JP Beeghly and Josh Maynard have a blunt argument for anyone rolling out enterprise AI: treat it like a junior employee, not a magic tool — give it access, train it on your terminology, and supervise the output. They make the case that dimensional modeling isn't obsolete in the age of LLMs, it's more important than ever, especially when a system is writing its own SQL.
Behind the Expert
JP Beeghly and Josh Maynard both returned for a second appearance — the two were guests on the very first episode of this podcast, about a year earlier. In that year, JP's role at Sonos evolved from a CRM-focused marketing position into martech leadership, where he now owns the platforms and strategy behind how Sonos delivers customer messaging — a role he jokingly calls "senior manager of sweeping floors of martech," and unofficially, Chord's "chief question officer" for how heavily he's used its AI-powered commerce copilot. Josh made a bigger jump: from Head of Technology and CTO at home-goods brand Ruggable to GM of Commerce at Beast Industries, the commerce arm built around Jimmy Donaldson's MrBeast media empire, where he's about six months into building a still-nascent commerce business behind an already massive audience and content engine.
The Quick Hits
- Both guests describe training AI the way you'd onboard a new hire — give it access, teach it the business's specific terminology and shorthand, and grow it from simple tasks toward more complex ones — rather than treating it as an instant expert or handing it the hardest problem first.
- The real bottleneck isn't the model, it's data structure. Maynard argues that structuring data so an LLM can write reliable SQL against it now matters more than the raw data itself, since two people uploading slightly different spreadsheets to the same tool will get two different, untrustworthy answers.
- "Context" is becoming as important as data, and the two are getting conflated. For Beeghly, context means the company-specific quirks, acronyms, and historical exceptions that turn a generic AI answer into a brand-specific one — the difference between a Sonos-specific answer and a Beast-specific answer built on similarly-shaped transactional data.
- Sonos keeps individual AI tool choice locked down centrally for data safety, but doesn't try to reconcile everyone's separate AI conversations after the fact — strategy gets agreed on by humans in a room first, and that shared, human-approved context is what then gets used to prompt AI.
- Both are skeptical of "AI slop" as some new category of risk, and land on the same fix independently: treat an AI-generated deliverable exactly like a junior engineer's Stack-Overflow-sourced code — check it for context and editing, because it can look functional and still fall apart at scale.
- Maynard's advice to someone early in their career: "mine for conflict" — use AI to produce a first draft (a go-to-market plan, a piece of code), then take it to someone senior and ask them to explain or critique it, rather than treating the AI output as finished work.
- Neither wants AI to replace the human "conduit" between brand and customer. Their shared worry isn't job loss — it's a future where brands get treated as fully automatable, single-person operations, when what actually built the "iconic" brands they work for was specific, hard-won human context and curiosity.
The gang's back together — a year of role changes
A year after their first appearance, both guests have moved. Maynard left Ruggable, where he was CTO, for Beast Industries, running commerce for the media empire built around Jimmy Donaldson's MrBeast brand.
Beeghly, meanwhile, stayed at Sonos but watched his role evolve from a CRM-focused marketing position into martech leadership — and became, in Mahoney's words, Chord's "chief question officer" for how heavily he's used its AI commerce copilot to interrogate Sonos's own campaign data.
Business and technology are converging — AI is the accelerant, not the cause
Both guests pushed back on the idea that AI itself is why business and technology functions are merging. In their view, better tools did that first — AI just sped up a convergence that was already underway.
Beeghly agreed that AI's role has been to accelerate a shift that was already happening as marketing platforms matured — not to originate it.
Superpowers for the humans in the loop, not replacements
Both guests returned to a theme from their first episode — using AI to give the people between a brand and its customers "superpowers," rather than automating those people out of the loop. Beeghly's live example: using AI to iterate quickly on messaging for leads the business doesn't yet know much about.
Maynard echoed the augmentation framing directly, describing his own go-to-market work as dramatically faster without feeling automated away from him.
Telling AI slop from real craft
Pressed on how to tell a genuinely good AI-assisted output from "slop," Maynard drew a direct line back to his engineering background — the same judgment call he used to make about junior engineers copying code from Stack Overflow.
His conclusion: it's the same teaching moment engineering managers have always had with junior developers, just supercharged — you can still tell when something hasn't been edited, and the fix is the same as it's always been, showing someone what good looks like.
Structure your data before you trust the SQL
Maynard described a moment of genuine doubt about his own field — wondering whether dimensional data modeling was obsolete now that AI could seemingly generate its own structure. His conclusion went the opposite direction.
Context, not just data, is where the next advantage sits
Beeghly framed the next competitive layer as company-specific context — the acronyms, exceptions, and institutional shorthand that make an AI answer feel like it actually understands the business, rather than a generic model output.
Maynard, asked directly whether an LLM is even needed for everything, drew a firmer line — some questions deserve a pre-approved, deterministic answer rather than a generated one.
Hiring AI like a junior employee
Asked to extend the hiring analogy — onboarding AI the way you'd onboard a new person, with access, training, and a growth path from junior to senior — both guests admitted they probably aren't doing enough of it themselves.
Advice for people early in their careers
On what someone with a year or two of experience should do to build real judgment rather than just AI fluency, Maynard's advice centered on deliberately seeking out senior scrutiny rather than trusting AI output on its own.
What they'd tell their daughters about AI
Both guests are fathers to daughters, and the conversation closed on how they think about AI outside of work. Beeghly's advice to his college-age daughter, who's studying to become a veterinarian, was to use AI to sharpen her own understanding rather than outsource it.
With his 14-year-old, who just got her first phone, his focus has been media literacy about AI-generated content itself — helping her recognize when what she's seeing on TikTok has been shaped by generative tools, not just consumed as-is.
Maynard, whose daughter is eleven months old, framed his advice in more general terms — less about any specific tool, more about the disposition he wants her to have toward whatever "AI" ends up meaning by the time she's old enough to ask.
Sound Bites
- “I think it's all about context and editing. You could tell when a junior engineer was just copying from Stack Overflow, and the function didn't quite make sense contextually with the rest of the application, but it sort of worked. It certainly wouldn't work at scale — things start to break down.”
- “I woke up one day and thought, is dimensional modeling dead — do we even need to structure our data anymore, can't AI just do it? What I realized is that the structure of data matters more than ever, because we have to train the LLMs.”
