What happens when you use generative AI tools as interns?

What did Kyle Monson learn from his ‘AI agency intern’ experiment? Find out with the latest episode of the Lexicon podcast
Roland Ellison
Codeword's 'AI Interns' Aiko (left) and Aiden (right)
Codeword's 'AI Interns' Aiko (left) and Aiden (right)

Courtesy of Codeword 

Back in January, Kyle Monson, partner at digital marketing agency Codeword, began a 90-day trial of the use of ‘AI Interns’ as part of his creative team. 

Part publicity-stunt, part serious experiment, Kyle wanted to see how his teams could incorporate the use of tools like ChatGPT, Bard and Midjourney to assist with creative work, learn, experiment, and have some fun. 

In this episode, IE caught up with Kyle to find out what he had learned, and whether he planned to take Aiko and Aiden on full-time. 

Listen to the full podcast here. 

Here are some highlights from our chat, edited from the transcript for typos, brevity and flow. 

Roland: Firstly, could you give us some background on Codeword, the type of work that you do, and the clients that you work with? 

Kyle Monson: We're a mid-sized marketing agency of about 100 people, mostly in the US. Most of our clients are in the tech space. So we spend a lot of time working with tech companies, engineers and developers, helping them turn very technical stories into content, social media, and PR campaigns that people can understand. 

Roland: Can you tell us a little more about the generative AI experiment that you've undertaken, and really how it came about?

Kyle: Back in November, these tools were really taking off. Everybody was talking about them. And there were all these kind of like, deep existential conversations about how they were going to impact every different industry. 

We decided that instead of pushing back and resisting, we were gonna try and embrace it and see how these tools could actually exist in a creative environment like an advertising agency. 

Can we use them? Are they a threat to our business model? Are they a threat to our employee’s careers? Or, are these useful tools that, if we learn how to use them, can be a competitive differentiator for us? 

The team was a little bit apprehensive at first, but we've had a great time with it and I think we've come to realize some things about the tools. We see it as a positive thing, at least for right now.

Roland: Why is it that you think your sector is particularly well suited to running an experiment like this?

Kyle: Yeah, that's a good question. I mean, especially with your audience where you've got engineers solving very complex challenges and probably some particularly sensitive challenges through engineering, code, developing and science. 

We're not doing anything that, we're marketers. So, if we get the wrong answer from Bard or ChatGPT it's like ‘Uh? Okay, we'll rewrite it next time’. I'm being a little bit too flip with it, but there aren’t lives hanging in the balance, based on what we're doing with generative AI tools. There are branded blog posts hanging in the balance, which is a very different thing. 

I think we're well-positioned to have some fun here and test the limits of these tools in a way that some industries probably shouldn't. 

Roland: I know you've received a lot of press coverage about the experiment. I just wondered what the reaction has been to it from the sector generally.

Kyle: It's funny. There's a lot of sort of anguish. I think these tools are really scary for people and force them to confront a lot of questions about their careers in the spaces that they work in, and their own levels of creativity. Which has always been true of technology.

That was true of the Industrial Revolution. That was true of Gutenberg's printing press. These tech breakthroughs have always kind of forced human conversations. 

We were one of the first companies to be like, ‘Hey guys, we're really gonna embrace this and experiment with it and see how it works. Not a lot of companies had done that in any kind of official way…’

So I think it prompted this kind of outburst of both positive responses like ‘good for you’, ‘this is the exact right move’ and also, ‘how could you do this?’ 

We knew when we went into it that we were wading into some choppy waters. But it's been good to see the responses that I care most about, are my internal team, and our clients. They've actually been quite supportive. 

We can either sit back and be behind as other companies and organizations take the lead and learn how to do this stuff, or we can take the lead and learn how to do it and kind of control our own destiny. 

Not to get on my soapbox but I really think for my writers and designers, we've got about 60 writers and designers at Codeword. 

Look, if you're a writer in 2023, your career is already fairly precarious. The opportunities are shrinking by the day. 

A writer who can walk into their next job interview and say, ‘I really know how to use these tools to work more efficiently. I'm really good at writing great generative, AI prompts. I know the landscape and I'm gonna do really good work for you by combining technology and my own human creativity and ingenuity.’ I think that that's a really great career move for them and it's gonna open a lot of doors for writers at a time when writers need every open door, we can get. 

I'm a writer myself, so that's why I feel quite strongly about that. 

Roland: So you had Aiko on the design team and Aiden on the content team. Can you tell us about what you learned from them? 

Kyle: Yeah, we learned a lot. I mean, one of the first things we learned is that these tools are not ready for production environments. They’re just not. 

There are some things that they can do and save us a lot of time, as creatives, and there's a lot of stuff that we really can't rely on them for which, you know, no surprises there. I wish I had something earth-shattering to tell all your listeners but you know, that's sort of where we're at with it right now. 

