Latest / Inside Commerce: Ecommerce Strategy, CX and Technology Podcast / EP294: How AI Code Generation Will Evolve The Relationship Between Brands & Agencies, with Jim Tattersall & Seth Bindernagel
Transcript
- 0:00We talked about value and empowerment and ultimately as an
- 0:02agency, you have to provide value and you should be looking
- 0:05at how you can empower so that your collective velocity as fast
- 0:08as possible. AI is amazing.
- 0:10I mean, it's incredible. It's an amazing opportunity.
- 0:12It's changing weekly and it's so exciting in terms of how we can
- 0:14increase the value that we've delivered to customers.
- 0:21You listen to Inside Commerce, the weekly ecom strategy,
- 0:23technology and CX focus podcast hosted by experienced digital
- 0:26consultants and re platform specialist James go to Paul
- 0:28Rogers, get insights into ecom strategy, leading brands, hear
- 0:31from industry thought leaders and leading tech vendors with
- 0:34practical advice to help you improve your online CX.
- 0:39This episode is sponsored by Shopify.
- 0:41We've got a really interesting debate topic rant going on
- 0:45today, which is about we're looking at the impact of AI,
- 0:48specifically Gen. I on how ecommerce teams are
- 0:50going to evolve their relationship with technology
- 0:53agencies in the future and the concert.
- 0:56The I guess the key context to this is, is that AI in the hands
- 0:59of an expert versus AI in the hands of a tech novice and how
- 1:03those are very, very will drive very, very little outcome.
- 1:06So before I do a little bit of preamble aware with some of the
- 1:08discussion boys, let's introduce our our amazing guest today.
- 1:11So we have Seth from who's the VP of marketing at Commerce
- 1:15layer. Hey, Seth, welcome back to the
- 1:17podcast. Good to be here.
- 1:18Thanks. And we have Jim, who's founder
- 1:22of CT at Rotate. You've also been in the podcast
- 1:24for Welcome Back, Thank. You very much good to be here,
- 1:27gents. Look, gents, thanks for doing
- 1:29that. And Seth, I really like the fact
- 1:30that you reached out and suggested this because for those
- 1:33of you listening, I'm if if anyone isn't on LinkedIn, which
- 1:35I'll be a standard by you should get on then just see how much
- 1:38noise. There is so much out there.
- 1:40There's a lot of constructive inflation.
- 1:42There's also a lot of sensationalist.
- 1:44You must do this away or you must reposition your whole
- 1:46business model. If you don't do this, your
- 1:48business will fail. There's a a lot of hype crap out
- 1:51there, which is really irritating.
- 1:52And what we want to do is get beyond the hype and actually
- 1:54talk about the reality here. Look at how AI tools are
- 1:58fundamentally helping E Comm teams increase automation, how
- 2:01they're helping them do things more themselves and find
- 2:04efficiency gains. And wherefore what that's going
- 2:06to do in terms of what people can do themselves with good tech
- 2:10and governance versus how they can speed up processes but still
- 2:14need proper technical oversight. And, you know, technology
- 2:17experts to own the delivery, the QA, all of that piece around
- 2:22code that will get released into a production environment.
- 2:25And there's so much going on. We got leading tech vendors at
- 2:27shop. I just dropped additions with,
- 2:29you know, the AI code generator for content layouts.
- 2:34You've got AI specialist vendors in different areas around
- 2:37product discovery or chat bots continuously recent new
- 2:40capabilities. You've got agencies and Jim, you
- 2:44can share some of the things Jordan who are integrating AI
- 2:47into not only their coding practices, but also building out
- 2:50tooling for customers to help them with Automate.
- 2:52There's so much going on. So what we want to tease our is,
- 2:57is what what is this? What's it going to change?
- 3:00And what are the gotchas that you really need to be thinking
- 3:03about before drinking the kool-aid and assuming that AI is
- 3:06going to solve all the problems of the world.
- 3:09Seth, Jim does that, does that, is that a good enough set up or
- 3:12is there anything you want to add in terms of what we're
- 3:13trying to achieve in this discussion ramble?
- 3:16I think that's a very starting point.
- 3:18Yeah, let's get going. Let's go do this.
- 3:21Excellent. We none of us on this on this
- 3:23podcast know where this is going to end up.
- 3:25This is this is an exciting podcast for me to do.
- 3:29And I think this is where the industry is generally is trying
- 3:32to work out what the direction of travel is.
- 3:34So we're going to start. So the context and why this
- 3:36discussion came about is Seth recently challenged himself to
- 3:39build a website without talking to developer using a comma slate
- 3:42API and AI tools. So why don't we start then?
- 3:46Why did you do that, Seth, and why did you think that was an
- 3:49important thing to go through? Yeah, Thanks, James.
- 3:53The, the initial thought exercise was not just to build
- 3:56like a simple website. I wanted to build a, what I
- 4:00would consider a robust e-commerce application.
- 4:03And I've been leading the marketing for commerce Layer for
- 4:06like 4 years now. And I talked to so many people
- 4:08about, oh, use the commerce Layer API and this idea of
- 4:11composable commerce and it's, it's easier to build and you can
- 4:14swap things in and out and all the talking points that's around
- 4:17our industry. And now I have AI coding tools
- 4:22and I thought to myself, like I wonder if I could build a
- 4:25reasonable website prototype of a store that would be an online
- 4:30e-commerce store that would use all of the components of a
- 4:33composable commerce set up. So that I could learn very
- 4:37deeply how the commerce layer API gets integrated from the
- 4:41back end into a web application. And so I came up with this
- 4:44concept. It was, you can check it out.
- 4:47It's at the URL Seth dot house, Seth dot Haus.
- 4:52And it's, it's an e-commerce website for a triathlon store.
- 4:55It's called Seth's triathlon store.
- 4:57And I built it over probably about a two week span from
- 5:03everything from using Commerce layer's command line interface
- 5:06to like build out the entire e-commerce back end and then
- 5:09using a series of AI coding tools to write all the front end
- 5:13code. And, and I ended up with a React
- 5:16application that I hosted on Netlify.
- 5:21And so there's a lot of things to talk about there, but I
- 5:23challenged myself to do it because I felt like, look, as
- 5:25AVP of marketing for a very technical company.
- 5:29And any marketer right now needs to like really use the tools
- 5:33that are at our fingertips, particularly in AI, to learn
- 5:37deeply about not only the tools that they're marketing or
- 5:40they're charged with marketing, like the commerce layer API, but
- 5:43also to learn like how much more a marketer can do in this day
- 5:47and age. And a starting point for me was
- 5:49to like start with a core, like get a web application built.
- 5:53And then I started to experiment with a bunch of other stuff
- 5:55there, like putting in analytics and putting in a data layer to
- 5:58track things and all these other things that a marketer would
- 6:00care about. But I needed that foundational
- 6:02application to start. So yeah, so I, so I did it.
- 6:05And the challenge, the only, the only requirement I put on myself
- 6:08is don't ever talk to a human being developer to get any help
- 6:11debugging it. Any, any other resource I can
- 6:14use, I can go for, but it's just, I'm not allowed to ask
- 6:16anybody for help. And yeah, so I got there and the
- 6:20website's live. I hosted the code on GitHub and
- 6:22put it out there. I even put it on the MIT license
- 6:25in case people want to like, fork the code and build their
- 6:27own web app. I don't recommend it, but it's
- 6:30out there. So yeah, it was fun.
- 6:32It was a ton of fun and I think I learned a lot.
