI use cases tied to either manufacturing or merchandising and inventory planning that you guys are leaning into as part of your 5, 6, 7% improvements that you're able to, to speak to?

Yeah, I mean, a couple. So we run a weekly demand planning cycle, right? Because we own the factory. So we basically have to create our own work orders for our own product where we try to anticipate demand. And I think that, uh, just bringing in like very basic things like having Claude Sanity check those plans every week, like all the work orders. Here's what we think. Here's the data. It's, here's the data we're making decisions off of. It's actually been incredible. Like it's, it's saved us money. It's made us faster. It's made us more productive. And that is just with like a head of supply chain using that in a very her own workflow. Um, another one, and this is for all of you like large catalog people who listen to this show. Probably my biggest individual win that I did was, uh, I built the equivalent of the TikTok For You page on our site in our bestsellers section so that when somebody shows up and starts scrolling, it reads the signal of the scroll. So like, do they stop on our product? Do they like it? Uh, do they look at the images? And then it starts to resort the page as the person scrolls based on like implied preference, right? That was like a 28% profit per session increase when we split tested it.

Incredible.

I did that with Claude Code in like a, in like a couple all-nighters on a weekend. And to me, I was like, okay, that was like my red pill moment where I'm like, oh, we live in the future. You know, TikTok is a $100 billion company because of that algorithm. And I built a very simple version of it for my one little business. and it ripped. So there's my pro tip for everybody.

I'm, I'm literally jaw-dropped by that use case. And when we talk about the threat that, you know, AI could pose to software companies, I, I mean, it's exactly that. There are poor sorting, you know, algorithms that, uh, you know, people spend tens of thousands of dollars a month on, and this is just something you built based on your brand. That's, that's truly mind-blowing. Matt, thank you so much for being here. Obviously, if our listeners want to learn more about you, hear your takes, et cetera, where can they find you?

Man, I'm on all of our podcasts. I seem to be everywhere right now in the show. So 9operators.com, you can kind of find me on X. I'm not as prolific as Sean or Cody. I don't like social media that much. I much prefer these things. So yeah, you can find me there.

Come back anytime.

Yeah, man.

Thank you so much.

Okay, welcome to the show, Arius Giri, the head of operations at Block Blue Light. Uh, Arius, tell us a bit about who you are and, uh, what your company does.

Awesome. Yeah, hey, I'm Arius. I'm the head of ops at Block Blue Light. We're a light-based wellness company based out of New Zealand, but we operate globally. We mainly focus on, um, lighting products such as, uh, light bulbs and red light therapy. And we also have a range of blue light blocking glasses as well. Um, we manufacture mainly in Asia and then we sort of, uh, have local regional warehouses, uh, that we service the, the world from. So, um, based in sort of the US, UK, New Zealand, and Australia, uh, soon to be Canada. And also, um, on the roadmap this year is for the EU as well, opening up there.

Very cool. Very cool. You mentioned multiple locations here. So walk us through BlockBluelight's supply chain in about a minute, like manufacturing, warehousing, shipping, and then tell us where AI actually entered that stack in the last 12 months.

