10-80-10 - Working with AI

Most marketers are either terrified of AI or completely deferring to it.

The terrified ones are waiting it out, hoping the pressure to adopt passes and they'll get away without it. The deferring ones are hitting generate, skimming the output for about twelve seconds, and posting it. Neither approach is correct.

I've been doing something different. A framework I've landed on after eighteen months of working out, through trial and error, where AI actually makes me more efficient, and give a good result. I call it the 10-80-10 rule.

It's not something made up myself, it was actually stolen from Steve Jobs. He explains the rule this way:

• Spend the first 10 percent of the time communicating your vision for the thing.

• Allow others to spend the next 80 percent of the time moving the thing forward.

• Spend another 10 percent of the time polishing the thing, and helping others understand why and how you’re tweaking.

I’ve adapted it for how I’m using AI to support my day to day work. To be honest, sometimes, it 10-70,20 or 15,50,35 – it really depends on the task or approach, but it’s a guide that’s working for me

What the 10-80-10 rule is

The idea is simple.

The first 10% is human setup. The 80% in the middle is AI execution. The final 10% is human judgment and edit.

The reason I use it as a guiding principle, isn't the percentages, it's what happens in each stage, and more importantly, what goes wrong when you skip one.

The first 10%: the brief

This is the part most people skip or do half-heartedly. It's also the most important part.

Before I ask AI to do anything substantive e.g. write a piece, structure a strategy document, draft a brief, research a topic, synthesise a report, I spend time on the setup. That means working out:

• Who is this actually for? Not "marketers", ”what kind of marketer, at what stage, with what specific problem?”What do I want them to think, feel, or do after reading this?

• What's the one angle or observation that makes this worth writing at all?

• What does good look like? Are there examples I can point to?

• What are the guardrails? What shouldn't be in here?

Sounds obvious but In practice, most people skip all of it. They type "write a blog post about AI for marketers" and then wonder why the output is generic.

Garbage in, garbage couldnt be more true when using AI. It's exactly what happens when you take a shortcut at this stage.

The first 10% is the hardest part of the whole process because it requires you to have done the thinking before you start. You have to know what you want to say before you can brief someone else to help you say it. That's true whether "someone else" is a junior copywriter or an AI.

The upside is that this discipline makes you a better marketer regardless of AI. If you can write a clear, specific brief, you're ahead of most people in the industry and you'll produce better output from every tool and every person you work with.

The 80%: where AI earns its keep

Once the setup is right, this is where AI does the heavy lifting.

In a typical week on my team, the 80% covers a lot of ground.

First drafts. I give a brief, get a structured draft back, and work from there. The draft is never finished, but it's a starting point that would have taken two hours to reach manually. I get it in ten minutes. That's a genuine efficiency when you're running a lean team.

Research and synthesis. Pulling together what's known about a topic, summarising competitive positioning, identifying patterns across a dataset. AI handles this faster and more completely than I can do manually. I verify anything going out publicly, but the synthesis is usually close.

Structure. Sometimes I know what I want to say but I'm not sure of the order. I'll write out rough bullet points and ask for a structural recommendation. Useful about 80% of the time.

Variations. Headlines, subject lines, ad copy options. Give me five versions of this. That used to take a copywriter and a chunk of time. Now it takes thirty seconds, and I pick the best one or mix elements from a few.

Editing for clarity. I'll write something and ask "where does this lose momentum?" or "what's the weakest sentence in this paragraph?" It's a useful second read when you're too close to your own work.

What it doesn't do well: anything that requires actual judgment, lived experience, or a genuine point of view. It can write around an opinion, but it can't hold one. It can describe a situation but it can't have been in one. That's where the final 10% comes in.

The final 10%: the edit

This is where the real work is - but now I'm feeling fresh because I haven't had to do the grunt work.

I've had people on the team show me AI-drafted content and ask if it's ready to go. The answer is almost always no, not because the AI did a bad job, but because nobody's done the final 10% yet.

The edit is where you:

- Replace the over-formal phrasing with how you'd actually say it out loud

- Cut the sections that are technically correct but add nothing

- Add the specific detail or example that makes it real

- Fix the rhythm. AI writes in a predictable cadence that reads fine but sounds slightly flat when you read it back

- Put in your opinion or experience that makes it yours

That last point matters the most. AI output is averaged. It draws from an enormous range of sources and produces the most likely version of whatever you asked for. The most likely version is almost never the most interesting version.

Your experience is not averaged. The specific observation you have about why something works, drawn from the situation you were actually in, is the thing that makes a piece worth reading. That can't come from the AI. It has to come from you, and it gets formed in the edit.

Without the final 10%, you're publishing someone else's average. But after the edit, it’s writing that has a point of view - which is truly yours.

What this looks like in practice

On a lean team, the 10-80-10 framework changes how you plan your week.

Instead of "we need to produce five pieces of content," the question becomes "how many briefs can we properly write?" Because the number of quality briefs you can write sets the ceiling for what AI can produce for you. The 80% is essentially unlimited. The first 10% is the constraint.

In practice, for a team of two or three people, I'd expect to properly brief four to six pieces per week and edit them down to the three or four that are genuinely good. That's a realistic output for a small team running this workflow, and it's substantially more than most teams produce without it.

The other thing it changes is what you look for when you hire.

I'm less interested in people who can produce volume. I'm looking for people who can think clearly about what we're trying to say and why, write a sharp brief, and edit output with taste and judgment. Those are the skills that increase our team’s velocity. They're also the skills that AI doesn't have, which means they're the skills that matter more now, not less. If AI raises the floor for marketing output, the ceiling is set by human judgment. So now we hire for judgment.

The hidden trap

The most common mistake is treating the 80% as the finished product.

It's fast. It's coherent. It looks like it's done and Damn it did that in 2 mins? So people publish it. And then they wonder why it's not landing, why nobody's sharing it, responding to it, or reaching out about it.

It's not landing because nobody did the final 10%.

The AI does the heavy lifting. You do the work that makes it human.

One thing to try this week

Before you use AI for any task, write out the brief first. It takes 5 minutes to answer these six questions:

1. Who is this for, specifically?

2. What do I want them to think, feel, or do?

3. What's the angle that makes this worth reading?

4. What does good look like?

5. What are the guardrails - what shouldn't be in this?

6. What experience or observation can I add in the edit that the AI can't have?

Answer those before you open the chat window. The output will be better. The edit will be faster. And what you produce will actually sound like it came from you because the part that matters most did.

---

Note: Setting up your tools and telling LLMs like chatGPT and Claude how you write, and you background is an important step that will further speed up your workflows. But that's a whole other post - keep your eyes peeled for it soon!