- 6:34Let's go. So before we come to you, Jim on
- 6:36this, because This is why we got different viewpoints on this
- 6:39podcast because there are with great, great like tools comes
- 6:44great responsibility. But Seth, it sounds like you
- 6:48type here, right? And and there's enough bias out
- 6:51there and some of the noise about actually this.
- 6:53Yeah, you can just do it. Just create perfect production
- 6:55ready code. Sounds easy, sounds perfect.
- 6:58What's the reality? What do you ask about?
- 7:00What issues did you encounter? It is not easy and you know, I
- 7:07was trying to build a complex web application and Jim can
- 7:10testify to this. When you're really trying to
- 7:11build a multi market like complex web app that can serve
- 7:16multiple products and with multiple currencies and and
- 7:19build a check out like all the things that I did you, you need
- 7:23to have some knowledge. So I would say that I don't, I
- 7:25don't have an expert knowledge of the commerce layer API
- 7:27because I'm not an actual developer, but I have AI have a
- 7:29pretty good working knowledge of how the API works.
- 7:32If I didn't have that, it would have been very, very difficult
- 7:36for me to do what I did. So I had a good starting point
- 7:38on like how to at least think about how an API like commerce
- 7:41layer could deliver e-commerce data to a front end.
- 7:43What I had no knowledge of was how to build a React
- 7:46application. And that's where I got into
- 7:47trouble because like I, I got to a point, I, I went through so
- 7:51many different coding tools. I got to a point where I had one
- 7:55tool, Windsurf, which many people have heard of as my IDE,
- 8:00which had a coding companion built in.
- 8:02And I would take the recommendations that Windsor was
- 8:05saying. I copy paste them into either
- 8:07Gemini 2.5 or Cloud 4.0. I'm migrated to Cloud 4.0.
- 8:12When that launched, just a gut check.
- 8:15And at several points Gemini and cloud were like, yeah, that AI
- 8:18thing is hallucinating right now.
- 8:19Hit stop, Stop it from continuing to change code
- 8:23because it's not doing the right thing.
- 8:24And like, I never would have known that.
- 8:27And so I had to get these tools working together because there's
- 8:31no way that I could have looked at what Windsurf was doing.
- 8:33And then like, this is bad. And it was going into these
- 8:36loops and it would be like kept changing code that like, I
- 8:40didn't really know what it was changing.
- 8:42So anyway, that's the great responsibility that, like, comes
- 8:45along with these tools is like, you know, Jim would look at it
- 8:47and go, Seth, what what are you doing?
- 8:49Why are you even hitting go? But I was just like, yeah, yeah,
- 8:52yeah. Go, go, go.
- 8:53I'm just trying to get this thing live.
- 8:55Bad mistake. So I I learned a lot throughout
- 8:57that process. What do Mattel, Victoria,
- 9:02Beckham, Sennheiser, and Glossier all have in common?
- 9:06They power their businesses with Shopify.
- 9:10Shopify is the most innovative and scaled commerce platform on
- 9:14the planet. It also happens to have the best
- 9:17converting check out on the planet Shopify, Powering the
- 9:22future of Enterprise commerce. Find out more at
- 9:26shopify.com/enterprise. Yeah, that's a bit like me with
- 9:29my WordPress site for my for digitaljuggler.com where I go,
- 9:33go, yeah, I'll upgrade the PHP and I'll I'll edit that.
- 9:37What could go wrong? What could everyone.
- 9:40It's offline. OK, Oh, Jim, let's come to you.
- 9:44You know, you are founder of CTO, you have immense amounts of
- 9:47technical experience. You work across different
- 9:49platforms, different projects, Composable, non composable,
- 9:51headless, non headless. Right.
- 9:52So you've seen the good, the bad and the ugly.
- 9:55I I'm going to do to give you a A1 line statement, right?
- 9:59Econ teams don't need good technical leasing developers
- 10:02anymore. They can just code themselves.
- 10:03Discuss. Yeah, I think, I think the key
- 10:09thing that kind of came up was comes as mentioned, it was like,
- 10:12you know, what are the risks? You know, what are the risks
- 10:14here, right. So Seth didn't use a developer
- 10:17and what, you know, what are the risks involved with that?
- 10:19And it's like, well, the risks are exactly the same as when you
- 10:21use a developer, right. So you're a developer, you've
- 10:25got to worry about security, privacy, brand integrity, like
- 10:29exposure of sensitive data, all these things.
- 10:32Now, development isn't just like the coding.
- 10:35Like I love Marty Kagan, you know, author of Inspired and
- 10:39Silicon Valley Products Group, isn't it?
- 10:42You know, he was like, product engineers are defined by how
- 10:45they code, but like the creative problems that they solve.
- 10:48And I think that's it, right? You know, actually you're an
- 10:50engineer and coding is part of it.
- 10:51It's probably the least valuable part of it, actually, when
- 10:54you're developing software from end to end, you, you're right,
- 10:57right. The risks are actually massive.
- 10:59And so you have to kind of have automated testing, you know, any
- 11:02code that we write, even if it's AI, you know, a test is written
- 11:06with AI, then that's reviewed by a human, then that's reviewed by
- 11:08another human, you know, and throughout those steps, you're
- 11:12kind of going through all this process.
- 11:13Now what's the risk? It's like, well, it depends on
- 11:15your, the level of exposure. You almost need a developer to
- 11:17even assess the risk, right? Because for Seth's application,
- 11:21the risk is actually huge. I mean, you could have card data
- 11:24exposure. You could have like, you know,
- 11:26privacy exposure you can have, I mean, there are issues in the
- 11:29code base like I did have a quick look through.
- 11:32But yeah, you can join the team, still find Seth, you can still
- 11:35join as an engineer. But but I think what's there are
- 11:38so many volunteers, right? And I suppose we'll get into
- 11:40that. But from a risk perspective, it
- 11:43can be pretty huge. And I think that's a concern is
- 11:45that actually AI can be leveraged in a multitude of
- 11:48ways. And actually when people talk
- 11:50about Cogen, I think what we forget is actually what's the
- 11:54objective, what we're trying to achieve.
- 11:56Seth done this as an experiment is great.
- 11:58It's great as a learning kind of like initiative and like to
- 12:01understand the technologies and all this kind of stuff.
- 12:02But actually, if you went out and you said, I want to create a
- 12:06production grade website that looks great and functions well,
- 12:10you wouldn't approach it like this like there's no way, right?
- 12:12You would probably just go on to Shopify, say, you know, use, you
- 12:16know, I mean, even like the horizon thing now, right?
- 12:18You'd kind of go, OK, this is the style they want.
- 12:20You'd kind of generate, you know, the layouts and stuff and
- 12:23it's like, cool. And that would be pretty safe.
- 12:25You know, it would be safe. And I suppose we can kind of go
- 12:28into a bit more depth around very different approaches, just
- 12:32kind of solving the same problem, right?
- 12:33You kind of go, this is the end output I want.
- 12:36But actually there's so many different ways I can go into it.
- 12:38And I think in terms of code Gen., there are loads of there's
- 12:42so much potential. And so like for people who are
- 12:43non developers, but I think you know, understanding the risks,
- 12:48understanding what you're exposing yourself to is
- 12:51obviously super, super critical because it it can be pretty,
- 12:54pretty dangerous if the guardrails aren't in.
- 12:56Yeah. And then there was a thought
- 12:59that came to me this week because I was playing around
- 13:01with Shopify's new AI block generator.