Yeah. So, uh, like I say, China or Asia, we, we, uh, manufacture in and, uh, we ship to our local, uh, warehouses. So we have a consolidator in China that we actually consolidate from around 15 suppliers. In there and then we ship out to each regional warehouse. So that's a bit quite complex and sort of getting that step right. And then from there, the last mile is done to each, each local, to each country in our local regional warehouses. But then we also have rest of world shipping from some of those depending on where we're shipping to in the world as well. In terms of the AI side of it, it's been a big part of our business and how we're sort of moving into the future. You know, it is for everyone at the moment and it's kind of, we're all finding our feet and how that works. Personally for me, one of the biggest things I've done is created sort of an AI operating system, a personal one, which is helping me balance all the operational tasks that are coming my way as the business is scaling, the workload's growing quickly, you know, so we need to make sure that we're, you know, staying on top of all these tasks that come, get thrown your way really as you work in operations. So what I've created is is, uh, inside Cowork is basically, um, a personal, uh, I guess, AI and, and a business mentor that sort of runs my life across, uh, all my Slack inbox, um, and, and also all of the tasks that we're doing sort of on the day-to-day. But then also like the big strategic pillars and, um, you know, sort of big rocks, I call them, um, goals that the business has over the sort of coming 6 months. Um, so Claude basically sits across that and scrapes my inbox, scrapes my Slack. And then inside that, I have some like 3, 6 live files that Claude reads every morning. So it understands how I operate, how the business operates, all of the things it needs to know and my brand Bible, my voice, how I like to be talked to. And on top of that, it layers in all of the daily things, all of the big rocks that I mentioned and sort of like a handoff file. Whereas I'm working through day to day, it's updating those in real time. And then every morning I'll get a Slack brief from this Cowork file that will outline everything that I sort of need to focus on today, sort of the big strategic pillars and how they're tied into my day's work. And then anything that's come in across Slack, across my emails throughout the night, as I mentioned, we are, you know, based in New Zealand. So we quite often work in different time zones with our suppliers in Asia and and warehouses all over the world. So messages are coming 24/7 through Slack and through inbox and that. So having that morning brief for really can focus on the important things I need to get done immediately versus the things that can wait till later in the day is really helpful. And then what it does at the end of the day as well is gives me sort of a rundown of how, how I've been going and sort of reviews the day as well, which is really good. And then I sort of roll that up into a weekend review as well. And it kind of forward thinks of asking me questions, how I can be better and, you know, improve and ask me some uncomfortable questions so that I'm not avoiding anything that I should be not avoiding.

Incredible. It's like a chief of staff, executive assistant, and a mentor all in one. And when, you know, all of us operators, we have multiple moving parts and having it all in one place and having reports beginning of the day, end of the day, I think is super helpful. So that's very inspiring, Arias. Thank you for that. Moving on to talk a little bit about compliance and regulatory in ads and creative. So wellness is a highly regulated category where claims get scrutinized quite often. Uh, and, and so how are you making sure your team uses AI to run faster, but also that it doesn't get you in trouble with unsubstantiated claims in, in marketing assets?

Yeah, that's a good question. Um, you're right. The wellness category is a minefield and you can imagine with, especially with electrical compliance, cause a light bulb, for example, you know, you have to comply in every single country and every country has different rules. So there's literally hundreds of compliance issues you have to deal with, and sometimes they change quite frequently as well. So you're constantly having to stay on top of that. And we have people dedicated basically just to that within the business. AI, I guess, is helping you move faster, but it's not making any claims. So it's really helpful to do the research and scan for new things that may have changed and pull that information to the to the front for us, for then us to review and ensure that, uh, we're happy with any of those claims that go out to market. So it just speeds up that research process because that's what it is really good at is, is collating a lot of information fast and then bringing it to our attention so that we can review for the correct information before that goes out.

Arias, want to go back to your earlier use case, which is your, your personal chief of staff. I've created a similar thing. I want to focus on the end of day recap. So for me, I've got a sales coach. Where after every call I'm getting an email back from my sales coach telling me what I could and could, what I could have done better, what I should do differently next time. And it's personalized to me. And I just find the act of seeing that feedback is so helpful going into the next day. Can you take us one step deeper on just kind of the end of day recap and some of the value that you're seeing there?

Yeah, sure. So at the end of day, I get another Slack message and that pings me to go and have a chat to my coworker and And I kick that off and it filters down sort of 6 questions or 6, depending on how the day's gone around, you know, what did I ship today? What was blocked today? You know, an uncomfortable question, which is really gets that reflection moment going of what I'm avoiding. And it builds a picture and pattern recognition of things I have or have not been doing over the week, over the previous weeks. And it really starts to, like I say, sits on my shoulder like a high-level business mentor. I've done quite a lot of work in building out what I want it to ask me and how I want it to, to, to answer my questions as well and give me lots of pushback. So it understands how to talk to me in that space. And then it's collating all of that information and just making sure I don't miss any blind spots.

My—

oh, sorry, miss my blind spots because we're all prone to that. So having someone there that can really push you is really helpful. And then, and then the next step after that is basically performance monitoring that you can do on yourself daily, weekly, monthly, and you can roll that up really quickly so you can see how you've performed. You can talk to your, you know, your peers or your supervisor, your boss or whatever leaders and share that information quite quickly.