- “I've literally uploaded Google Sheets and Excel files of data, and if my dataset is slightly different than somebody else's, we're going to get a different answer. Having a centralized repository, structured in a way that's optimized for an LLM, is incredibly powerful — organizations that adopt that quickly are going to win.”
- “It's a relationship built on language — you have to find the common language between the data you have and how you speak to it. That's what we're finding when we test Chord — when I start being very basic in how I phrase things, it has a harder time understanding.”
- “If I were to spend more time training it, I could have these good answers. I don't think we're doing enough to train it. You would never hire somebody and immediately give them the most complex task, without them first proving they can do the simple tasks well.”
- “The advice I'd give somebody early in their career is to mine for conflict. What that means is ask questions constantly. Take the thing you had AI write for you, and go to somebody more senior and say, what is this, how does this work, what does this mean — I actually don't even know.”
- “You don't use AI to help you learn — that's not going to work that way. What you do is ask good questions, like, how can I memorize this in a strategic way that will help me do better on the test. Don't think about it as cheating — think about it as helping. The goal is to learn.”
The Chord take
Beeghly and Maynard run commerce and data functions at two very differently "iconic" brands — an audio hardware company built on decades of engineering reputation, and a media empire built in a few years on one creator's ability to make virality feel kind. Despite that, they converge on the same non-obvious conclusion from a year of hands-on AI use: the discipline that actually produces good results isn't AI fluency, it's the same discipline that's always separated good work from sloppy work — structured data, checked context, and someone senior enough to say "this doesn't hold up" before it ships. The hiring analogy both guests kept returning to is doing real work, not just serving as a metaphor. A junior employee who's given the hardest task on day one, with no context about the business's specific quirks and history, produces exactly the kind of confidently wrong output both guests worry about from AI — the difference is that a human junior employee's inexperience is visible, while an AI's fluent, well-formatted answer can hide the same gap in understanding. Their shared fix is the same in both cases: onboard deliberately, keep a human as the final editor, and treat curiosity and asking-for-review as the actual skill worth building — in an employee or in a prompting habit — rather than treating either humans or AI as something you can just plug in and trust.
Put it to work
- 1Before trusting an AI-drafted deliverable — a go-to-market plan, a piece of code, a customer message — apply the "junior engineer" test: would this make sense to someone who understands the full context, or does it just look plausible?
- 2Audit whether your organization's data is structured well enough for an LLM to write reliable SQL against it. If two people uploading slightly different versions of the same spreadsheet get different answers, that's a data-structure problem, not an AI problem.
- 3Document your organization's specific context — acronyms, historical exceptions, the "why we do it this way" — the way you'd onboard a new hire. That's the layer that turns a generic AI answer into one specific to your business.
- 4If you manage people early in their career, encourage them to "mine for conflict": use AI to produce a first draft, then take it to someone senior and ask them to explain or critique it, rather than treating the AI output as finished.
- 5Decide deliberately whether your AI guardrails are only about security — which tools, whose data — or also about consistency of output. If everyone's using a different assistant with different context, the same question will get a different "right answer" depending on who asks.
- 6When introducing AI to someone learning something new — a new hire, a student, a kid — model Beeghly's approach: use it to ask better questions, not to skip the thinking. The goal is still to understand, not just to produce an answer faster.
Full transcriptShow ↓
Bryan Mahoney
And now you're at this company — I had to look up to see what they do, Mr. Beast.
Josh Maynard
We're selling more than just a product. We're really selling lifestyle, we're selling movements. This idea that Ruggable was all about alleviating stress — how do we live a stress-free life. Moving to Beast, it's also about how do we make kindness viral.
Bryan Mahoney
Yeah — like, a real person did this. I don't care how they got there, I'm shocked at how fast they got there and how good this is. But I know that someone truly cared about doing this, and that's going to have an impact downstream for our customers.
JP Beeghly
How do we use AI to help us do more with less? I think the roles hiring is always the same — you find the right person for the role you're hiring for.
Bryan Mahoney
It can be incredibly powerful, but it's not a shortcut. It's about being really curious, being discerning, and using it to learn — not to cheat. Welcome back to another episode of the Brilliant Commerce Podcast. This is not déjà vu — this afternoon I'm sitting almost in an interview-style setup with JP Beeghly and Josh Maynard, who you may remember were my guests on the inaugural podcast that kicked us off. That was about a year ago now — I can't believe how fast time has gone.
Josh Maynard
The gang's back together.
Bryan Mahoney
The gang is back together. Different room, same cast of characters — well, one of us has changed. We just had sort of a hot debate about agents. We weren't talking about agents a year ago — we're talking about them now, so there's tons to unpack. I brought notes this time — last time, guys, we did this in a single take, no edits. Who knows what'll happen this time. But I'm sort of betting on us. All bad jokes aside, thanks for coming back, and thanks for doing this one in Santa Barbara — last time we did it down in Los Angeles. It's great to have you back.
Bryan Mahoney
Really excited to dig into this conversation and just sort of catch up generally on where you guys have been, and where you're going.
Josh Maynard
So, yeah, welcome back — thanks for having me back with you.
Bryan Mahoney
Yeah, yeah, I appreciate that you color-coordinated. I also like that the laptop's out — I feel like the hard questions are coming.
Josh Maynard
That's true. Yeah.
JP Beeghly
We spoke.
Bryan Mahoney
We cool.
Josh Maynard
We prepared a little bit for this time.
Bryan Mahoney
Look at that.
Josh Maynard
Not last time.
Bryan Mahoney
I was just afraid I was going—
JP Beeghly
—to look worse than you.
Josh Maynard
That's never happening.
Bryan Mahoney
You were worried about that last time, and you both look equally good. Okay, let's do a quick catch-up in context. JP, you're at Sonos — you were at Sonos last time and you continue to be at Sonos, but your role has evolved. One of the things I promise people who tune into this podcast is that we'll take a look behind the scenes at what goes on at iconic brands, and for me, Sonos remains one of those iconic brands. But your role has changed a little bit since we last spoke — catch us up on what you've been up to and what you're excited to be doing next.
JP Beeghly
Yeah, I wouldn't say my role has changed — it's probably evolved, but it's been sort of a natural progression. I was sitting on a marketing-focused team working for what we call CRM, which is really focused on the customer. As I started taking more ownership over the platforms we use, and the strategy for how we're going to deliver those customer messages from a technical standpoint, it started becoming a martech role. And I don't know — martech, senior manager of sweeping floors of martech.