- 13:04I was using it on my development store and I was creating FAQ
- 13:08blocks, carousels and going through the, the learning
- 13:12process of, of, of actually what you're right, what are you
- 13:15trying to achieve? Because anyone can create a, a
- 13:18content that's very easy. So the question is what, what
- 13:21are you trying to achieve from a, a functional point of view,
- 13:23from a customer interaction point of view, from a design
- 13:25execution point of view? Then you have all of the, the
- 13:28safeguards behind the line. Is it touching any data that has
- 13:31a privacy implication? Is it stuff that has to be
- 13:33properties? You have all those questions
- 13:36and, and actually what it's made me realise is that that
- 13:39potentially it's going to force people to be more rigorous in
- 13:46thinking about the specification behind it because I see too much
- 13:49content generator, whereas we've got a layout and then it and
- 13:52then somebody hasn't thought about functioning exactly how
- 13:54it's going to work. If you go through from an
- 13:56engineer's mindset, you're querying, OK, what does it not
- 14:00just what does it need to look like, but how does it, what are
- 14:02the elements? What are the interaction between
- 14:05the elements? Is there a hierarchy?
- 14:07Is there configurable elements? And what shows by default?
- 14:10And you have all of these sub questions that you have to
- 14:13answer in order to get a detailed technical spec to
- 14:15create relevant code that works across devices.
- 14:18And I was going through thinking actually, the intelligence in
- 14:21prompt design to these generators, the need for that is
- 14:25huge. If you want to get production
- 14:27ready for a, you know, if you're a serious brand and you're not
- 14:30just, you know, you've launched your own store and it's a test
- 14:32and just want to get running and selling.
- 14:34But if you're a proper brand where you need your brand
- 14:36integrity protected, you cannot just go into this without
- 14:41understanding the impact between the input and output.
- 14:45Yeah, that's a really good point, James, because like, you
- 14:48know, what's like part of the magic of these AI coding tools
- 14:51is like you can just come in and get going, you know, the vibe
- 14:54coding feeling, right? And it gives a lot of people
- 14:58this feeling of empowerment that they've never had before in
- 15:00their career. But it immediately short
- 15:04circuits the requirements that are typically brought forth in
- 15:08like a Sprint process or like a product manager or in my case,
- 15:12like a very technical marketer sits down with like the
- 15:14engineering manager and says like, you know, did you fill out
- 15:18the Jira story and tell us everything about it?
- 15:20What are your user stories? What are the requirements?
- 15:22And you know, they never say like, hey, Seth, like what are
- 15:25the security requirements for this particular thing you need
- 15:28to do? It's never that because like
- 15:29they come into it knowing they already know fundamentally what
- 15:34needs to be built in to make sure that everything meets like,
- 15:38you know, best in class coding practices.
- 15:41A lot of these tools and I try, I tried like four or five
- 15:44different ones and you name it, like, you know, the lovable and
- 15:48Windsurf and Gemini, all of it. I tried so many of them.
- 15:52They give you this open canvas and they try to prompt you like,
- 15:54hey, send us a brief, just like a brief you'd send anyone else.
- 15:57I want what you want to do. But yeah, you spend some time
- 16:00writing that brief. But like, I got into so much
- 16:03trouble. And Jim, I'm sure you saw it in
- 16:05like the GitHub repo, but they're at one point and
- 16:08unbeknownst to me, Windsurf had exposed all of my clients secret
- 16:12new client ID keys required for the commerce layer API in the
- 16:17public code. And then I somehow got Gemini to
- 16:20look at the code and they're like, dude, you have exposed all
- 16:23of this security stuff and we had to refactor the whole code
- 16:25base to make it secure. That never even thought it would
- 16:29never even came into my mind when I was like started just
- 16:31like this wild prompting of like what I wanted to do.
- 16:35I was kind of the cat behind the keyboard, just like moving fast,
- 16:38you know, I mean. That's that's totally the
- 16:39example, you know, when it's saying you expose something like
- 16:41that and it's like, cool, I'm potentially exposing admin API
- 16:45access to my entire account, right.
- 16:47Like how incredibly dangerous is that?
- 16:49And you know, someone who's, you know, not as knowledgeable as
- 16:52you set as well, might kind of go, cool, I'll just change that.
- 16:55I'll make a new commit. It's like, well, it's in history
- 16:57now. It's in the history of like a
- 16:58publicly available repo, right. So in a matter of you change it,
- 17:02you actually have to completely kill, yeah, you have to get rid
- 17:05of this because it's there, right.
- 17:07And then there's all other things like was that crueled?
- 17:10You know, was that repo when it was like when it was public
- 17:12crueled, you know, and is that now somewhere else?
- 17:15There is so many considerations within that.
- 17:17And and like you say, I suppose people just see the coding
- 17:20aspect and actually kind of go, well, no, this is this is an
- 17:23entire process, right? Like a product mindset where
- 17:26you're kind of going actually the first question is like,
- 17:27what's the objectives of the business, what the goals we're
- 17:29trying to achieve, all these kind of things.
- 17:31Yeah, actually code is literally just the, you know, code is the
- 17:36US kind of being the conduit, right, the translator of those
- 17:38ideas once we have them, and kind of interfacing the machines
- 17:41to kind of make those a reality. And I think the amazing thing
- 17:44with that AI, the most amazing thing out of all of it, is kind
- 17:47of moving from this interaction with computers that we've had
- 17:50historically, right? You know, punch cards and
- 17:52keyboards, mice, touch screens. So something where we can
- 17:55express ourselves in a way more humanistic, kind of like a way
- 17:58where the machine understands intent on a more humanistic
- 18:02level, right? Where we can say actually, I
- 18:04want this, I want this, and the machine can just understand
- 18:06that. And you're interfacing in just a
- 18:08far more, you know, human way, which I think is an absolute joy
- 18:12and actually is one of the huge policy positives of AI in terms
- 18:15of how that's going to affect all kind of user experience,
- 18:18right? You know how we interact with
- 18:20machines and how we make requests and all this kind of
- 18:22stuff with. Yeah, yeah, I've got.
- 18:25So, Jim, a good question from the, the, the agency lens here,
- 18:28because one of the great challenges between clients and
- 18:31agency teams is on optimising use of the retainer to, to ship
- 18:36row map items to do things proactively rather than just on
- 18:40maintenance. And when you're having to field
- 18:44code for new capabilities and features, there's, there's the
- 18:47overhead of doing that and there's the, the QA.
- 18:50Then there's all of the tests that then, you know, if there's
- 18:54a tweak made to it, because in the staging or the test, same
- 18:57environment didn't quite hit the mark, then you have to go back
- 19:00and redo it because see, any change code can impact other
- 19:03parts. What people from a client's side
- 19:07want to get to is where the agency's speed of coding is
- 19:13faster so that that there's cost taken out of it.
- 19:17And that cost is either redisputed into more ideation,
- 19:20like you said, or into being able to ship things faster.
- 19:23I'd love to know where, like what, how, how are you embracing
- 19:27this at the moment as an agency in terms of using AI to help you
- 19:29get efficiency gains and, and do you see this being one of the
- 19:32main areas where you can bring benefits to merchants in the
- 19:35future or is it an unrealistic expectation?
- 19:38No, it's not unrealistic at all. And I think it's the great thing
- 19:42about discussion is we don't know where ultimately where
- 19:45things are going. I mean, no one knew when the
- 19:48smartphone came out the impact it would have on dating.
- 19:51I mean, teenagers, mental health, all sorts of like
- 19:54massive impact that these things kind of have with AI.
- 19:57Every expert, you know, kind of, you know, said something
- 19:59different about where things are going in terms of the near term.
