You're losing nearly a day a week to Slop Fixing.

Time to write: 110 minutes

“Hey AI, write me a LinkedIn post on the new WFH laws in Victoria”

AI: Sure, here is the most average of the average piece I can produce that uses lots of impressive words and at least three uses of corrective antithesis (It’s not X, it’s Y).

“Urgh. That sounds like AI Slop.”

AI: Offensive.

“Okay AI. Make it sound more like me”

AI: Okay, I’ll add in a bit more warmth and professionalism, because you are warm and professional.

“It’s still sounding like AI. Can you at least try to say something insightful?”

AI: Certainly. Here is another draft, with the word “insightful” in it.

“I’m going to close this laptop now.”

AI: Great idea! Would you like me to summarise what we’ve achieved today?

AI promised to bring all of us knowledge workers humungous productivity savings. Yet CFOs around the world are looking at the numbers and feeling pretty grumpy about the lack of return.

One of the more comprehensive studies, involving around 6,000 executives across the US, UK, Germany, and Australia, found that 89% report no measurable impact of AI on their firm’s productivity.

But the reason isn’t the tech. The tech is good!

The reason is how most people are working with the tech.

Last week I wrote about AI Slop and a potential solution. And in case you are new to the term, AI Slop is AI-generated output that looks impressive on the surface, but actually has no substance underneath. And at its worst, it is also just plain wrong (i.e. has factual errors called ‘hallucinations’).

Back in 2023, AI Slop actually didn’t seem too sloppy. Many of us would read something AI-generated and think “Wow, I sound really smart” (of course, forgetting that there is a big difference between sounding smart and saying something smart).

But now, in 2026, if you use AI everyday (as many people do), you are probably pretty good at spotting it.

My team at Inventium.ai are in workshop rooms almost every day with people who have never been taught how to have a productive conversation with AI. So they do what we call “Gunslinging” - they shoot from the hip.

Gunslinging resembles the conversation that opened this post. You give the AI next to no context and expect it to produce a miracle. (It’s just not that good….yet).

And when we gunsling our way through a conversation with AI, we are left with the hidden cost of Slop Fixing.

Slop Fixing involves going back and forth with AI, arguing with it, telling it to “do more of this” and “sound less like that”. Not only does this waste time, but it also wastes tokens (and if you don’t know what tokenomics is and why that matters, check out the How I AI episode where we cover it).

There have been all sorts of estimates around how much time we are spending Slop Fixing, ranging from 4.5 hours per week, through to 6.4 hours per week. And research by Workday found that for every 10 hours we save using AI, we lose around 4 of those hours to fixing its output.

I think we can agree this is not ideal.

So why does almost everyone default to Gunslinging and the inevitable Slop Fixing?

Because they haven’t been on a course with Inventium.ai!

Because arguing with AI feels like progress. Words appear on the screen. Something is happening! Every round of “make it punchier” gives you a fresh draft, which activates the same part of your brain that a ticked-off to-do list does. Progress!

Writing a proper brief before you start feels like homework. You type for five minutes and produce nothing you can show anyone.

So we take the option with the immediate payoff and the delayed cost, over the one with the immediate cost and the delayed payoff. Which is pretty much the mechanism behind every bad decision humans make.

There is an alternative. We call it Architecting.

Think about building a house. If you were to engage an architect, you don’t just say “Design me a house”. Instead, you work with the architect to talk through your requirements and your lifestyle. You have three kids who all need bedrooms, you work from home so you need a study, and you have a dog who needs some grass to wee and pooh on (although my neurotic dog Luna refuses to pooh in the backyard and only defecates on nature strips, but that’s the topic for another post).

And importantly, it’s a discussion, not an order.

One of the things that makes a good architect worth their fee are the questions they ask that you’d never have thought to consider yourself. Where does the afternoon sun hit? Do you want to see the front door from the couch? Where does the exercise bike live once you stop using it?

You didn’t know you had opinions on any of that. But of course you do.

That’s what a lot of people miss with AI. You need to ask it to interrogate you.

What Architecting actually looks like

Let’s take a task that you might give to an AI: Create an onboarding plan for a marketing manager.

If you’re a Gunslinger, you’d probably enter those exact words and then spend the next hour arguing with your AI (or producing AI Slop for your teammates to clean up).

If you’re Architecting your way through, you would:

1. Start with your goal: Create an onboarding plan for a marketing manager.

2. Give the AI some context (e.g. about your organisation, your brand/s, your team, the new person’s role and priorities).

3. Start a conversation with the AI: what’s best practice for onboarding for this kind of a role?

4. Ask the AI to ask you questions to fill in gaps it might have.

5. Get the AI to critique it’s own work.

6. And only when the blueprint is all bedded down and looking good, then you ask the AI to build it.

That list is easy to read but can be hard to do, so here is a template to steal to get things going:

You need to reframe the effort curve

You might be reading this approach and thinking to yourself, “But that sounds so time consuming!”

Yes, it can be.

But a better way to think about it is where you’re distributing your effort.

Gunslinging involves 10% effort upfront in the initial prompt, and 90% of the effort in cleaning up AI’s (and by association, your) mess.

Architecting involves spending 90% of the time upfront in working with AI to set it up for success, and 10% of the time doing a minor clean up to the final output.

And overall, when it comes to quality of output and not spending your days Slop Fixing, Architecting is 100% the way to go.