Bryan Mahoney
Also, JP proudly wears the title of our — what did I write in my notes here — our chief question officer at Chord. So we launched a commerce copilot, our first agent. Let's put a pin in the agent conversation for now, but our first agent that's there to act on behalf of our customers — one of the things it's good at, and getting better at, is answering questions. And there's no one who has asked more questions than you. My team loves watching your questions come in.
JP Beeghly
And so I think it's interesting, and I really appreciate Chord — I think it just speaks to the paradigm shift we're having in how we think about data and understand what's happening in campaigns. Even at the simplest level, we're just thinking about a campaign report — now that we're thinking about that in a complex way, it's not as easy to get that data. When you ask Copilot a question, it has to reach into a lot of tables to really understand that question. I think it's fascinating to watch the results, and I also think it poses another challenge — there's a lot of data.
JP Beeghly
However, for something like that, a Q&A process — discussion is around how much time is saved, efficiencies you gain, and that's probably true. But there's also a lot of institutional knowledge in the people that I build the data model under. So it's been fun testing Copilot, really testing it against our internal knowledge center.
Josh Maynard
Yeah.
JP Beeghly
And, you know, sort of collaborating with you to hopefully develop that into a really great platform.
Bryan Mahoney
I think you're being very gracious — you're doing more than testing, I think you're helping us build it. Josh, last we spoke you were Head of Technology, CTO at Ruggable. You've changed roles — Ruggable, I still put in the category of an iconic brand, and you're proud of all the work you did there — and now you're at this company, I had to look up to see what they do, Mr. Beast. And you've shed 'technology' from your title, you're now GM of Commerce for Mr. Beast. I think that's a fascinating switch, seeing more and more technologists get a little bit closer to business, or business getting closer to technology. So you're sort of leading, I would say — but yeah, how's it been? A bit of a whirlwind for you, I'd imagine, landing in the Beast empire.
Josh Maynard
Yeah, it's been wonderful. I love Ruggable, and I'm very proud of the work we did there, and the work the team continues to do now. But it was time for me to make a switch and figure out the next move — and this has been great. I think both brands are — we're selling more than just a product, we're really selling lifestyle, we're selling a movement. This idea that Ruggable was all about alleviating stress, how do we live a stress-free life. Moving to Beast, it's also about how do we make kindness viral.
Josh Maynard
Mr. Beast — Jimmy Donaldson is the best in the world at virality, and it's been a very cool mission to get behind: how do we blend ambition and kindness in such a way that I don't think a lot of other brands are doing, and prove to the world that you can be successful not at the expense of being kind — that those two things actually augment each other. I've been very lucky in my career to work now for two companies where I really get behind the brand. So it's been wonderful, I've had a great time — I'm about six months in, and we have a lot of work to do, but the work is really fun because—
Bryan Mahoney
The commerce business is nascent.
Josh Maynard
Right.
Bryan Mahoney
Like, you're trying to figure out how you can take that virality of content and translate it into commerce — am I getting that right?
Josh Maynard
Right. Yep — it's 100%, and it's not new, it's been around, but I think we're rethinking positioning, what we stand for, and then figuring out how we can—
Bryan Mahoney
So 'nascent' isn't fair — what I mean is leveraging what you ultimately have, and figuring out what it can become within this — what is ostensibly an empire now — of kindness. I love that.
Josh Maynard
Yeah, it's great. I mean, how lucky am I — I'm a technology guy who's now surrounded by the most creative and brilliant people in the world.
Bryan Mahoney
And well, maybe we jump in there, because I don't know if you guys are feeling it, but for me — I've been in technology for the last 30 years, and I think about these different evolutions. It used to be there was no tech, and then it became technology, and technology was always there to serve the business — the business wants this, so tech is going to do that. Recently, as recently as the last 18 months, exceptionally so, it feels like business and tech are becoming one. I think we're seeing that in your role — you're inherently someone who's really technical sitting in a business role.
Bryan Mahoney
And JP, it feels like the business teams, marketing teams, and technology teams are getting closer and closer together at Sonos too. Do you guys think that's because of AI — which we didn't really talk about on the first podcast — or is it just an inevitability that these two worlds were going to collide, and AI is just something you can't walk two steps down a sidewalk without hearing about?
Josh Maynard
Sure, I'll start — no, I don't think it's AI. I think AI has helped speed it up, but I've always said that at the end of the day, e-commerce is technology — it's all code. And we've seen these e-commerce tools get better and better over the last decade, and because inherently what you're doing is using technology to build product, I think the closer you can be to the technology, the better you can run them, because you understand exactly how things were working in the back end.
Josh Maynard
I think because Shopify is incredible, and because Playbio and Rise and all these other wonderful marketing platforms have allowed technology people to step out a bit and understand the business better, and vice versa — it's allowed business people to step into the technology in a way they weren't able to do before. So I think just because the tools are getting so good, the two are converging.
Bryan Mahoney
Yeah.
Josh Maynard
And AI is just speeding that up.
Bryan Mahoney
It's an accelerant. Yeah — easy, agree 100%.
JP Beeghly
I think there's — the technology is just advancing at a pace, and when you're in a space, technology tends to be technology-forward, but as it evolves it really becomes purpose-forward. Technology is really serving the purpose of why we're in it, on the marketing side — marketing platforms are now focused on people using them and really being able to implement great ideas with them. So I think AI — I don't know if AI's even really changed anything yet, that conversation. I don't know that it's really sped it up so much as it's just happening anyway.
Bryan Mahoney
Yeah, yeah, I would agree. This is a great segue, because I wanted to talk about humans in the loop, and superpowers — I went back and looked at the transcript from our first conversation, and we did talk about how you could use data to give the people in the loop — the people between the brand and the consumers, our customers, our advocates, our champions — superpowers. I've been talking about this for a really long time, but I think practically it was kind of difficult to realize. One of the things I'm seeing now with some of these tools is — I'm not excited for a world where AI replaces me as the conduit from brand to customer.
Bryan Mahoney
But I am excited to use AI to make it easier for me to connect with my customer, to give them a better experience. So as I think about accelerants, and about this human-in-the-loop problem — I'd be curious what maybe you've seen, or what you're experimenting with, that makes those superpowers ultimately possible. Or put another way — if we move beyond the idea that it's not just about data, it's about tooling, and the accessibility of that tooling that allows these superpowers to actually develop within an organization, so we can deliver better experiences.