- 20:02There are lots of things we're doing and yes, the sort of
- 20:04things you talk about in terms of high quality.
- 20:07Lower costs are already kind of starting to come in.
- 20:09So in terms of things that we're doing at the moment, I mean,
- 20:12Seth mentioned like Windsurf, but like all of our engineers
- 20:15are using cursor, which have, you know, obviously AI agents
- 20:19kind of like built in. Pretty much all of our tests are
- 20:22all written like purely by AI and debugging is becoming better
- 20:27and rapid prototyping might becomes better.
- 20:31I mean, we're even like starting to experiment with things like
- 20:34in terms of our flow. So you know, you have like a
- 20:38ticking system where you move things along and actually,
- 20:40rather than assigning it to an engineer, assigning it to a, to
- 20:44an AI agent, right? So you kind of go, actually,
- 20:46here's an idea. Cool.
- 20:49Actually write the PRD and you assign that to the PIPRD agent,
- 20:52right? And that writes up the spec, you
- 20:54know, and it writes it up in a standardized format.
- 20:57And so I suppose across the agency, you kind of go actually
- 21:00it's not just about the Cogen aspects of it.
- 21:03And actually, that's the part where it's good at the moment,
- 21:07as in, well, it's good, it's OK the moment if you're very narrow
- 21:10and you're very focused with something like Seth where you're
- 21:12talking about a massive code base, it falls down pretty,
- 21:15pretty quickly. But when you're kind of saying,
- 21:16actually, I want this function, it's doing something very
- 21:18specific and you've got autocompletes and all these
- 21:20things, it does speed up the engineer's time.
- 21:23And most importantly, a cost prohibitive thing, especially
- 21:27with agencies where costs are like, we have to keep costs low,
- 21:29is you always want to pay a program, you always want to work
- 21:32with another engineer, and you want to kind of like bounce
- 21:35ideas off them. Now that's quite hard because
- 21:38obviously you've got the extra resource of just the luxury of
- 21:40having two engineers working on the same problem.
- 21:42The idea is it does speed up velocity, but not not twice as
- 21:45fast. You know, in reality, the idea
- 21:48that the tool becomes more of a collaborator is definitely the
- 21:51case. Like rubber ducking is what they
- 21:53call it, or you know, pair programming, but you're doing
- 21:55that with the machine. And that's, that's really great.
- 21:57So for all the code bases now we're adding extra documents
- 22:01within the code bases. We're adding rules and stuff as
- 22:03well. So we're adding way more context
- 22:06for the AI so it can be more informed and help us move kind
- 22:09of quicker. We're not far off getting to a
- 22:11place where, to be honest, a client issues like an issue that
- 22:16comes in as a ticket, it's automatically signed, assigned
- 22:18to a debugger that kind of as an agent that reviews the code,
- 22:21spots issues, does APR, and then and then the human comes along
- 22:25and kind of goes, OK, cool. And so we're already seeing like
- 22:29some like early days, but we're seeing like efficiency gains
- 22:33already. But more than that, like we've
- 22:35got way better code coverage, the quality, the outputs better.
- 22:39So, yeah. And I think obviously the more
- 22:41important thing is the AI tools that as an agency you can help
- 22:45that brand leverage, you know, from automatic, you know, all
- 22:49tag writing, translations, tone of voice, you know, all these
- 22:52kind of things like automatically building that into
- 22:54their, into their kind of like business OS, as it were.
- 22:59Yeah. So we're doing a slew of things.
- 23:00Part of it's around engineering, part of it's like the whole life
- 23:03cycle as well. Yeah.
- 23:05Yeah. And I think for merchants out
- 23:08there, e-commerce leaders, team members, these are the
- 23:11conversations you should be having with your agency partner
- 23:13to understand how they're operating and how it's impacted
- 23:16and, and what that could do to the way that you work in future
- 23:19is such an important conversation because you're, we
- 23:21touched this earlier. There's no right answer to this.
- 23:24Like people are on different learning paths at different
- 23:26speeds and it's dependent on the ecosystem you're working in, the
- 23:28tech tools, your, your platforms you're using.
- 23:31But this conversation is important.
- 23:33So you obviously work for one of the leading econ platforms.
- 23:37Where do you, where do you see it evolving to at the platforms
- 23:40side? Like what what is the role
- 23:42within commerce life, for example, of AI?
- 23:44Like how, how fast is that adoption?
- 23:47How much is being used to get to automate process and to to get
- 23:51those efficiency guides? Yeah.
- 23:52So internally from like an engineering standpoint, it's
- 23:55like a lot of the same stuff Jim mentioned.
- 23:57You know, there's going to be like efficiencies that the team
- 23:59can now build into how it creates code, how it reviews PR
- 24:03is how it ships code. But then we're at an interesting
- 24:09use case which is wholly different from agencies and SIS,
- 24:13which is, you know, we're an API.
- 24:15And so a lot of the times to engage with our API requires the
- 24:19customer, the merchant to give our API data formatted in a
- 24:26Jason object that's very complex to create.
- 24:30And like the most recent example would be like our, our promotion
- 24:33rules engine. And it's incredibly powerful
- 24:36promotion rules engine that can allow merchants to do really,
- 24:40you know, multiple different combinations of promotions.
- 24:44But to get the API to, to, to do that, you have to give it
- 24:48something in a very like code formatted way.
- 24:51And so the typical like e-commerce manager or marketer
- 24:53who might be trying to put together some type of like sales
- 24:56campaign for a suite of products is not going to be writing Jason
- 25:00that the commerce layer API can parse.
- 25:03So they need a developer to do that.
- 25:05OK, not ideal. So what we're able to now do is
- 25:09build AI agents, typical kind of what we hear a lot of these days
- 25:14of just using like an MCP protocol where like end user
- 25:18puts in some natural language. I want to do a set of promotions
- 25:21with this. These are the skis I want to do
- 25:23the promotions around and give me the Jason that I need for
- 25:26commerce layer. And then we can build out an
- 25:30actual interface that communicates with our API,
- 25:32returns the Jason, has that Jason validated and tested
- 25:36before the end user even gets it and gives the end user
- 25:38something. And as a response that they can
- 25:40just immediately go and create a promotion for.
- 25:43They can then copy paste that into the commerce layer UI to
- 25:47have built the like an actual user experience and user
- 25:50interface around the the promotions engine would have
- 25:54been incredibly difficult because of like, I won't say
- 25:58infinite number of combinations, but you understand, I'm saying
- 26:01like just the limitless number of combinations that any
- 26:03merchant could put together that utilizes the power of our rules
- 26:06engine. But with AI, with like an agent,
- 26:10we can expose a user experience that someone just types in like,
- 26:13hey, I want to do a promotion. This is what I want to do.
- 26:16And they get the Jason back already tested, already
- 26:19validated, and then they can move forward.
- 26:22So we have the the development stuff Jim talked about, but then
- 26:25there's like a product layer that we can now build into our
- 26:28API that gets people using the API much quicker.
- 26:31Yeah, I realized I was having a conversation at the.
- 26:33I was at a centre event last night and I was chatting with
- 26:36Martin. The CEO is now going to be the
- 26:39chief product officer about where platforms sitting and the,
- 26:43and this is a bit where I was saying is I think the the the
- 26:45agent part of helping people to solve like day in day out,
- 26:50common tasks where you don't want to take up your, your
- 26:55engineering resource, which should be focused on like fixing
- 26:58major problems and real value and thinking.
- 27:00Like you were saying earlier, Jim, but you've seen it now in
- 27:03things like psych, psychic and Shopify.