How did you build this, right? Are you a deeply technical guy? For people in supply chain who might be kind of like thinking, man, I'd love to get there. Like, I don't even mean the technical specifics. I just mean like, walk me through your journey. You weren't necessarily a cloud expert the first time you used it. On how you got to this specific use case.

Yeah, no, I wasn't at all. I literally followed a few people on LinkedIn, few, few videos. There's quite a few good ones out there. And, and setting up your Claude coworker is really important and doing that right first. So having a folder dedicated so that Claude is only working inside that folder and it's not going to other places. And then inside that, you control exactly what it can read and what it can write to. So that was the important part is, is okay, you are only allowed to read these files and you must do them before you have any chat with me. So you have all the context. I know, you know, we talk a lot about it, but feeding it context is the most important thing because you can give it all the information in the world. But if it doesn't have your business context and your personal context, it's not going to be super relevant to you and important. So that is the most important thing, getting that context right. And it's something that I worked quite hard on to get the markdown files correct for my personal context and business context. And then once you have that, you're just layering in everything you're doing every day at business. And then it really works quite well after that. And you just iterate and you build on it, you add to it and like an employee, right? You're just teaching them what they need to know and it just goes and goes and goes.

Preach. I could not have said it better myself. The importance of context, the importance of small, modest improvements over time. All right. So let's go back to the supply chain world of it all. So you live in New Zealand, you're a multinational business. So tariffs, customs, cross-border logistics, like there's a lot going on here. Is there a particular AI workflow that has saved you a lot of time and money over the last, let's call it 6 months?

Yeah, there has been a few. It is difficult with a lot of, you know, like you say, cross-border legal is another hard one because every country again has different legal standards. So one, like a really small example I'll give you right now is, is our Canadian 3PL we're about to stand up. So we're just doing some negotiation around the contracts and things there. And it was really helpful for me to sort of plug in, uh, what we had been sent as a sort of offer contract from them and compare that to all of our other 3PLs we have around the world. And, you know, the standards that we set and what we expect, run that through. And it gave me a really detailed explanation of what was good in the contract, what was bad, what we could push back on, all these sorts of things, and laid it out in a really simple form for me, which made it really easy for me to go through, highlight, and then push back with my sort of, I guess, commercial lens of what I've experienced in supply chain over the years and what we really need in our business and what's important to us. And I guess the light bulb moment for that came when I got an email back from our account manager in Canada. And he goes, man, our in-house counsel loves you because you just, you outlined all the points in the contract so quickly, so easy. It just made their job really simple to go yes, yes, no, no, rather than, you know, a messy email with, hey, Can you change this? Can you change that? And they're just, what section are you talking about? You know, you know, so it was really, that was a bit of a light bulb moment for me. And he, and he's, he said he hadn't said that before. So that was great.

I, I love it. So you've got a personal chief of staff slash mentor coach who's guiding you on your day to day. You've got negotiation support, et cetera. If we were to close with one use case that an operator in the supply chain world should start to leverage, let's say it's not somebody who's even interacted with Cloud Cowork yet. What would you guide them to their first step to sort of get to your level?

I think, you know, yeah, yeah. I think the personal command center is kind of what I like to call it is definitely one I think we should all look to use first because it's really easy. I started with just a daily tracker that just tracked what I was doing every day. So instead of, you know, I have a notebook and, and that, and I write down sort of things I need to do or things like that, but just start with a daily tracker, pipe everything into there. Either at the end of the day, you know, use your notebook during the day and at the end of the day, just pipe it in there. So you're not losing track of important tasks that need to happen and then build from there. That's kind of how I started. And then I layered on, you know, strategic visions for the company and everything else kind of came after the fact. It wasn't a one-go, I did this all. It was definitely over, you know, quite a period to build it up to where it is now. But I would say start with that, just a tracker. I think that works really well with AI.

Amazing. I will pass it off to Ben, but there's so much from your story that is so consistent, right? We talk about it, context, small refinements over time, start to use it in one way and then pattern match. Hey, it helped me as a personal advisor. Could it help with negotiation?

Wow.

It did that really well. Can it do this? And you're just building up the muscle memory to start to really expand from there. So Ben, I'll let you close at home, but Arias, thank you so much for joining us. I, I learned a lot.

Thank you.

Arias, last question for the operators listening. How do you roll out AI across your team?

