What Happens to Your AI First Draft After You Hand It Over to Your Team?
There's a certain kind of message I've been getting a lot more often lately. It goes something like: "I've made a start on this. Can you just tidy it up and get it built?"
Attached is a sales page, or a slide deck, or a graphic, made in whichever AI tool the sender has been playing with. And I completely understand how it happens. You had an idea, you wanted to see it straight away, and twenty minutes later there it was. Or at least a very convincing picture of it.
From where you're sitting, you've just given whoever has to finish it a really good head start.
From where I'm sitting, what usually arrives is a PNG. Or a flat PDF. Or an HTML file written for a completely different kind of website than the one you have. There's nothing in it I can open and actually edit, so "tidying it up" means making the whole thing again from scratch, using your picture as a reference.
Which means it's now been made twice, by two different people, and one of them was you.
Before we jump into the post, if you prefer to listen, I’ve covered this topic in more detail on the latest episode of the Connect + Create podcast! Listen below, or add the show to your favorite podcast app to get notified when I release a new episode!
First, where I'm coming from
I'm not going to tell you whether you should be using AI. I have plenty of my own concerns and frustrations with it, I also get real use out of it in a few specific areas of my work, and if you want nothing to do with it at all, I completely understand that too. But it's turning up in my client work either way, so what I can usefully talk about is what happens when the founders I work with start experimenting with it.
And to be clear, these are not people trying to replace their team with a chatbot. They're testing things and working out where it might be useful, if anywhere. Which is pretty much what you'd expect from people who built their businesses by figuring things out themselves.
Why it's so tempting to make a start yourself
Most of the founders I work with grew their businesses a long way on their own before anyone else was involved. Even the ones with a team now keep it small, and that habit of jumping in and tinkering never really goes away. (I'm also that person, so no judgment here.)
So when something comes along that will make you a version of almost anything, of course you're going to have a play. And the reasoning is usually pretty sensible. Maybe you wouldn't need to pay for that app anymore. Maybe it would take less time to explain what you want if you had a draft to point at. Maybe it saves someone a few hours somewhere down the line.
Then there's the waiting. Going from an idea to something real used to take days, weeks or months, and usually other people, some of whom you'd have to hire just for that project. Now you can have a rough version before your coffee gets cold.
That speed is where a lot of the trouble starts. Being able to start straight away means you can start on things you haven't thought all the way through yet. It's always "I'm just testing it," and then the test is sitting there looking suspiciously finished, and it's very tempting to treat it like it is.
You skip the part where you sleep on it and think about how it fits with everything else in the business. And you skip the part where an even better version comes to you in the shower a couple of days later, because by then you're already a fair way down one path, and the shower idea has become a bit of an inconvenience. It's like getting the sunk cost fallacy way too early in the process.
What actually arrives on the other end
When I say what usually arrives is a picture, I mean that pretty literally. Most of these tools give you the finished image of a design, the same as if you'd taken a screenshot of it. It isn't the working file underneath, and you can't just open a picture and change the text on it.
On the occasions I can get something open, or convert it into something that will open, what is actually properly editable is hit and miss. The headline might be real text, or it might be flattened into the background with everything else. And when the draft arrives as code, it's often written for a different platform or theme than the one you're actually on.
So what I've really got is a visual reference. Sometimes that's useful, and sometimes not so much.
To be fair to my clients, nobody is sending me those truly awful AI flyers with about ten billion bits of text crammed onto them. What I get usually looks more or less like the brand. The trouble is that a picture of a page doesn't know anything about where it's going to live.
It doesn't know what your website platform can and can't do.
It doesn't know how your checkout works, or what the page before and after it look like.
It doesn't know which fonts you use, or that you don't allow substitutes.
And it definitely doesn't know about the workaround from three years ago that the whole thing is going to be sitting on top of.
So a lot of the time I'm working out which parts of the draft can exist inside your setup, and then reverse engineering something that will.
Why it can take longer than starting from nothing
More often than not, this takes me longer than if you'd just asked me to create it from scratch.
Without a draft, I'd look at the content and, from what I know of your platform and your brand, work out which layouts are available to me and go from there. With a draft, I now know exactly what you want the finished version to look like. So wherever part of it isn't possible, I have to find the next best version that is. Sometimes that's not particularly close, which means coming up with an alternative that still holds onto the essence of what you gave me.
Then there's how confidently wrong these tools can be, way too much of the time. They'll tell you something has been checked, or that it meets a particular standard, or that it's going to work, in exactly the same tone whether it's true or not. Inside your own area of expertise, you catch it. Outside of it, which is often why you reached for the tool in the first place, you have no real way of knowing.