JP Beeghly
I think a lot of the impedance to getting things to market is the amount of time it takes. Where I find and see a lot of use for AI, including myself, is being able to work through ideas very quickly, to get to something that lets you write a script to communicate with all the other stakeholders you're going to have in building that project. I found it really useful just working with you on some of the ideas we have with Chord, developing an idea where we can use anonymous identified profiles in a safe way that will hopefully generate new customers.
Bryan Mahoney
Yeah.
JP Beeghly
If they aren't the actual customers we intended to reach, we could use AI to be more human about it — let's test it, let AI generate a few versions of the same message, let AI figure out which person gets which message, and over time understand which one works better for particular characteristics. I could work through that concept in a day rather than a couple of weeks.
Josh Maynard
Right.
Bryan Mahoney
Yeah, well — the thought partner.
Josh Maynard
Yeah, I agree. I'm more of an inherent optimist, I think, so I'm not an AI-doomsday person or anything like that. I think AI augments, it doesn't replace — it's helped us speed everything up. I can write a go-to-market strategy in five minutes now, which used to take me a day at least for a draft.
Bryan Mahoney
Right.
Josh Maynard
And it's incredible how fast it's allowed us to iterate. To your point, I like what you said about people having superpowers — I think AI has just allowed everyone to potentially do five, or ten roles. Which also, as a leader, means I now expect a lot more from people — if my go-to-market strategy doesn't get to me within a couple hours, I'm like, why?
Bryan Mahoney
Or — can you tell if that go-to-market strategy you got is slop? That's what I'm struggling with right now, and I'd love your point of view — is it a superpower that's being misused, or a superpower that's been refined, where something of quality has actually been delivered to you? Because you're right, you can create a go-to-market strategy faster and better than ever before — but do you know what a good go-to-market strategy looks like? How do you know what's AI slop, which I'm seeing a ton of right now — versus something where you go, yeah, like a real person did this — I don't care how they got there, I'm shocked at how fast they got there and how good this is.
Bryan Mahoney
But I know that someone truly cared about doing this, and that's going to have an impact downstream for our customers.
Josh Maynard
Yeah, I think it's all about context and editing. This problem doesn't seem any different to me than back when I used to code every day — you could tell when a junior engineer was just copying from Stack Overflow, and the function didn't quite make sense contextually with the rest of the application, but it sort of worked. It certainly wouldn't work at scale — at scale, things start to break down.
Bryan Mahoney
Right.
Josh Maynard
So I guess, going back to being an optimist — I just think this is the same problem, but it's supercharged. Yes, you can still tell bad — you can tell if somebody hasn't done editing. But that's no different than the teaching moment that already existed with Stack Overflow. So it's just a teaching moment — you work with whoever sent whatever they sent, and say, this is how I would edit it, this is what I want it to look like in the future, let's try to make it.
JP Beeghly
Yeah, you're not cheating — don't hide the fact that you use these tools. In fact, no — I celebrate the fact that you're using tools to do it.
Josh Maynard
You must use them.
JP Beeghly
Yes, yeah, I agree with that. I think we're at this stage of AI where we're asking how do we use AI, rather than talking about how do we achieve the end goals and, as we always did, choose the right tools to get there. Just thinking about writing, I think it's an interesting one — I have a boss who's probably the best writer I've ever witnessed, she can really just convince people with an argument really well. I've tested AI against her writing, and every time, she has corrections that always make it better. I think that speaks to the fact that AI is a tool — in the end there's not a human replacement there, it's a tool.
JP Beeghly
And if you try to pull that tool up to be human, you find that it doesn't interact with humans very well.
Bryan Mahoney
Yeah, it is brand and dashes for me — this content, and this content are my own, and that's getting a chuckle.
Josh Maynard
A lot of people — I had like an HR line too. I've had to tell ChatGPT to stop using emojis multiple times — please stop using emojis, no, the telltale sign.
Bryan Mahoney
But I think that's really important — there's this sort of training. I think we can agree that what's happening in organizations now is that most people are using something like Claude, or they're using ChatGPT — that's their thought partner, their individual thought partner that they're training. Josh, you sort of said it's all about context — as you're thinking about the enterprise or the organization, how do you try to get people on the same page in terms of context? Do you imagine these thought partners as being better if everyone's operating from a shared point of view — in terms of the nuance of a brand, or our voice, or what's acceptable — like what data we have access to? Because I think that's a really hard problem that not enough people are talking about.
Bryan Mahoney
I'd just be curious — is Sonos tackling this right now? Is there an AI task force? How are you thinking about this at Beast? You know — capturing everything that's in Jimmy's brain as context, so that if the team is using some of these tools, that brand voice comes through. So, yeah, JP — what does Sonos do? Publicly traded company.
JP Beeghly
Yeah.
Bryan Mahoney
Is AI scary about that?
JP Beeghly
But yeah, AI is definitely a main part of the conversation these days — it's a main part of how we run these days. We have AI as part of the discussion, part of the tool set, from top to bottom. We do have an AI task force, probably living in, I'd say, two sides of the business — one on the business/marketing side, one on the technical side — really making sure that everything we use AI for is safe, contained, and ultimately serves the customer best. Making sure it's well-guarded within the data that we know serves the customer best. So I don't know if that answered the question, but that's where we are today.
Bryan Mahoney
So you're thinking about it mostly from a security point of view, but in terms of training it so that — instead of you individually using ChatGPT and your colleague using something else, and someone else, because these models are non-deterministic and everyone's getting a slightly different version of the same answer — is that a problem you're trying to solve today?
Josh Maynard
No.
JP Beeghly
And actually, it's controlled from the technology center — so we roll these things out, you don't have permission to just use whatever AI tool you want, because we protect our customers' data.
Bryan Mahoney
Right.
JP Beeghly
We make sure that whatever we use, internally or externally, is extremely safe for all purposes — that's one control. But really, the control around the message of the output is done by humans. It's a top-down story — the leaders of our company say, this is our strategy, we all sit in human sessions on Zoom or in person together, we really grok that, and we walk away with that, and then we use AI to write our next go-to-market strategies built around that leading strategy. That's it.
JP Beeghly
So I think as more people use it, I don't know if it's understanding what my colleague has written over here and what I've written and what another colleague has, and sort of putting that all together — but I'd always start with, this is what I want to do, and this is how it meets our current strategy, and let AI go from there.