- 27:05And you know, every platform has its own way of giving people
- 27:08that that interface, but ability to create discounts are great.
- 27:13When I was doing it shop wise, I create me a free shipping
- 27:16discount that's only eligible and orders over X and is 1 use.
- 27:19And as long as you know your as long as you know the functional
- 27:22limitations of of the tool, what you can and cannot do from an
- 27:25eligibility exclusion criteria thing and you're very precise in
- 27:29it. It's Bam, it's done, it's set up
- 27:31and there's, there's and I for me, this, this, it's this part
- 27:35where platforms can add so much value to E Comm teams by
- 27:39speeding up decision making and tasks for like low complexity
- 27:45stuff. But then it's also the
- 27:46predictive bit of the chat interface that you can bring in.
- 27:49So you see that there are some predictive analytics tools out
- 27:52there where they will do analysis of the data and
- 27:56automatically give you like tasks of you could improve
- 27:58margin by doing X or this product is not selling well.
- 28:02But the ability for somebody within a a ecom platforms to be
- 28:05like, tell me which products are are best selling at Easter in
- 28:10France and have margin of over 25%?
- 28:14Yeah, things where strain teams just want answers and they can
- 28:17just get quit. Totally.
- 28:19Yeah, like we have a metrics API, but to have made that query
- 28:23in the past was difficult. Now we can build interfaces with
- 28:27agents that make those queries very simple.
- 28:30Another thing that I think is interesting about like commerce
- 28:33in general is that there is a physical product sitting in a
- 28:36warehouse somewhere that is an actual thing.
- 28:39It's not artificial. And so like the API still serves
- 28:43the commerce layer. API still serves A tremendous
- 28:45function of being able to like validate that that thing exists.
- 28:48It has a price, it has a certain currency that it can be
- 28:51purchased in. It has a certain market.
- 28:52There's all these business rules that need to be defined by the
- 28:54API and the API can be accessed by anything.
- 28:59So like, well, I think we're excited about is that future of
- 29:01like where agents just interact directly with our API because
- 29:06we're very defensible in that regard because like we know how
- 29:09much stock there is in a stock location.
- 29:12It's real, you know, this is an artificial stock and we will
- 29:16only make a transaction happen if that stock is there.
- 29:20And so I think like that's like a really interesting place to be
- 29:22as well as an API. There's a very huge promise
- 29:25there for like what this new agentic experience can bring to
- 29:28us. Like we have, we're like, hey,
- 29:30Shachi PT wants to own check out.
- 29:31Cool. You'll still need to know if
- 29:33there's actually a product there to be sold.
- 29:35You know I. Think Swans of agents, right
- 29:39where you are in this, you know, exactly as you say, Seth, right,
- 29:42you kind of go, we've got an MCP server, right?
- 29:44And it's you know, Shopify has have theirs.
- 29:47You know, you've got yours and you the great thing is it
- 29:50decouples it from the actual agent.
- 29:51So you have a custom agent and you as an e-commerce like
- 29:56merchant, you're there and you can say, OK, I've got a new
- 29:58promotion and your agents kind of swarm off, right.
- 30:01And there's these NTP connections with each with very
- 30:03defined somps and also they've got build up context specific
- 30:07range of business. The analytics agent is feeding
- 30:10back other like previous experiments.
- 30:12He knows what works. It sets up that you know,
- 30:14they're all kind of spawning off and doing these different
- 30:16things. You're right in terms of, and
- 30:17again, I suppose it comes back to that idea of you go to it's a
- 30:21commerce layer or even like Shopify, the amount of
- 30:24configurability around some of these things.
- 30:25It's just a level of like the abstraction where you can
- 30:28express your intent and that could be and you know, at the
- 30:31moment that's tends to be quite platform specific because the
- 30:34MCP stuff is pretty, it's really new still, right, in terms of
- 30:37it's not ubiquitous. So the idea that we'll get to a
- 30:40place though, where it's like actually, I can build an apps,
- 30:42an MTP server on top of the Shopify NCP server or a custom
- 30:45one, which starts down the amount of functionality that
- 30:47Shopify's offering or augments it and adds more.
- 30:50And as long as programmatically as like Seth was saying this
- 30:53programmatically, it's exposed, right?
- 30:55As long as you have AP is and stuff, you can kind of build out
- 30:58those servers for custom kind of fight rules.
- 31:00And I think you said right at the beginning, James, right,
- 31:03actually different, you know, different businesses are going
- 31:05to want have different requirements, right?
- 31:07The depths that they want, the level of configurability like
- 31:10they want, it's going to be completely different.
- 31:12If you think of CMS at the moment, some clients we give
- 31:14vast creative control too, because they have teams that
- 31:17have got creative expertise, you know, really amazing art
- 31:19direction and everything. And for others, we don't give
- 31:22them though that exposure because, you know, it more
- 31:25choice isn't always more freedom and more empowerment, right?
- 31:29Actually, often it's quite the opposite.
- 31:30You're kind of a choice paralysis and actually it's not
- 31:33allowing us to focus on what's really important.
- 31:35But yes, I think the next step where you know, through NCP
- 31:39services is kind of, you know, spoke about where you can
- 31:42actually have a agent, which speaks to all of your kind of
- 31:45core tools and kind of kind of, you know, orchestrate this stuff
- 31:49is is definitely where it's going to go, right.
- 31:53Yeah, link to that. Let's talk about this.
- 31:55This point about business is different because it's, it's
- 31:57always, it's always saying the hype is always every business
- 32:00must do this or they're going to die.
- 32:03Whereas the reality is we know that you have a start up that
- 32:07just wants a basic store and to sell.
- 32:09You have ADC brand that's in fast growth and actually they're
- 32:13accelerating rapidly and their teams expanding almost monthly.
- 32:17You have huge enterprise and I'm talking enterprise from a
- 32:20systems complexity, not from a revenue point of view where they
- 32:23have, you know, a crazy systems architecture and they have
- 32:26insane compliance requirements and massive discipline around
- 32:30they, the, the security piece you talked about earlier.
- 32:33And then what I'm working with a brand that's part of the Nestle
- 32:35Group that every brand in that company has to adhere to the
- 32:38security and compliance policies of, of the, of Nestle.
- 32:41And when you've got public listed companies, we all know
- 32:44the amount of scrutiny that they get under and they need to be,
- 32:48to be really spot on on that. So what, what in you in you how,
- 32:53how does this play out in terms of how do you put safeguards in
- 32:58to the relationship between a development partner and a
- 33:03strategic partner like yourself and a company that might want to
- 33:08do some of that, they might want to do some cogeneration cells.
- 33:11What are the safeguards and how do you build those in so that
- 33:13you don't get this divergent where some some in house people
- 33:17have created some code that's screwed up a theme or a site or
- 33:21create a compliance risk? Yeah, I suppose it's how you
- 33:25leverage AI now for the business.
- 33:28We talked about like the whole life cycle of a feature, right?
- 33:31And we said actually a lot of it starts with the objectives and
- 33:35your goals and challenges and opportunities and then coming up
- 33:38with like an idea. And I suppose actually when you
- 33:41come up with an idea, it's a risk, right?
- 33:42You're going to spend money on engineering to kind of build
- 33:45that out. And then at that point you find
- 33:47out if it works, it actually has an impact to your business.
- 33:49So you're building up risk, you know, and as much as it's
- 33:53obviously valuable to build the, you know, something, well, it's
- 33:55more valuable to build the right thing, right?
