From Basic AI Use to Building Systems

My name's Andy, and I'm a marketer for, I've been a marketer for about 20 years, mostly B2B marketing. And recently I've been helping senior marketers and their teams implement AI to get real work done. So I very much see myself as a player coach, very hands -on.

I'm going to steal Adrian's tactics of polling. hands up who's currently using AI to do mainly basic work what they would consider basic stuff rather than really super super cool we're gonna go into that we're gonna go into that in more detail but I just want to get a feel for where you think you are all right hands up who is using I like to do some really sophisticated stuff yes and some experts in there okay great right so

So everyone who put their hand up for doing basic work, you shouldn't feel so bad. So Anthropic did a study of 400 ,000 Claude sessions, and they found that only 16 .6 % of those sessions can be classes, technical work. So we've got a lot of people making up that 16 % in today. today.

So yeah, the most common use of Claude is producing an explanation. So a lot of typical stuff is drafting emails, summarizing calls, rewriting text, and answering questions, so glorified search engines. That's where most people are at today.

The Three Levels of AI Capability

So for today's mission with my what probably 13 minutes remaining 13 minutes is to map out the three distinct levels of AI usage honestly help everyone honestly assess where you sit on this journey thinking about where you you put your hand up in the beginning and then reveal the most important skill to building with AI and

it might surprise you okay so the first objective the three levels of AI capability so I've broken them down into three broad categories obviously you can have a lot more and more than that but for today's purposes and I've called them the prompter the architect and the engineer and I'm going to explain what

Level One: The Prompter

each one of them mean in more detail now so level one the prompter so this is where most I'm a marketer so this is where most marketers live this is where they use AI for work which is classed as non -technical and I've created a laptop test so you can think about in practical terms what if you're at this stage so

when you're at work and you close your laptop does anything continue or does does the value die in that session, when that session's over? So if nothing continues, then the work you're doing right there is at prompting level.

So a practical example of doing prompter -level work is, so as a marketer, I do a lot of repurposing content for social media, such as making infographics. So, a typical example of doing that, kind of at a prompting level, would be pasting an article into Claude, asking for a LinkedIn infographic, giving it its brand colors in the chat, having several rounds of iterations and chat, and finally coming up with something which I'm happy with to publish. But when I go on to do it next time, I'm starting from scratch. Right.

Level Two: The Architect

Let's move on to architects. So this is level two. So this is where we're getting a bit smarter. So this is about delegating to AI, so describing outcomes and loading context one time and using features like co -work, projects, skills, and core design.

And the laptop test for this one is, when you close the laptop, does the job keep going? So can you walk away, get a coffee, go to the gym come back and all your work's done done for you as per outcome

um a practical example of that sticking with the infographic um would be um turning the infographic into a skill so a skill is a reusable package of instructions that teaches the ai how to do specific task a repetitive task exactly the way you want it time after time so

for my for my infographic skill I've combined I combine it with Claude design so it knows my branding by formatting how I like to have my text and the output is PNG ready for showing on LinkedIn or substack yeah and here's an an example of an infographic that my skill created.

And for this one, I actually used the content that's been repurposed. No guess would be this presentation here.

Level Three: The Engineer

So there's the three stages. And the third level is the engineer. So this is where you stop doing the work and you build the system that does the work for you.

so at level three you're producing the assets that persist so an example is creating a website that has agents that run it on autopilot and optimize it for you a social content engine that automatically updates your social media scheduler such as buffer is my personal favorite it and with social posts and the drafts. So your buffer is all loaded up with drafts and you can just check it and publish it after tweaking.

So no guessing, I've built those two particular workflows and I'm not a developer, I'm a marketer. So if I can do it, all of of you can do can do as well so uh level three engineering practice so this is the social content

A Social Content Engine in Practice

engine i've built uh using claude code um i've chained nine separate skills together in a workflow it runs automatically on using a routine it identifies what's trending um in the last seven days. It generates six different visual types and not just infographic carousels, cheat sheets, charts, things that look like I've scribbled something on a notepad. Yeah.

It's connected to Buffer via an API. So Buffer released its API this year, a new version. And with its free account, you can post 10 times a week just with a free Buffer account. So that can get you started if you wanted to try something out like this.

And then it's only, I don't know, less than 20 pounds, something like 15 pounds for Unlimited. So yeah, big shout out for Buffer. Don't work for them, by the way.