Josh Maynard
Yeah, I agree — I think humans are still the ultimate editor, and to ensure that things are sharing at least the output, or sharing the context, across the organization. But I'll say, I think the trick here — this is why I like Chord — is that structuring your data is so much more important now than ever. I'd argue it's potentially more important than the data itself, which seems like a sin to say — we've talked a lot about how BI as we know it is changing dramatically.
Bryan Mahoney
Right?
Josh Maynard
Like, the interface is no longer Looker — it's now a chatbot, and it should be, because it democratizes data in a really powerful way. And I think why that's powerful is because big decisions, in my experience, are not what changes a business. It's the collection of daily decisions made by people on the ground — optimizing that ad set by 10%, optimizing the email flow by 15%, whatever it is — and the collection of those optimizations leads to really big incremental changes.
Bryan Mahoney
It's like a big decision, but then it's executed incrementally — how are we making sure we're executing on the strategy we set out?
Josh Maynard
Right — so without a good structure of the data, you can't trust the output the LLM is giving you.
Bryan Mahoney
They stand no chance, right.
Josh Maynard
And especially — I mean, I've done this, I'm sure a lot of people have done this — where I'm literally uploading Google Sheets and Excel files of data, and if my dataset is slightly different than somebody else's dataset, we're going to get a different answer. That's a big problem — it's a big problem in every organization. So having a centralized repository, a database that's structured in a way that's optimized for an LLM, I think is incredibly powerful. Organizations that adopt that quickly are going to win, and the ones that don't — that let their account managers or whoever else re-upload Excel files — those are the ones that are going to get left behind.
Bryan Mahoney
Yeah. So we got to this part of the conversation faster than I intended to, but I think this—
Josh Maynard
Sorry.
Bryan Mahoney
No, no, this is great.
Josh Maynard
This is giving a hit.
Bryan Mahoney
This is the kind of meat of it. What I've learned in the last, call it 12-plus months of building here, is — I woke up one day and thought, is dimensional modeling dead? Do we even need to structure our data anymore — can't AI just do it? What I realized, to your point, is that the structure of data matters more than ever, because we have to train the LLMs — if they're going to write, at the end of the day, in a business-intelligence context, we're asking them to write SQL. So — how do I ask you this question, maybe it's a really hard question — the kind of question JP asks, that any senior analyst would take hours to answer.
Bryan Mahoney
But in order to give these LLMs a fighting chance, the data has to be really well structured. That was the first problem we tried to solve. And even once we solved that, with really well-structured data, what we learned is the answers coming back are still missing some nuance of the business. So the thing for me, bringing it back to context, is — how can businesses, if we think about these agents as people you're going to hire into your organization — you hire any net-new person, you need to onboard them, give them access to your data, give them access to tools, but they also have to learn the specific acronyms or shorthand that Sonos, or Mr. Beast, might use.
Bryan Mahoney
But by the way, if you look at this code, there's this little exception here because of this one period of time we did this thing a certain way. What I'm really excited about in developing that context is there's real value there — I think that's where businesses can start setting themselves apart, by training it that way. Here's the thing you need to know about our business, so that if we're asking these questions, I'm not just giving you a generic answer — I'm giving you a Sonos-specific answer, I'm giving you a Mr. Beast-specific answer, because your transactional data guarantee looks similar, but your business context is very different. I think that's where this war is going to be won.
Bryan Mahoney
The other thing too, before I let you react to this context thing — I was in San Francisco this week, and I was lucky enough to be in the audience to hear the Chief Scientist from OpenAI talk. His talk was equal parts terrifying but also really insightful. As I think about how we use AI in the context of business intelligence — he said, you know, the models we're building, we don't expect them to give the same answer every single time, they're non-deterministic. And I thought, wait a sec — if I ask how many orders we placed last week, I want a very deterministic answer, I don't want something arbitrary. So that actually brings me back to context.
Bryan Mahoney
How can we think about context not only as making sure the brand voice is consistent, but as the guardrails I can put in place so I can trust my entire organization — they ask a question, they're going to get an answer I can trust? I think we're very early in the game of talking about context — I think we're going to end up conflating it with data. But for me, this is the part I'm really excited about, and we're just getting started. I've seen some of the context you've contributed, and I'd just be curious — Mr. Beast, to me, sounds like you're sitting on this treasure trove of context you can use to train models. So how do you help your teams think about the difference between data and context? Does that even matter?
Bryan Mahoney
Am I barking up the right tree? Might be a hard question to answer.
Josh Maynard
Yeah, I need to think about it. I mean, I think context is king, right — context is just empathy. I can't speak for the creative side of the business, but on the commerce side, I agree with you — there are certain things you don't need to use an LLM for, for every customer. No — and I wouldn't use an LLM for, say, how many orders did I get — I guess I would use the chatbot, but I would train the chatbot to output a piece of SQL that I approved, to give me orders, for instance.
Bryan Mahoney
But so, can that piece of SQL that you approved also be context?
Josh Maynard
That's for sure context.
Bryan Mahoney
Yeah.
Josh Maynard
Okay — because it just means, this is what I need when I say that.
Bryan Mahoney
Don't go and interpret this differently — you know what the answer is, because I've given you this context.
Josh Maynard
This is what we want — I think you just train the models on that context. But at the end of the day, this is why it doesn't work as just humans — hallucinations are a real thing, and you need somebody double-checking. Yeah, always.
JP Beeghly
Yeah, I mean, it's just — by the fact that the UI is a chat, it's a relationship built on language. You have to find the common language between the data you have and how you speak to it. So part of the modern data-scientist role now is coming up with terminology that's acceptable across the organization. I think that's what we're finding when we test Chord — when I start being very basic in how I phrase things, it has a harder time understanding. It'll come up with an answer, but then it's taking a lot more of its own latitude in terms of context.
JP Beeghly
And so, you know, you guys have done a good job in the background of really helping us find context, because you know our business well after being our partner for the last couple of years. So I think that's important.
Josh Maynard
Do you need to use an LLM for everything, though?
JP Beeghly
Like, I don't — and actually I kind of struggle with that even in Chord. I do like the visibility of SQL — sometimes I struggle with just being able to do segmentation, and I want to verify the segmentation. I want to know that every time I run the SQL, it's doing exactly the same thing. By hiding that, I feel like there's a little loss of control, and a little loss of trust that happens at the same time.