- 33:57So in terms of how I suppose businesses now can like really
- 34:01leverage AI within like safe, safe constraints, I would say
- 34:05like first of all, PO CS like use lovable, use like a bolt to
- 34:09kind of actually build out your ideas, test them internally.
- 34:12And when I say internally, I mean, you know, maybe get small
- 34:14pools of customers, you only need like 5-6 customers to start
- 34:18actually getting a good feel for things.
- 34:19You don't need massive like sets of data.
- 34:21And what you're doing here is you're de risking any initiative
- 34:24that you do. You're also like flexing your
- 34:26kind of products thinking kind of muscle within the business to
- 34:28kind of go actually we've got an idea of something.
- 34:30We're going to do quick proof of concepts.
- 34:32We're going to understand, you know, the kind of edges of this
- 34:35and the end of that. You're in a position where you
- 34:37can actually get AI to look at your proof of concepts and
- 34:40actually create a spec and then you can hand that over to an
- 34:42engineering team to kind of produce production kind of
- 34:45quality code. If you if you're not doing that
- 34:48and you actually want to take it step further and get things into
- 34:50production, you have to do it within really safe parameters,
- 34:53right? I mean, Shopify is probably one
- 34:56of the few exceptions where you're like, actually this is
- 34:58kind of safeguarded within Shopify.
- 35:01You know, you're not really getting even if you're an
- 35:04engineer in there, there's like quite good safety, kind of like
- 35:06restrictions, you know, the check out changes to check out
- 35:08changes to account area. These are all about compliance,
- 35:11privacy like and and safety around kind of like commerce.
- 35:16So if you're going to go to that step and you're going to do
- 35:18things like the code generation, you know, block generation that
- 35:21you mentioned earlier, James, you'd kind of, you know, fine.
- 35:24If you want to take that step further, you do that.
- 35:26You speak to your technical team first, understand that you're
- 35:29not going to create any risk around this stuff.
- 35:30And then when you've got something that you're happy
- 35:32with, then use, you know, testing in Shopify's not great,
- 35:36but use something like intelligence, use 5%.
- 35:39Make sure you've got testing for like automated testing for end
- 35:42to end. Make sure you've got alerts
- 35:44through, you know, for tracking when things kind of drop and
- 35:48then just keep monitor it, keep an eye on things, make sure that
- 35:51you are seeing statistical significance that this is an
- 35:53improvement. And at that point, again, you
- 35:55can either get it built flight production quality, you can have
- 35:58an engineer just review to make sure you're not introducing
- 36:00technical debt. So in terms of, and again, this
- 36:03comes down to where you are as a business, right?
- 36:04As you said earlier, James as well.
- 36:06So you know, what's your app type for risk, You know, what is
- 36:09the true risk here? How much revenue is going
- 36:11through this store? All these kind of questions, but
- 36:13as a very minimum, you can start as an as a, you know, as a
- 36:17merchant, as a team to kind of experiment with PO CS, you know,
- 36:20push that through internally as much as you know, we try and
- 36:24understand as much as we can about the merchants, but
- 36:26merchants, but they ultimately are the product experts and they
- 36:28understand their customers and understand their products better
- 36:30than we do. And for them to experiment with
- 36:32some of these things to kind of validate ideas to a certain
- 36:35level is definitely going to save costs and also help.
- 36:38I said like building that kind of products model muscle within
- 36:41the internal teams. I think that's like where I
- 36:44started, you know, I never wanted this to be anything but a
- 36:47prototype. So if you were to imagine myself
- 36:49being like a mid size or a large size enterprise merchant who's
- 36:52considering shifting to something like commerce layer or
- 36:55any of the other composable approaches and the internal push
- 36:58back from like the CTO or the VP of engineering is like, well, I
- 37:01don't know, like this works. If it if it's like not broken,
- 37:04why? What are we trying to fix?
- 37:07I could prototype a non engineer like myself.
- 37:09I could get a working prototype using commerce layer up in like
- 37:11under a week. It allows me to then show make
- 37:15my case internally where it's like, look, we have like I have
- 37:18multi market support with like SKU lists across all the markets
- 37:23with different currencies. And these are things that we're
- 37:25limited by with our current platform.
- 37:26And I did it in a week using this thing.
- 37:28That internal prototype can really drive change internally.
- 37:32So like that's like a great use case of these coding tools for
- 37:35for sure. It's like internal prototyping
- 37:37to prove more than just like a a concept, like show a thing that
- 37:42actually works. Massively de risking it, right?
- 37:44Hugely. You know when you do something
- 37:46like that. But this is where from, I mean,
- 37:49that's a a more advanced level at the I'm, when I'm thinking
- 37:53about most, most things I work with, it's this trading of the
- 37:58site that is their priority, right?
- 38:00They are day in, day out, updating content, managing
- 38:02products, doing promotions, campaigns, landing pages, you
- 38:04name it. And they want quick turn around.
- 38:06They want to see stuff and and you know, we've moved the lobbis
- 38:11has moved from the old legacy staging, production replication
- 38:15thing and doing a lot through like themes and like, you know,
- 38:21visual previews and the ability to have a test theme that is not
- 38:25impacting your production one and just to be able to mess
- 38:28around with code generation on for me, it's ideation.
- 38:32It's not production ready code, but it's OK.
- 38:34I want to landing pages got this kind of structure and these
- 38:37types of layouts and functions, right.
- 38:40Hey, AI build me a layout that now let me have a look at the
- 38:45the shell. Actually, that looks all right
- 38:47now. I kind of wanted to do this,
- 38:49though. Great.
- 38:49Now I've I've gone through the ideation, I've refined my
- 38:52viewpoint of what that page should be.
- 38:55Now I've got a tight design spec to give to my expert development
- 38:59partner to turn the raw code into production ready code and
- 39:02think about those other things. So I can see those AI generators
- 39:06not being the final end product, but being brilliant at speeding
- 39:09up that that design thinking within within client side teams.
- 39:15Yeah. For sure, I can go really
- 39:17quickly on this. And then Jim, I know you can
- 39:18take more of it, but like what I did for mine, James, is I like I
- 39:23did a as part of my initial prompt, I submitted screenshots
- 39:27of like websites that I liked that I know our commerce layer
- 39:31customers. And so like Solomon, for
- 39:34example, is an outdoor gear store.
- 39:36And so I was quickly could do some screenshots of like the
- 39:38flow from their main landing page to like a like a product
- 39:41listing page to product details page.
- 39:43I submitted all of those as part of it.
- 39:44I said, and by the way, make the design kind of look like this
- 39:46and the initial design sort of look like that, which is really
- 39:50cool. So you could imagine like if
- 39:52internally I had a pre-existing design and I thought there was a
- 39:55competitor website that I really liked, or maybe not even a
- 39:58competitor, just another website that I really admired.
- 40:01When I submitted those design requests to the prompt, it gave
- 40:04me something that sort of looked like that rather than just like
- 40:07coming up with something out of the blue.
- 40:09And the nice thing about that is if you're getting these like
- 40:11pre, you know, contrived like themes and you're continually
- 40:15told by your engineering team, like, sorry, that's not going to
- 40:17work with the theme builder we have, you know, now you can
- 40:19actually maybe push the push the envelope there.
- 40:22But anyway, Jim, I'm sure you have a lot more to add to this.
- 40:26And yeah, I mean, I suppose again, it comes down to, you
- 40:29know, what you, what you're trying to achieve.
- 40:30I think if you look at Shopify, you know, Horizon, right?
- 40:33And you kind of go actually this is 1 theme.
- 40:35There's loads of different iterations of it because it's
- 40:36got so many variables under the hood, like.