The design team has been using a mix of Midjourney, Dalle-2 and Stable Diffusion. We've been using a mix of Bard and Chat GPT on the language model side. 

A lot of the stuff that we’re getting from them isn’t quite usable. 

The two things that we have learned positively are this space is changing so fast. The three months we've been doing this experiment, have felt like three years in terms of tech innovation. It's been wild to watch, which is another good reason for doing the experiment. Keeping up with these tools takes real time and intention. 

And the other thing that we learned is that while they can't necessarily do production work, they can get us through lots of creative roadblocks and hurdles. 

You know, as a writer sometimes, the trickiest thing for me is staring at the white page and being like, ‘Alright, I got this thing with words. Where do I start?’ 

Generative chat tools are actually great at helping you just start just kick this thing off. Here are some starter ideas to go with. You're gonna need to do your own research, but the page isn't blank anymore. 

For designers, if we're doing a logo exercise or a rebrand of a tech company, typically that kicks off with quite a long research phase where you're looking at the design elements of their competitors and comparing logos in the space and colors and fonts and creating mood boards. 

Generative design tools can do that for you in seconds, whereas it would take weeks of research to do it. So, it does provide meaningful shortcuts. The nice thing about those shortcuts is that our human teams don't like doing that work. Designers don't want to spend their time doing research, they want to spend their time creating things. 

Roland: Has there been anything that's really surprised you about their output? Like things that you thought they wouldn't be able to do well, that they have, and vice versa?

Kyle: Yeah. I mean two of the most time-sensitive types of creative work on the design side are hand-drawn illustrations, and photoreal. If you're doing photo-real stuff, sometimes you actually have to get people in the room with the camera, and the tools can actually replicate that fairly effectively already, and they're going to get a lot better. 

And then hand-drawn illustration, this is ironic but for brands that want that kind of handcrafted organic feeling, that takes quite a lot of manpower to do and it's quite expensive. The tools can replicate that quite quickly and make it look good enough. You're not gonna hang it in an art gallery, but for a client presentation, it looks great. 

It’s sort of ironic as that's the last refuge for human designers, that handcrafted feel, but the robots are quite good at doing that.

Roland: Did you conduct a performance review?

Kyle:  We did. Yeah. And we're gonna publish our next one next week, I believe. But you know, they’re ambitious, they got big dreams for their careers. 

They've got a lot of dentist appointments on the calendar, which is a little suspicious, so I think they might be interviewing at other places. We'll see.

Roland:  How do you see the future of generative AI tools and how do you think they’re going to change the way your industry works in the next five years?

Kyle: My hope is that we get to a point where creatives always have a tab open in our browser and we know how to use it. 

Even right now, I've got ChatGPT open, I've got my Midjourney discord channel that I turn to from time to time. And so those are just always open, they're just two tabs on my browser that I've always got open and I'm still trying to figure out, how to best use those tabs. 

And I think we're all gonna go through that over the next year. I think this is a near-term shift. 

At the same time. Putting on my like prediction hat, the Internet's gonna fill up with like a 100X-level s***. 

Just like crap everywhere online. 

A flood of content, the likes of which we've never seen, and it's all gonna be low quality and it's gonna be terrible and you're gonna know that it's coming from generative sources. 

And honestly, I don't think it's bad. It might be an okay thing because it's gonna teach us that you're going to need to trust your sources of information. Maybe a little more responsibly. Which I think in the UK and the US has been a real challenge for a lot of people over the past say, six years.

Maybe the flood of s*** content, misinformation and absolute crap on the Internet is going to finally end this kind of ‘age of credulity’, we've been in since Gutenberg invented the printing press. 

We're gonna learn, you know what, you can't trust anything you read, so you better pick your sources pretty carefully.

Roland: Do you think there will be big ethical questions arising for businesses using these tools, like declaring their use on content? Particularly I'm thinking of publishers.

Kyle: I don't know how I feel about that. I think that there's always technological pressure to keep innovating and to adopt new tools. There was a call, maybe two weeks ago that we should have a six-month suspension of any AI research and I'm like, ‘Come on, like who's gonna enforce it?’ 

Even if you do enforce it in the US, that just means that some other tech scene elsewhere in the world is just gonna pick it up and keep moving. 

I do think that there's real pressure to always be pushing forward and always adopt new tools. In the media industry in the tech industry and marketing, I don't think we're gonna slow down. I do think that tech companies are gonna do more to integrate these tools into existing products that we use every day. And that's gonna make this feel safer for people.

There's AI all around us all the time we just don't always know it. Like we're talking on Google Meet right now, which has really great noise cancellation tools that are powered by machine learning. If my dog barks, or if there's construction outside, and birds are chirping, the AI in Google Meet is going to cancel a lot of that stuff out on its own without me ever knowing it, and without there being a big sign, that's like the robots are helping you right now.