Bryan Mahoney
We do show you the SQL, you—
Josh Maynard
—have to show the SQL. I think Shopify does a good job of that too. Yeah, let's show you the SQL. I don't know — it's like, a car is great, you don't need a car to walk, or — you don't need to use it. Sure, not my kind of car. You don't need to use AI for absolutely everything.
JP Beeghly
No, no, I think there's—
Josh Maynard
You shouldn't.
JP Beeghly
Yeah, there's a laziness to it. I do think that business intelligence, when it burst onto the scene — the Lookers, the Tableaus, the Domos, the Omnis of the world — it was like, we're going to democratize access to the entire organization. They're all independently wonderful products, but they can be difficult to use. One of the things that's been really eye-opening for me is we released Copilot to some design partners, and there's been a number of — we give seats to almost everyone in the organization for a customer that signs with us, and we're seeing all these net-new logins, people asking questions who've never logged in before. To me, that's really great — I love seeing these questions.
Bryan Mahoney
My favorite thing to do in the morning is I wake up and look at the questions people have asked Copilot overnight, and think — how well did we do? But it's neat for me to see new people — I was like, well, why—
Bryan Mahoney
—they had this data before, we didn't just give them new data, what we gave them was an easier way to access it. So I am excited about — how do I say this the right way — I'm excited about the role AI can play in truly democratizing access to the business. But what I'm realizing is that really what we've done is we've allowed companies to hire a new person onto their team.
Bryan Mahoney
We've allowed you to hire an analyst you maybe didn't have the budget for before, to be able to say, I've got this question, or I have this hypothesis, can you go work on that for me — and now you've got this army of agents who are like, I'd be happy to. I didn't mean to steer us into the agent conversation, but I have — we're going to come back there. But I think it's less about AI and more about the fact that there are these new jobs that can be satisfied within an organization, in a friendlier format — it's conversational. But again, it's sort of — do we trust them to do it? So if you'll indulge me, let me use this analogy of hiring new people.
Bryan Mahoney
So if you were to hire a junior person on your team today, you'd think about onboarding them, giving them access to things, training them. Are you doing enough within an organization when you onboard AI right now to do that? First of all — do you think about it as a junior-level employee? Is that fair? And are you trying to train it so you can grow it from junior to mid-level to senior? Or are you happy with it just hanging out at this level?
Josh Maynard
How do you set expectations?
JP Beeghly
I'm not going to get in trouble and answer that one.
Bryan Mahoney
I don't mean to get anybody in trouble, but the—
JP Beeghly
Yeah, go ahead — you go first.
Josh Maynard
I was just gonna say no — probably, like, we don't do enough of that. I am consistently surprised by how much better my answers are getting, and I realize now that if I were to spend more time training, I—
Bryan Mahoney
—could have these good answers. Is it because you're getting better at prompting it? Because prompting is almost like employee training.
Josh Maynard
I could say yes to that. Yeah.
Bryan Mahoney
Okay, just the models—
Josh Maynard
You're getting better. Yeah, I don't know if the model's getting better — it's just the amount of context I've given it.
Bryan Mahoney
Okay — so, learning about it, students—
Josh Maynard
Well, it's also, without prompting anything about Beast Industries, now I get answers that are specific to Beast Industries — I'll ask a general question and I'll get a general answer, and then I'll get another blurb about how that directly applies, which I didn't ask for — which is great, you know, 'cause I should've — I'm not great around prompts. I certainly think I've gotten better, but my point is, I don't think we're doing enough to train it. And I think your analogy is a good one — you would never hire somebody and immediately give them the most complex task—
JP Beeghly
Yeah.
Josh Maynard
—without them first proving they can do the simple tasks well. So I think probably everyone, but certainly me, could do a better job training new AI models as we adopt them.
JP Beeghly
I mean, I could. I'll put a big frame on this — economics change everything, right — it comes and, when you're in a period of growth or a period of decline, what do you change, sort of your lens on that. Let's just take a place where you have to do more with less — right, and this is where AI is really a big part of the language now, like the part of how we speak. It's how do we use AI to help us do more with less. I think the roles hiring is always the same — you find the right person for the role you're hiring for, and you make it very specific about the responsibilities in the role.
JP Beeghly
But a person grows by kind of getting outside of that and really discovering out of that — so we're going to bring that back to AI. If they can come in and sort of be self-starters, use AI to really build themselves into a better person that's fitting the role, then they're going to evolve — and often people don't like growing into new roles within a company, they kind of build their role because in doing that already. I think that's just like Google — you go from school, where you had to go to a library and pick up a book, then you have Google, you can Google the research you need. Now you have AI to do that research and hyper-speed it for you. I think it's just really evolving everyone's potential capabilities.
JP Beeghly
And that's what I'm really excited about — everyone around me who was already the most brilliant person I've ever worked with is now even more brilliant. They're even quicker about bringing ideas to the table.
Bryan Mahoney
So, go ahead.
Josh Maynard
I was just going to say that's a good point — I was just talking to my wife about this, that memorization used to be so important, and then that skill — I think, kind of with my generation, how I got through school — right, and no longer is that as important. My generation was, you know, how quickly could you find the answer to something, which was much more important than memorizing the answer. In fact, it was probably much better for you to not have it all in your memory — I think now it's much more important how quickly can you ideate.
Bryan Mahoney
Yeah, right.
Josh Maynard
And so it's not anyone can find the answer immediately now — but now it's, how well can you ideate and iterate? AI is so powerful, and then you do that, just as Google was so powerful in helping you find the answer. So it's just the skill sets are changing — what's important today is not going to be as important tomorrow. And I think the idea people and the creatives are the stars — I mean, they always have been, but they really are now, because they have such a powerful tool to help them ideate, iterate, and execute.
Bryan Mahoney
So, but the two of you are talking about a very hands-on approach to AI. I didn't mean for us to talk about AI as much as we have, but I think it's interesting to me — I'm drawing this parallel to being a really good manager. You get out of the employees you're bringing on what you put into them, and I think what I'm so far seeing is the more I put into training AI, or maybe it's an attention to prompting, the more I'm getting out. So I'm wondering if that ideation, or that willingness to work with the tools we have, is what's going to separate the next generation of commerce operators.
Bryan Mahoney
And I hope that's where we are, versus how can I automate the process of what it means to be a brand faster than someone else — because to me, that's where we lose the soul, versus I get out what I put in. Do you have a point of view on that? Like, are you trying to automate everything so you can do more with less, or is it really more about how do we be even better at what we do, so we can do more, faster?
