While enterprise has long embraced the AI change and is driving a narrative about how to use these tools, at Aperic we’re equally interested in AI within SMEs. With increasing pressure on the UK services sector a lot of businesses we talk to are doing great work, but seeing their bottom lines eaten away year on year. One of our passions is trying to fire up small businesses to innovate and take AI beyond Gemini meeting notes and a bit of Chat GPT. To whit, we sat down this week with Recording Artist, Producer and Small Business Owner Rupert Goddard to talk about these challenges and how he’s using AI across both of his lives.
Alex: For the readers, give me a bit of background about yourself. Who are you?
Rupert: I mean there's two sides of [AI] that I use. I use it creatively. I write music. I'm not really a filmmaker, but I have tried to experiment with making films originally as music videos, but also just because I wanted to see what could be done with it. So I've done a fair bit of that. And then I spent most of my adult life as a partner of Goddard Consulting, which is a specialist firm of construction health and safety consultants.
Alex: Okay. So shall we talk about the creative bit first?
Rupert: Yes. I write music as Neon Transmission. I've changed the name slightly to Neon Diffraction for the slightly more experimental and breaks stuff.
Alex: I know that you have started with using some AI in your production chain. How does AI fit into the mix?
Rupert: I've changed what I'm doing recently and I'm being a bit more careful about it because the level of backlash that there has been is…significant. So actually now anything that I do that's 4/4 as Neon Transmission, which is the stuff that I mainly release through labels, I won't use generative AI at all. It's interesting how you get to what people are happy with and what they aren’t because we've had AI mixing tools for a long time.
Alex: That’s true. Everyone was cool with the AI mixing tools.
Rupert: EQs and compressors, all of them have got a little bit of it in there. I've obviously been doing this a while, so I was quite interested in the generative part of it as well, but I have been surprised at the level of backlash that you get against Generative AI.
That’s part of the reason why I've now split it slightly, so that the stuff that I do as Neon Transmission will not have any AI in it. It's officially AI free. It has to be officially AI free because otherwise you can't get anything out on a label. Whereas I can put stuff out that does use Generative AI as Neon Diffraction.
Alex: So it’s forced you to split your creative output?
Rupert: I think that you have to. I've tried to explain this on forums and things to other people. [AI] doesn't necessarily make it any easier. It's just a different process. If anything, it takes a lot longer to do some of the stuff that I've been doing that has used Generative.
I think people just imagine if you're using an AI creatively, all you’ve done is write a prompt. You can say "Suno, make a song" and it will make you something. Or you can try and use it the way that I use it, which is generating song specific samples that you can mix into a wider piece.
Alex: So like a vocal or a drum sound.
Rupert: Like a vocal. And that's the bit that people got really bent out of shape about, is people creating vocals. I have used vocals and I have used various products to generate some instrumentation. I've not found an AI that can create good drums.
Alex: The reaction to Generative is so mixed, do you think that if it's replacing something that's obviously the creative input of a human being, is that the red line that's emerging?
Rupert: Yes, but replacing people is not new. AI has already replaced a lot of Mix Engineers. It’s not particularly perfect, but you can press a button in Logic now and it will master your track for you using AI.
Alex: So do we just not care about mix engineers? [joke]
Rupert: I do find the morality of it, particularly in dance music, quite interesting in that we were happy to rip off people's samples in tracks in the 90s and hope to get away with it. But the fact that it’s now gone through an AI is somehow morally intolerable? Essentially, if you use Generative you can’t put anything out on a label.
I think it’s an argument about how AI is used. Nearly all music uses synthesis of some kind these days. What if you use Claude Code to write a VST plugin? Is that not ok because an AI made the synth? Or to go one step further; should angry Violinists be lining up because we synthesized violin sounds? These were arguments made at the time going back to the 80s.
I've even had some people saying that we're not happy with AI sampling but we don't mind AI stem splitting because it's built into Logic. Why? Stem splitting is really honing in on people's samples. And most people aren't seeking clearance, I don't think, for that.