Right, and yeah, it works while I sleep and I still approve every draft. And shout out to Adrian.

Creating a Feedback Loop with Analytics

So when I was talking through this presentation a couple of days with you, you came up with a good idea about how I could turn this into a loop, a flywheel. So I double -checked this, and Buffer's API also, you can grab the analytics. So how I can make this smarter is I'm going to be able to pull in the analytics of how my social posts are performing, and then it will lean towards what's working for future drafts so yeah thanks Adrian good idea

Assessing Your AI Operating Level

right so let's go back to objective two which was assessing where your AI operating level is so having gone through those three examples prompter architect engineer I just want you to think to yourself where you really are, where you're really at

so when you're working does AI any work while you watch it or does it work while you, does it still work when you walk away and grab coffee or does it work while you sleep so just have a think about that

you may be wondering where where everyone else is with AI.

The Gap Between AI Adoption and Production

So once again, this is from MarTech's Scott Brinker's report. And from that, they say that 90 % of marketing teams use AI agents. So you go, oh, wow, everyone's using AI agents. So everyone's on it already.

But actually, only 23 % of those are in full production. and most of them importantly are vendor bought so they're off the shelf products so anyone with a credit card can buy an ai agent but building one is a capability which apparently

a lot of companies don't have currently in -house right you may also be thinking i'm not i'm a a non -technical person, I can't engineer AI systems, Andy. Well, I want to dispel that myth, if you are thinking that.

Domain Expertise Is the Key Skill

So going back to Anthropix data here. Yes, Anthropix data, yeah. They analyzed 400 ,000 Claude Code sessions. and they took every cloud code session that produced code and asked one straight question did it verifiably succeed so did it pass tests did it commit work and what they found was it wasn't software engineers that came up top they hit 34 % successful sessions but managers of all people came up top with at 37 % successful sessions.

So why is that? Managers are used to explaining things to people of what good looks like. It's literally their job. Checking whether they got what they wanted and they're not shy going back and asking for things exactly the way they want it. So they're good at directing.

So, ta -da, the big reveal is The most important skill when building with AI, the key takeaway from the study was the ability to steer Claude towards success comes more from command of a domain than from the ability to write code.

So do you really know your job well? The nuance, the sort of thing that you just can't get from. claude's not going to know what's going on in your work and how you've been how how how things actually work to the nitty -gritty uh nitty -gritty details that's the real world experience that you have you've built over years doing your job that ai doesn't have so

my favorite actor he gets a hard time sometimes but um he's done some good films uh ben affleck he is wicked smart anyone seen the film goodwill hunting hopefully there's a few

Ben Affleck’s AI Lesson for Filmmaking

right so ben affleck he is secretly a bit of a wizard so he um he is uh he he founded an ai startup a few years ago called interpositive and he recently sold it to netflix for 587 million in this year and what he's done is he's trained he's used his domain expertise in filmmaking so over 20 years of crane films and post -production work and he's trained open source models on jobs that filmmakers actually struggle with in

post -production so this so when he first saw what was going on with AI movies and and AI animation and so forth. He called Matt Damon apparently, I read the story, and he said, we've got to make as many films as possible in the next few years, otherwise we're out of a job.

But then he actually thought about it and he actually realized that AI as it was couldn't do the sort of things that he needed to do for his actual movies. So he went and built the company.

I will add a little joke. How do you like the apples with the 587 million? Right, hopefully there's a few people who got that.

Put Your Expertise to Work Tomorrow

Right, do this one thing tomorrow. So you might not create a 600 million pound startup, but you can make a start.

So if most of your work is more at the basic level, what you can do is pick one job that you really know, like the back of your hand, describe the outcome, brief it like you would to a new hire at work, with all the contacts, everything they need.

And then when it goes sideways, don't quit. So a lot of novices quit when things go sideways with AI. Don't quit. Stick at it. Use your real -world expertise to steer it back on track.

So thank you.

Free Resources to Get Started

I'll just end it as all the skills that I showed today, they're all free. All you need to do is scan this QR code and you'll get my trend radar infographic builder carousel builder the sketchy thing

um and oh yes ai slot linter so it takes out stuff that sounds aie and um a b2b landing page builder there you go all free for you thanks a lot cheers