- 40:39And I think that's really interesting because actually you
- 40:41get to a point where you're not worrying about, you're not
- 40:43actually doing code generation. You're doing, you're kind of
- 40:46taking intent and you're setting, making loads of
- 40:48settings, entering kind of copy images and all this kind of
- 40:51stuff to actually realize that. And I think that's where it
- 40:54gets, but I think that's the side where I think is near term
- 40:57really, really powerful. So if you kind of went OK, and
- 41:00there's no risk there as well, not in the same way because it's
- 41:02kind of, you know, it's kind of protected.
- 41:03You're not doing original kind of code generation.
- 41:05So if you're kind of going actually, and Shopify will move
- 41:08forward to this way more aggressively than they are at
- 41:10the moment. But if you kind of said actually
- 41:11I want a new landing page and then this is what I'm trying to
- 41:13communicate. And this is the sort of audience
- 41:15and they're like, OK, here you here, you blocks.
- 41:17Here's how it's laid out. Here's how we're communicating
- 41:18this. You're like, cool.
- 41:20And that is production ready, right?
- 41:22And that's kind of ready to go. And I suppose I said in terms
- 41:26of, I suppose there's a few different things we're talking
- 41:28about. There's prototyping and
- 41:29validating ideas early on. So you're de risking things.
- 41:32Then there's actually code Gen. within quite a walled garden,
- 41:35quite a guarded and safe space where you can do
- 41:38experimentation. But that actually that
- 41:39experimentation isn't code generation, it's settings and
- 41:43underlying technology changes, which is again, really, really
- 41:46safe. And then you've got actual, you
- 41:48know, unique code, which if you're doing that, then there's
- 41:51obviously a slew of safety precautions that you have to
- 41:53make and definitely split testing and making sure you've
- 41:56got those kind of garden rounds in place.
- 41:58Yeah. And I've got, I say one further
- 42:00question. Christ, I could talk about this
- 42:02with you guys for hours because I think it's really, really
- 42:03interesting and it's so valuable for for all of us working in the
- 42:07ecosystem to have these conversations because this is
- 42:09evolving and it's evolving fast, as you both said at the start
- 42:12and, and having a short term and then a mid to long term view.
- 42:15And this is really important and the mid to long term will
- 42:17evolve. We know it.
- 42:19But I think having a short term view of how you will use and
- 42:21explore and change is important. But my question, I'll come to
- 42:26you first. And Seth, because you raised
- 42:27this to me when we, when we first discussed this episode,
- 42:30which is e-commerce teams are going to need to understand
- 42:34prompt engineering and writing accurate prompts because they
- 42:37will, whichever tools they end up using, whether it's within a
- 42:40platform like a Shockfire commerce layer or it's a, yeah,
- 42:43a specialist AI tool that they've got all.
- 42:46They're using interfaces like the chat GBTS, understanding the
- 42:51structure of, of the data that you put in and the impact it has
- 42:55on the quality that comes out. So do you think all ecommerce
- 43:00teams need prompts and a new strategy or is this on a case by
- 43:04case basis? I think every team, every
- 43:08commerce team will need a prompt engineering strategy to some
- 43:11extent. I think every marketing team is
- 43:12going to need a prompt marketing strategy.
- 43:16And I think agencies and GS is, will need to build in higher
- 43:21engineers who have a vast like knowledge of like how to write
- 43:24code and then prompting in the most efficient way because like,
- 43:29if done properly by an expert, the velocity by which you can do
- 43:34something is tremendous. And like, I know this is sort of
- 43:39like a, it's sort of like pop psychology, but there's that
- 43:42famous like book by Malcolm Gladwell that talks about 1010
- 43:45thousand hours to master something.
- 43:47And so Jim, you've been a developer like your whole life.
- 43:50I don't, you know, it's certainly well over 10,000 hours
- 43:52of like writing code. And I, I, I'm not saying that I
- 43:57produced anything that was expert, but like in a matter of
- 44:01a, you know, several hours over the course of a week or two, I
- 44:04was able to create something I never would have been able to
- 44:07create before. But took that, I took the amount
- 44:09of time and I like divided it by 10.
- 44:12I divided it into 10,000 and I saw that like I had a 500X
- 44:16increase in my own productivity based on how many hours it took
- 44:19me to build this thing. I think it was like 20 hours
- 44:22total. And, and like, the point being
- 44:26that like, I, I, I, I think prompt engineering would have an
- 44:30even higher factor of like productivity gains.
- 44:33So everybody in my opinion, needs to be thinking about that.
- 44:36If you're hiring an agency or a GSII would ask about what your
- 44:39prompt engineering strategy is. If you're interviewing as an
- 44:42engineer or interviewing at an agency, I would ask them what
- 44:45their prompt engineering strategy is because yeah, I
- 44:49think it's critical. Jim, what's your take?
- 44:53Is this something that's already coming up with clients you'll
- 44:55work with? Is this something you're
- 44:56advising them on? Just getting over Seth's
- 45:00comments. Yeah, no, I, I think it comes
- 45:05down to a few things. I mean, I think agencies, their
- 45:09roles going to change a bit, right?
- 45:10As in terms of, you know, Ron just being creators, they need
- 45:13to be educators and curators, right?
- 45:15And a part of that kind of prompt strategy, I suppose, is
- 45:17actually being the being, you know, we want to empower and
- 45:24accelerate and, you know, the people we work with.
- 45:27And part of that means actually making sure that it, rather than
- 45:31having another kind of conduit, you know, to kind of like to
- 45:34these things actually, we want to kind of make sure that the
- 45:37tool is curated around them. At the moment, you're going into
- 45:39fairly generic agents like Seth is right?
- 45:42And he's kind of going and you're saying here you go.
- 45:45Now we as an agency said, actually here's all the context
- 45:48that you need to kind of know AI.
- 45:49And here's like a bit about the brand and all these kind of
- 45:51things like you're saying for marketing for like all the
- 45:54different kind of verticals within the business.
- 45:56And actually I'm having quite a distinct identity, right?
- 45:58And so you're kind of going, OK, I want to engineer something and
- 46:02it's like, OK, as a business, this is what you use here.
- 46:04The systems that you use here are the considerations that I
- 46:06need to be aware of. And I suppose part of the agency
- 46:11role, I suppose evolving as part of this is actually kind of
- 46:13going, how do we make sure that you are getting the most out of
- 46:16AI, right? How rather than just like the
- 46:19agency kind of having a strategy around this, how does the agency
- 46:21make sure that the client has a strategy and that we empower
- 46:24that and we make sure the efficacy is, is, you know, as
- 46:27great as it possibly can kind of be.
- 46:29And I suppose that's where I see at least like business, like
- 46:31rotate, you know, where that kind of moves to, right?
- 46:33Like we are composable in our DNA as as an agency.
- 46:38And if we talked about MCP servers and all this kind of
- 46:40stuff, how do we make sure that we are empowering, empowering
- 46:43the, you know, the the merchant as much as possible, you know,
- 46:46across this kind of like tool. So yes, you know, in terms of
- 46:50Seth's comment about like I've done my, my 10,000 hours in an
- 46:53in five hours away, like I suppose it's, you know, it's not
- 46:59as simple as that, I suppose. And also, you know, and I know
- 47:02that probably to be to be controversial, but I suppose
- 47:05it's a bit like you can recreate Picasso.
- 47:08But I suppose the kind of, you know, the journey you've gone on
- 47:10to kind of to kind of get to that is obviously quite good.