The first time I saw Claude Code build what looked like a fully working app from a few lines of an idea, it felt like something out of Harry Potter. But when one of these tools spits out a design, I can see straight away where the issues are, the same way you probably can when it spits out copy or marketing strategy in your own field.
Where the time actually goes
When I think about the projects that have gone this way, the time tends to go something like this:
You, prompting it.
You, going back and forth until it looks right.
You, reviewing it and sending it over.
Me, working out what it is and which parts are usable.
Me, making it again, usually effectively from scratch.
Me, building it, testing it and checking what else it affects.
You, reviewing it all again (plus a revision round or two).
Before, you would have told me what you wanted, and I would have made it. Once.
And honestly, in a lot of cases, the time that goes into making a version so you don't have to explain it is more than it would take to just explain it.
When an AI draft really does save time
It's not that it never works. There have been a handful of times a client has sent me a mockup of a sales page or a graphic that was more or less good enough, and I built it exactly as it was. A rough draft can also be a much easier way to show someone what you mean than trying to describe it, which I know some founders really struggle with.
The places where I've seen it genuinely help tend to have two things in common:
First, there's enough existing work and documented guidelines to point the tool at.
And second, the person in charge of the tool knows that work well enough to tell it what to pay attention to, and to know when what comes back is wrong.
A lot of people assume that if you give the tool enough examples, or a thorough brand guide, it can take it from there. In my experience that helps, but it doesn't replace the second part. No amount of examples can tell it why something was made the way it was, or what sequence of decisions led to it. It can guess, but the guess is about as good as a coin flip.
And the oversight has to start before the tool has made anything. If you can tell it that the page is going into Kajabi, give it examples of your typical pages, and explain what the theme and the interface can actually do (and sometimes how the code is structured, and why), you can get a real head start. But you have to know all of that to even think to put it in the prompt. If the well-intentioned people sending me these drafts knew what to ask for, they probably wouldn't have needed to hire me in the first place.
Which is why, in my experience, it's usually the person who already knows how to do the job who ends up saving the time. They know what to tell it at the start, they can spot where it's gone wrong along the way, and they're the best judge of which parts of a project it could actually speed up.
If you're curious, ask the person on your team who would finish the job
A lot of founders are doing the experimenting on behalf of people on their team who'd be much better placed to do it. If you're curious what an AI tool could do for your graphics, your sales pages, or anything creative outside the area you usually sit in, one option is to ask the person who looks after that work to go and find out.
It's not always an easy conversation, so be really clear about why you're asking. For a lot of creative people right now, "can you look into what AI could do for your job" sounds an awful lot like being asked to train your replacement before anyone has said you're being replaced. Tell them what you're hoping to get out of it. (If you are trying to replace them, maybe you need to be honest with yourself and with them about it.) And if someone you trust properly tries it and tells you it won't make much of a difference for that particular thing, trust their judgment.
Can you do it yourself, or do you want to?
I think two questions have gotten a bit tangled up in all of this. One is, "Can I do this myself now?" The other is, "Do I actually want this to be something I do?"
The answer to the first has changed a lot in the last few years. I'm not sure the answer to the second has changed much at all.
Once you can do something yourself, it starts to feel like something you should be doing yourself. But even if it really does only take twenty minutes (and I think we tend to vastly underestimate how long these things take us), that's twenty minutes on something you had already handed off and decided wasn't your job. You handed it to somebody for a reason. Meanwhile, there's a whole list of things in your business that only you can do, all the work that needs your ideas, your vision and your expertise.
If the time these tools give back goes straight into jobs you'd already got off your plate, I'm not convinced you've ended up with any more of it.
Personally, if something saves me twenty minutes, I'd much rather spend it drinking an Aperol and booking a flight to the south of France than making a first draft I'm about to send to somebody who'll have to make the whole thing again anyway.
So, where does that leave you?
I don't have a big position on AI to give you. Mostly I keep seeing how easily founders fall back into work they'd already handed off, and don't always notice, because the way the work happens looks a bit different now. The method feels new and exciting. The end result is the same, though: you're still doing it.
Helping clients work through this is increasingly part of how I work as a creative partner. That means which tools are actually worth it for the business, where they fit, and how to use them without everyone going slightly insane in the process. If you'd like someone inside your business who can help with this alongside everything else, you can learn more at emilybankscreative.com/work.
And if you've made a start on something with AI and then watched it turn into a whole thing, I'd love to hear about it. DM me on Instagram, or leave it at emilybankscreative.com/podcast.
See you in the next one!
Emily Banks