Yeah, again, I think it's one of those things we're all still learning it. You know, you have the worries of security, what people use it for, what information they'll give it to. You know, currently we use Claude across the business and have the enterprise version. So a little bit more secure than, say, a personal account. But the big thing I'm on, the big thing for my team anyway, is I'm pushing them to use it every day and learn from it because it's the one thing I can guarantee if you're good at AI and using it, you know, you're going to have a job in the future. Well, we hope so. We don't really know where this is going and things are moving so fast. So just learn it, just get involved and start using it because, you know, we've heard this saying before, AI's maybe not coming for your job, but it's coming for the people that aren't using AI. So, you know, get in and use it, learn from it. It's a lot of fun. So I just encourage everyone to get their hands dirty and build.

Great advice, Arias. For listeners who want to get in touch with you, learn more about Block Blue Light, where do they go?

Yeah, blockbluelight.com is our primary website. You can also find me on LinkedIn as well. I started to share share a little bit more of my story and personal and work life on there. So you can just search my name and I should pop up on LinkedIn as well.

Arias, wonderful to meet you. Thanks for coming on the show.

Later. Awesome.

Thanks, guys. Great to meet you.

All right, everybody's favorite section, AI 101. And today we are going to talk about model context protocol. So in past sessions, we've talked about tokens, we've talked about agents, We've talked about skills, and today we are going to talk about the thing that brings them all together. Ben, WTF is MCP?

Great question. So a prompt is like asking a question. Context is the background AI already knows. A skill is a repeatable playbook. We spoke about all these things in previous episodes. MCP, Model Context Protocol, is the standard connector. That any AI can plug into to get live data out of your business. So real quick on why this matters versus a regular API integration. So an API is the existing way two pieces of software talk to each other. And when the company on the other end updates how their software works, which happens pretty frequently, that connection often stops working and breaks until somebody on your side rewrites it to match the new updates. So every tool, every connection, your team is on the hook. With MCP, the vendor builds and maintains the connection once, and then any AI plugs straight in. When something changes underneath, the vendor fixes it, not you. That's why MCP took off so fast. Anthropic shipped it in November 2024. OpenAI adopted it across their API. Google rolled it out into Gemini. And last December, the spec got handed off to the Linux Foundation, so it's no longer just one company's project. Anthropic concrete operator example. Say Matt at Pela wants to ask Claude, what's my reorder point on cases in UK warehouse this week? Without MCP, that's a custom engineering project. Someone has to build a connection between Claude and his inventory system from scratch. With MCP, he flips on a pre-built connector. Triple Whale has one that exposes blended e-com performance, ad spend, profit numbers. Klaviyo has one for email and SMS data. Claude plugs in and your questions get answers from live data. Now MCP works everywhere AI does. It's in Claude.ai, Claude Code, ChatGPT, Cursor, Gemini. Basically every serious AI client supports it. The easiest place to start is Claude.ai itself where there's a built-in connector library for Pro, Max, Team, and Enterprise plans. Klaviyo is one-click, every brand has it, the data is clean, you may want to start there. The thing worth tracking, start asking your other vendors, your help desk, your 3PL, your ERP, do you have an MCP server? If they do, you can connect cloud to basically no-code, right? If they don't, you need to push them on that, right? Early MCP support is becoming a real competitive mode for vendors and SaaS makers because they understand they need to allow their customers to use AI easily with their platform.

So Ben, if I'm understanding things correctly, based on everything you've said, you've talked about MCP, Model Context Protocol, being about connectors. So ultimately, am I thinking about it right in that effectively it's like the pipe that allows these AI models like Claude to connect to other datasets? Is it that simple?

Exactly right. And the beauty of it is that you don't need to maintain it. The people that create the MCP on the SaaS side need to maintain it. So it breaks, I would say, less often, put it that way, than API connections.

And that was it. Episode 6 of the AI Operators. Thank you listeners for joining. If you appreciated the conversation, please like, rate, review, subscribe. Thank you for listening. Thank you to Matt. Thank you to Arias. Thank you to Finn and Mahin and Sam. I am so sorry. I know you made a lot of time to attend this one to talk about new ChatGPT 5.5. I am sorry we left you in the waiting room. Next time we will have you. Ben, any closing thoughts?

No, excited to see you all next week.