Alex: Is there a world in which you use Generative for something that's placeholder or use it like a way of sketching and then re-record it with a live artist?
Rupert: You could do that. And I don't think people would mind if I then, for example, made this album and found a blues singer to re-record the vocals. But it’s not guaranteed.
There's some stuff I've done with spoken word. I don't really like the sound of my own voice so I've used ElevenLabs to change it. Is that cheating? I wrote the lyrics. The intonation is mine, all I've done is changed the tone using ElevenLabs.
The worry is that there’s a perception that if AI has been used, it’s slop. There's been musical slop forever. It's just been made by people who aren't good at making music. I have to be careful because I don’t want to be labelled, unfairly, for making AI crap.
In fairness, I think there’s some moralizing and hair-splitting as to what people like and what they don't like. And my general view is it's the result that matters. It's only slop if it’s rubbish. If it's good, then does it matter how it's been made?
Have you heard my curry analogy on this? If I order a curry from Deliveroo and serve it up to you and say, "I've made this curry," then clearly I'm lying. If, however, I invent a curry in a Michelin-starred restaurant, but I happen to use as part of that a pre-chopped jar of garlic, then no one's going to argue that I haven't made a curry. I've just used an ingredient to speed up the process.
Everyone seems to be happy with those two extremes. It’s the halfway point that people argue about; If I make a curry using a jar of curry sauce, have I made the curry? And some people will say yes. And some people will say no.
And it's a bit like that with Generative AI. It's a battle for authenticity.
Alex: I get it. I can see all sides of the debate. I just don't think the world has made a final decision because we haven't got to the final stage of the tooling yet. The AI tooling is super accessible and so I think people react to it because they feel like “I spent years learning to do this thing.” Now a machine can barf something out instantly. It’s depressing.
Other people, and I probably sit in this camp, have a problem with it because the Generative training data is not always paid for.
Rupert: Well, I really feel that. I've had my music ripped off by people who are more famous and well known. I was a bit annoyed when I released a track called "These Games"and then I heard the vocal ripped off pretty much completely in [someone else’s] song. Is it theft, or is it inspiration or homage?
Alex: So there’s danger of doublethink; it’s ok if a human “trains” on your output but not if an AI does? I think Sam Altman would like that argument.
Rupert: I worry that particularly with dance music, there’s a certain amount of moralising and hand-wringing, when everybody's been perfectly happy to “be inspired” by what has come before, forever.
Alex: Really? Are you accusing the dance music community of occasional hand-wringing? [joke]
Rupert: Oddly enough, using Generative for music has been much more recent for me because Suno has only recently gotten good enough to be able to generate anything usable. But for video I started off using Stable Diffusion. And now these days I would do everything in Midjourney. The ability to keep characters constant and animate, it allows you to do it all in one place.
Alex: I think you've chosen the hardest possible things to try and keep on track creatively. Generative by its nature will just wig out and go off on crazy tangents. It requires a lot of wrangling still.
Rupert: Yeah. I think the bit that's slightly annoyed me about all of it, is that you get this massive backlash and the assumption is that it’s super easy to do. And I think unless you've actually tried to do it in production you don’t know how hard it is.
Alex: So do you think you’ve got to a place with Generative where you’re almost being restricted by people saying “I don't want your thing if it's got AI in it”?
Rupert: Exactly. And that's slowing adoption in a creative space. The real pushback's been in the last six months, I would say, with people saying "I really don't like this." And I can see why given that there is a lot of slop that is infesting social media and creative spaces.
Alex: When it’s so easy to create stuff in volume at low quality there’s a worry about a tsunami of crud. It’s creative Spam basically. We’ve been here before.
Rupert: I wonder if it’s about creative appropriateness or context though? Some of it's quite good. Did I tell you about the care home, dad's care home? I was with Dad at Christmas and everyone was sat there in front of a TV watching Christmas carols. And I was like, "Oh, some of these are quite good." so I tried to Shazam a couple of them and didn't come up with anything. And then I noticed the screen was AI visuals. And they're just sitting there. It was shocking. But also fine? I’m conflicted.