JP Beeghly
I mean, honestly, to your point — we're a brand, and brands are human by nature. Brand is a human definition you can't really put your finger on — you just know it's something more than the product itself. There's a broader understanding and emotion that goes on top of the product you're selling, and that's a human conversation that needs to happen. So, no, I actually don't think about it that way at all. We really take a customer-first lens at it — if we're going to talk about a customer experience, the way we approach it is, what is the customer experience, who is that customer at that given moment, what do we know about them at that moment, and how can we develop a really good customer experience that helps them understand the brand?
JP Beeghly
And it's very hard at the prospect stage, or the lead stage, where you don't know anything about that customer — you might know where they came from, or what attracted them, but really you know nothing about them at that point. There is a point where I'd say I could envision AI being a tool where I could iterate very quickly and try to learn something about that stage of the customer that might not reveal who they are, but obviously offer them something that's more attractive at that moment. And that's the thing — when you're thinking about a brand, people come to you because they're excited about having something more, you know, like more than a speaker. Sonos is a system — it is beautiful.
JP Beeghly
It has the best sound, but it's also got an iconic history that's all built around that passion — and people come there and they want that. So if they're coming in that moment, that's what we're selling them, and we've got to make sure we deliver on that. I won't trust AI to do that in all those moments — so I think the challenge is figuring out where it can assist, but still making sure you don't deviate from the brand, which is a very human conversation.
Josh Maynard
I think it's well said. Back to what I said before — I think it's augmenting, not replacing.
Bryan Mahoney
Yeah.
Josh Maynard
Certainly there are tasks that are no longer necessary, but that's normal, that's progress — and yeah, I don't think there's anything to shy away from. I also think it's great that AI has kind of allowed three technology guys to be here talking about brand — it's like, something that's true, the engineer-me ten years ago can't believe that I now have tools that have enabled me to be able to do this. So, yeah, I think it's all about augmentation — we're not trying to automate everything or automate what we can, we're trying to make ourselves better. And I think, back to this ideation and iteration — that's where the true power is, I think, on our side, with AI specifically, and that's—
Bryan Mahoney
What we're doing — I think more brands need to hear that. The more I talk to people, whether it's at conferences where it's Chatham House rules, or you talk to people, there's this idea that everyone else has automated it, everyone else has deployed AI within their organization. But I think the truth is we're still trying to figure out how we can use it to augment.
Bryan Mahoney
And I have two ideas that come to mind, and maybe we'll try to wrap up here — the idea of, for anyone who's earlier in their career listening to us this afternoon, wondering how they augment and how they use these tools. If you're a year or two into your experience, what advice would you have for someone to use some of these tools, but understand how to do quality control, how to really think about getting 10 or 15 years of incremental experience, hands-on on the job? Because I really worry about that layer now — how do you go from, 'I can look at something and know it's not going to work in production,' when I've written code for 30 years—
Bryan Mahoney
—like, I know this isn't going to work in production, but if I've never shipped production code, how can I use some of these tools that are incredibly powerful to give myself a shot to augment? And I have a tough follow-up to that, but let's start there. I don't have an easy answer for this — I hope you guys have something smart to say.
Josh Maynard
Never — I think that's true, actually. It's interesting, because I guess back to me being kind of an optimist about all this — it doesn't feel different. I used to have junior-level engineers whose entire job was to build landing pages — writing static HTML and CSS with some JavaScript. That job doesn't exist anymore, and that's not because of AI, that's because the tools have gotten better. There are, I think, lots of people learning how to code right now who are not learning JavaScript — they're specifically learning React.
Bryan Mahoney
Yeah, right.
Josh Maynard
They're learning the framework written on top of the language, rather than the language itself — which, I mean, I don't think that's a great idea. But my point is, I think all of this is progress. Jobs and roles change as tools get better, so roles allow you to do more. I think it's just all about how you can use the tools you have to ideate and output as much as possible. And the advice I'd give somebody — I'm rambling a little — but the advice I'd give somebody early in their career is to mine for conflict, which I'd tell anyone to do. I think what that means for them is to ask questions constantly. Yeah, be serious — yeah, this is what I outputted—
Josh Maynard
—and then go to someone who's more senior and get their opinion on it, or whatever it is you're doing. I mean, that's almost part of the playbook when you're an engineer.
Bryan Mahoney
Right.
Josh Maynard
You know, PRs are a thing, reviews are a thing. But I think for any role — take the go-to-market strategy that you had ChatGPT write for you—
Bryan Mahoney
Yeah.
Josh Maynard
—and go to somebody more senior and say, what is this, how does this work, what does this mean? I actually don't even know.
Bryan Mahoney
That's amazing advice.
Josh Maynard
And you can even ask ChatGPT to—
Bryan Mahoney
—do an initial code review through the lens of a senior engineer. If I'm worried, if I have anxiety about bothering someone or taking up their time, or I don't know if the code is good — if I'm curious and I care, and it's not just about getting to a fast answer, I can do that and say, how would I make this better, how would a senior engineer think about this, how is this going to hold up? As opposed to just thinking, I need to ship something out better and faster and quicker. I need to be curious, and I need to want to grow. So yeah, I think there's hope — those are the people who are going to be the stars.
Josh Maynard
I think so.
Bryan Mahoney
Yeah.
Josh Maynard
I just think that this is a tool that's going to make us—
Bryan Mahoney
Yeah.
Josh Maynard
—continue progress.
Bryan Mahoney
I can't believe how easy it's been to come back — I was like two years removed from writing actual production code, and I've tried some of these tools, and I was like, wow, I cannot believe it.
Josh Maynard
They let you write production code? Yes — still.
Bryan Mahoney
Yes, I don't remember — I have still used the production—
Josh Maynard
I'm impressed. I don't think I—
JP Beeghly
Well, like when you're writing the CEO—
Josh Maynard
—the business, like, yeah, right.
Bryan Mahoney
Yeah.
Josh Maynard
That's good. Yeah, I don't think I could write production-ready code anymore.
JP Beeghly
I would — no surprise, I agree with you, sir. I mean, the greatest person you're going to hire or add to your team is a curious self-starter, and the tools available today just allow someone who has the propensity to do great things, to do great things much quicker.
Josh Maynard
Yeah.