- 47:13It's quite different and the understanding and the expertise
- 47:16you can kind of bring around. That is obviously.
- 47:18I think the point I'm trying to make is not that I'm an expert.
- 47:21The point I'm trying to make is like, I had no capability to
- 47:25build what I built and I did something that I would consider
- 47:29an expert like you, if we're actually production level code
- 47:33would need to have spent their lives and being becoming an
- 47:37expert. And I'm not saying what I did
- 47:38was expertise. I'm just saying I spent about 20
- 47:40hours and built this like fairly complex web application, which
- 47:43was a was an impossibility 6 months ago.
- 47:47And it led me to this belief that like, if I were a if I were
- 47:51an e-commerce merchant hiring an agency, I would really want to
- 47:55know their fundamental understanding of all the
- 47:59security and guardrails they're putting in place for their
- 48:02prompt strategy, rather than just handing over the keys to
- 48:05the Kingdom to someone like a set who goes off.
- 48:08And it's like, I built you this thing in 20 hours.
- 48:10Ship it. Like I would want to hear from
- 48:12you, Jim. Like, no, no, guys.
- 48:14Like let's step everything you've just been saying.
- 48:16What do you want to do? Like, what are we building?
- 48:19Let's talk about the process. And yes, we use this stuff to
- 48:22make our code even better, but like, it still doesn't, it still
- 48:25doesn't pick all the boxes. And like, I think that's the
- 48:28thing that I would be prying into is like, someone needs to
- 48:31have a very solid answer to their internal prompt
- 48:33engineering strategy to give me the confidence that they know
- 48:36how to use these tools properly. Rather than just like, Oh yeah,
- 48:40we use cursor and all of our developers have a seat and we
- 48:42just all crank out code. That's that's a bad answer for
- 48:46me. Like I want to know how you're
- 48:48like making good code Knuckling go live.
- 48:52I think this is this is a really good way to summarise this is a
- 48:55is a that every e-commerce team should be thinking about what
- 48:59the role of of AI tools is in their business to give them
- 49:02like, you know, fast speed to mark great efficiency reducing
- 49:06cars. But they should understand that
- 49:08that using AI to generate code introduces risk to the business
- 49:12if you do not have the technical expertise to QA it and ensure
- 49:17that it is compliant. And that's why proper technical
- 49:20leads, whether you're have internal engine or you're
- 49:23working with a third party like a, like a rotate that that is
- 49:26critical. But yes, in the relationship
- 49:30with your agency, you need to explore and understand what the
- 49:34decisions they're making and how they are changing, how that
- 49:36impacts you and what the, is there anything there's a red
- 49:40flag that that it raises a risk to business?
- 49:43Or actually is it creating an opportunity where you can work
- 49:45smarter with them to do things faster or get more through
- 49:49because of the way that they're, they're building efficiency.
- 49:51So I think this is a really exciting time.
- 49:53But I think the word for me is caution is, is with all good new
- 49:57shiny stuff is don't get, don't think it's the, the, the magic
- 50:02pill that cures everything. You still need good governance.
- 50:05Yeah, I mean, that's the point I think Jim, you've been making.
- 50:08I'll give you the last word. But I think my final word is
- 50:10something I heard just yesterday in like a lecture and it was
- 50:12taught this person was talking about like AI will never tell
- 50:15you what it doesn't know because it doesn't know what it doesn't
- 50:19know. It's it's it's trained to give
- 50:21you an answer. And so like for a, for like a
- 50:25less structured, like team that uses AI for everything, you'll
- 50:30step into some hallucination and you'll be getting things.
- 50:33And if you're not an expert yourself, you won't know how to
- 50:36debug what you're getting back. And the same goes for marketers
- 50:38who are trying to use AI to do everything, AI to write copy,
- 50:42like you're going to get generic copy that isn't going to do
- 50:45anything, whatever the the case may be for, for marketers
- 50:47themselves. But I think like that's a really
- 50:49fascinating concept is that stepping into it with caution
- 50:52and understanding that like some of the best doctors that are out
- 50:55there are the ones who actually say, I don't know, I don't know
- 50:59what, I don't know what to do next.
- 51:00But we're going to have to experiment our way into like
- 51:02solving this like illness that might be hitting you.
- 51:06But doctors are always like, oh, this is what's happening for
- 51:08sure. And then like you go through
- 51:09this thing and it never gets healed right?
- 51:11Then that's a problem. So I think like that's like a
- 51:13the, the wisest people are going to be the ones that are like
- 51:15ready to say, we don't know. Sure, we use AI to help us, but
- 51:19we don't know how to solve that. Let's work on it together.
- 51:22That's why, you know, thematically, right, in terms of
- 51:25AI is an opportunity, right, when we talked about value and
- 51:29empowerment and ultimately as an agency, you have to provide
- 51:32value and you should be looking at how you can empower them to
- 51:35so that your collective velocity as fast as possible.
- 51:37I think James just said to me, you know, we want to be working
- 51:39on the problems that only we, you know, that we only that we
- 51:41have to because there's no one else who can kind of work on
- 51:43those. And we want to empower the
- 51:44client as much as possible to work on, you know, elsewhere.
- 51:47So AI it's an amazing, I mean, it's incredible, it's an amazing
- 51:51opportunity, it's changing weekly and it's so exciting in
- 51:53terms of how we can increase the value that we kind of delivered
- 51:55to kind of customers. But I suppose on both sides, if
- 51:58you're not moving as an agency, if you're not adopting this
- 52:00stuff, if you're not looking for new ways to add value, you're
- 52:03going to unfortunately become, you know, you will become
- 52:06irrelevant. And that's unfortunately always
- 52:07been the case. You know, that's never been any,
- 52:09you know, any different. If you're not providing value,
- 52:12you know that there's a problem, right?
- 52:13And AI will radically change what value means because many of
- 52:17the actions that we take today that are deemed as valuable will
- 52:20become productised and and diminished around that stuff.
- 52:23Yeah, and just for clarity, you know, listen, what we're not
- 52:25saying is if you don't do these three things, your agency's dead
- 52:29because we're not the hypists. But look, Jake, I could, I could
- 52:33carry on chatting for for hours on this.
- 52:35I'm not going to send everyone else listening to sleep, but but
- 52:37thank you. Oh, so Seth, thank you for soda.
- 52:40Thanks for coming back on Jim. Always enjoy chatting to you as
- 52:44well. I have a lot of respect for what
- 52:46you're both doing and anyone wants to reach out, anyone's
- 52:49listening who's interested in being Pete or fundamentally
- 52:52disagrees and wants to run stars do reach out and I'll have
- 52:56there's contact details on the landing page on the website, but
- 52:58you can find Seth and Jim on LinkedIn and they're always
- 53:03happy to chat. And if you've got projects where
- 53:06you are looking for somebody to give advice and guidance on,
- 53:08well, these are the these are the gens to speak to.
- 53:13You're listening to the Inside Commerce podcast hosted by
- 53:15experienced consultants James Gerd and Paul Rogers, Inside
- 53:18Commerce is your fast track to digital knowledge feature in
- 53:21interviews with leading e-commerce retailers, industry
- 53:24Thoughtleafs and the C-Suite key tech companies to explore the
- 53:26latest digital strategy, customer experience and tech.
- 53:30Discover what the latest trends really mean and get practical
- 53:33advice based on experience to help fine tune your E Comm
- 53:35strategy and tech. Follow us on LinkedIn and reach
- 53:39out to James of Paul if you'd like to discuss an e-commerce
- 53:41project. We hope you enjoy the latest
- 53:43episode.