Alex: This is what I find fascinating. Sometimes you stumble on these use cases and you can just see a glimpse of the future, right? Because to do that now you have to have a PRS license, you need to select the carols. You need to train your staff on how to put Spotify on or whatever. In the future you just press a button and it spooges out infinite carols.
Rupert: I don't know. I reacted to it but they all quite liked it. It fooled me. My initial thought was, "Oh, these aren't real songs." But they are real songs.
I can understand the pushback. I am respecting the pushback. And so anything where I’ve used Generative, I will just park for a bit and see how it shakes down. It’s not going to be too long before Generative functions are just built into Logic. And then you’re relying on people's honesty.
Alex: All tech companies are desperately trying to put AI in their products because it’s the only way they can get investment at the moment. So I think as a creative person, actively avoiding AI is almost going to be a job in its own right.
Rupert: It’s getting harder. Anyway, as a result, I am going to split my releases. I’ve got an EP which I will self publish which uses AI - and I’m entirely transparent about that - which will go out under Neon Diffraction and keep the Neon Transmission title for 100% human.
Alex: Changing the focus to Work. I know you have adopted AI in a very responsible way within Goddard Consulting. What and how are you using these technologies?
Rupert: We use ChatGPT as our main product. We pay for it, so we have made a variety of GPTs for different purposes and use Midjourney for anything graphical.
We have a Business Development GPT, which knows everything about Goddard Consulting and all of our projects.
We have a Contract Reviewer that reviews any contract that we have. There's particular terms that we look for, like insurance levels and things like that which are essential to catch.
We have a Business Bid Assistant. We enter a lot of formal RFPs and are asked a lot of fairly standard questions but need them rephrased according to the specifics of that RFP. This automates the process of pulling all that information together which is essentially the same every time but needs to be provided in a specific format or in answer to a differently worded question. It’s all about efficiency.
When we get asked to bid on things, we often get a huge amount of information about the project. All we really want to know is how much you're spending. What are you building? When do you start on the site? When do you finish on site? Once you know that, we've got a pretty good idea how to price it. What we don't need to know is what your ecology situation is and whether your social benefit analysis is significant.
And then we do have something that we call the Property Reseach GPT. We can put a Property into this and it will research the property and compile a report for us. That's often quite useful if you're going to an initial meeting as a briefing.
Alex: So in our tech verbiage we would call these “single use apps”. Tasks in business that you do pretty much regularly and have a reasonably sort of consistent structure to them. How did you train them?
Rupert: We’re a consultancy so we have versions of documentation going back literally decades. We curated that a bit and fed them the best examples but it picked things up quickly. We’ve set constraints on what each GPT can and can’t do and we’ve trained all of them on our tone of voice and things like that. How you want it to write.
Alex: And those are things that you've felt confident enough with to use them actively in course of business?
Rupert: Absolutely. These are really good now. The contract reviewer is so good! Contract reviews are not the most interesting thing to do and are often very complex so this saves a huge amount of time, even though a member of staff still reads them all. Most importantly, no one is allowed to sign a contract unless it's been through this tool.
Alex: Ah, so that’s already inverted your process? Rather than worrying about the AI output and a human has to review it, it's the other way around?
Nobody is allowed to sign anything unless it's been through this tool because it is very, very accurate. That having been said we are extremely careful with the training of our staff in terms of what they can and can’t do. But I think Elon Musk said it about auto-driving, “It doesn't have to be able to drive perfectly. Just has to be able to drive better than a person.” And I stress; everything still goes through a human expert.
Alex: Great, so in tech world I would say you’ve isolated and provided good quality training data. You’ve defined your prompts so that each GPT is nicely constrained - which is why single use apps are probably more accurate at the moment rather than one thing that does it all. You’ve also set context well and provided meta information like tone of voice, which is auxiliary to the actual task. I like that you’ve been able to do that quite instinctively.
Rupert: In addition to these, some people are using Grammarly, which is technically AI. And then Base44 is how we have developed our software product for the building trade.