JP Beeghly
But I agree with you that you have to be able to present it, and the biggest skill — the hardest thing to break through to be successful — is having that tough conversation all the time with your managers, making yourself visible, being willing to contest things, being willing to have the guts to present something when you know you're out of your lane. Those are the things that develop growth, not only in your career, but as a human, I think. Yeah, yeah.
Bryan Mahoney
All right. Speaking of developing humans — you're both fathers to daughters. JP, you have two daughters, both — yes, you're welcome — late teenagers, in fact. So, is that when you're still a teenager? Well, okay, yes — no longer a teenager, young adult and teenager. So entering the workforce, where AI is going to be kind of top of mind. And Josh, you have an 11-month-old who's still quite far away from joining the workforce, so by the time Madeline gets there, AI will not not be a thing — it'll just be sort of common language. But JP, for your kids right now, it's really something. So forget the next generation — think about your generation. What advice would you have for them with regard to AI? I think about this—
Bryan Mahoney
I have a daughter as well, and I remember people asking me, what language do you want to teach her next? And I was like, not a spoken language — I want to teach her JavaScript. Now I wonder if that's the right advice for her. So — do you have, does what you're seeing and what you're working on today give you advice you'd have for your daughters? And is it the same advice for your 20-year-old and your 14-year-old?
JP Beeghly
Not necessarily, and that's because they're different personalities — they sort of attack the world with different strategies. My older one is in college right now, and AI is very applicable, or less so, depending on what she's doing. She wants to be a veterinarian, so she's in chemistry, biology, all of it — these impossible subjects, tons of learning. And I said, you don't use AI to help you learn — that's not going to work that way. What you do is ask good questions, like, how can I memorize this in a strategic way that will help me do better on the test — phrasing things the way you would with other colleagues.
JP Beeghly
Or — she has the ability to have a veterinarian she can talk to every day, and say, how do I do well on this test? If you have that, that's the person.
Bryan Mahoney
Now you have a veterinarian you talk to every day.
JP Beeghly
How do I ace this test? Well, you can learn how to use AI to make these incremental improvements in your life every day. That's the advice I give her — don't think about it as cheating, think about it as helping, and use it to help you not cheat or try to get a better grade. The goal is to learn, and if you help it — if you ask it, how do I learn better — you're going to get some really great results. I don't think I talk to my kid about AI much. It's an interesting subject for my 14-year-old, because she just got a phone — I've been the strict dad who hadn't given my kids access to a phone, but yeah, they were later, and I just wanted them to socialize more.
JP Beeghly
I think social skills is one of the things we're not talking about here that's really important for career advancement. Some of us who were pre-phones obviously dealt with people one-on-one and developed a different skill set around how we communicate. I think for my 14-year-old, she's on AI all the time because she's looking at TikTok and social media, and she's sort of grokking what society is and how to be, from these 15-minute videos. So I've spent time really talking to her about what AI is doing in the Gen-AI space, and how it's developing this kind of content really quickly for her — so that she gets an understanding that there's a disreality in the reality she's viewing.
JP Beeghly
And to look up — put the phone down, look up, and look out at what's happening, and try to marry those two together to develop something. That's a big conversation for a 14-year-old — of course I don't talk to her like that, but that's my strategy overall.
Bryan Mahoney
I love that.
Josh Maynard
Yeah, I love that too. It's hard to imagine this — she had to stay a baby forever, it's crazy, she's already almost a year old, which is insane to me. I think what I'd encourage Madeline — that's her name — to do is be discerning, question everything, and form her own opinions. I think the dark side of AI right now is just how lifelike generative AI is becoming, and it's hard to discern what's real and what's not. Which is why I think it's so important to form your own opinions, and come up with your own solutions after looking at all the facts. And I'd encourage her to have empathy, try to see things from other people's perspectives, and try to walk in other people's shoes as much as possible.
Josh Maynard
I don't know exactly what I'd tell her about AI — to your point, I don't think it'll even be called AI anymore, it'll just be on the power of lagging. So, yeah, I think I hope she questions everything her mom and I say, and that it forces us to explain things honestly. Yeah — I know, be careful what I wish for.
Bryan Mahoney
Right.
Josh Maynard
But I am — I'm really excited for her, to see her personality forming, to see her becoming a little person with her own opinions. Anyway, I don't know if I answered that question directly about AI, but I think it's truly just about being critical about things.
Bryan Mahoney
And you both gave me the perfect way to wrap us up — and again, we did this in a single take. But as I think about it — first of all, the conversation is so different than last time, which was a little bit more technical. Now we're talking about the landscape changing dramatically, whether it's the brands we work at, or raising our daughters — AI is the center of the conversation right now. What I'm hearing from the two of you is that it can be incredibly powerful, but it's not a shortcut. It's not a shortcut to get to a shitty answer that you dive back in and fix. It's about being really curious, being discerning, and using it to learn, not to cheat.
Bryan Mahoney
And I think that's really important, whether you're doing it to get better at school, or to ship a better product, or a better go-to-market strategy, or whatever it's going to be — it's just a tool. But at the end of the day, it requires us to be critical thinkers, to be discerning, to be curious, and to always want to be learning. We're just learning something new, and ultimately that's really the goal of these conversations — I still don't think there's anything magic going on behind the scenes at the best brands. The magic is really the people, being curious, wanting to connect with their customers, and being willing to do the hard work.
Bryan Mahoney
And so when I hear people saying, well, we're going to automate what it's like to be a brand, and a brand is going to be a single-person operation — I'm like, that's not a brand. I want to be friends with a brand — is that a brand I want my daughter to be friends with? I want more critical thinking, more curiosity, use tools the right way. This is an incredibly powerful tool, and I think we're just getting started. That's the other thing I hope this conversation has revealed for people — go experiment, be a critical thinker, but don't think that it's a solved problem. I've been railing against this idea that commerce is a solved problem for the last decade. So this has been another amazing conversation with you guys — we did it again, in a single take.
Bryan Mahoney
So we're setting the bar high for round three.
Josh Maynard
Wrap us up here.
Bryan Mahoney
Thanks for coming up to Santa Barbara — great conversation, and looking forward to the next one.
Josh Maynard
It's great — I mean, just being surrounded by such smart people, I really appreciate being here.
Bryan Mahoney
Thank you.
Josh Maynard
Smart, but curious together.
Bryan Mahoney
Good night, figure it out.
Josh Maynard
Yeah — any good examples of being a great — I hope so. Thanks, guys.
Bryan Mahoney
Thank you. That was fun.