There’s a lot we still want to do. I think it’s going to be good for trackers. We have things like a Project Master tracker for the business which at the moment is an Excel spreadsheet. And I would like to turn it into one of these.
Alex: This is really significant in terms of the progress of business AI usage. This is halfway to agentics. The whole point of agentics is that it's one layer on top of this. So right now you have a person, they have to come to this tool and they use the tool. There's a lot of safety in that and it's easier to build compliance around that because the output is contained.
Agentics are sequenced AI tasks and the agent is smart enough to be able to say,
based on the outcome of one step, change or follow a set of commands downstream. At which point you get to “Here's an RFP, carry out the work.”
Rupert: Well, yes, we haven't got there yet.
Alex: We only have a few minutes left, so let's talk about your product. You have built a business system from scratch. The world would refer to it as vibe coding. I don't like that terminology. I call it non-technical prototyping, but that's because I'm boring and I've done this for too long.
Rupert: I wanted to make something that is accessible to smaller builders. There are very good systems for big projects which are very expensive. There's nothing available for smaller builders doing relatively straightforward jobs. This is a product where you can create a construction project. You can do all of your site inspections from this. And it works on a mobile phone as well.
Other than cost to purchaser, what we noticed was that all the big platforms are designed and priced around having tons of people using them. So we have divided this into single users and for a very small builder just doing it themselves, they could have all the rigor and the support of a big system but in a much easier to use package. We do support multi-user but it’s scalable rather than forcing that on you.
Alex: What I find so exciting about this is that in the Software Product Development world, everything was built around big teams doing quite long processes to extract this information out of you. The genius of this is that the AI tooling has allowed you to just - for want of a better word - brain dump your idea and make something real. I wouldn’t class this as Production Hardened yet but there’s so much here that you could happily Beta it with caveats or pay someone to get it properly into a Release state.
Rupert: What’s been good about this exercise is that it has cross benefits for us as well. As a small builder you buy into this system and the idea is that you would also be able to get a certain amount of advice from us either packaged in or ad-hoc. You can either have a template through the system or you can request that we write things for you.
So it works both ways. It's just not just a product. It's a route to market for us.
Small builders can’t necessarily afford or even need the full service we offer which scales to 8 figure projects. By giving them access to this system we open up an on-demand channel which allows us to really efficiently service projects of all sizes. And that’s ultimately a good thing; small construction projects are going to be safer and more compliant and we are able to address a market which we couldn’t before.
The other thing is the extensive testing of it. Because we could make this so quickly and can make changes to it, we have used this a lot. All of these jobs on here are real but just used by internal teams. We've had teams of administrators working together to try and break it and then give us feedback. But we’re able to make changes incredibly quickly and that means people don’t get frustrated and go back to the old way of doing things.
Alex: That is great to hear because one of the biggest problems with business transformation - to use an industry term - is actually getting systems adopted by the business. There's lots of reasons why people struggle but essentially running it in parallel is the best way because you've got failover. You can take feedback. People don't feel trapped into using something they don’t like and feel listened to.
Rupert: I genuinely don’t think we could have done this without AI. The cost, the effort. It’s suddenly meant that we have the confidence to do things that we couldn’t have done even a year ago. I get that there’s a stage at which you have to “go pro” to get help to put it live and start selling it, but this has meant we can really battle-harden the thing before committing to that next step as a business.
Alex: A lot of the narrative on AI is about the Big 5 Tech companies offering AI upgrades to existing monolith systems and that has turned a lot of people off because it all feels so out of reach and Big Brother-esque. What I am excited about is this exact scenario where businesses everywhere can make things that were financially impossible even 6 months ago. It’s going to mean an explosion of small business, grass roots, sovereign software because we’ve fixed all the issues of cost to create and maintain these products. Provided you know what to do.
Rupert: Provided you know what to do.
Alex: All right. So this year, the challenge is we've got to get this to market.
Rupert: Well, I'd like to get it to market because I think it's quite good.
Alex: Thank you very much. That is brilliant. We will end it there.
