Meta just caught up to the AI frontier, and it exposed a problem I see in almost every business.
AI is ready to do the work. Most businesses aren’t ready to give it the work.
Let me explain.
What makes the new AI models different?
Meta announced Muse Wonder, its most advanced family of AI models yet. After spending billions trying to catch OpenAI, Google, and Anthropic, Meta is claiming frontier-level performance (and people on X are roasting them for bench-maxxing).
But the interesting part isn’t the benchmarks, or Meta allegedly trying to game them. It’s what these models are designed to do.
They have something called Active Memory, meaning they can remember important context across conversations. And Active Focus, meaning they can stay focused on a goal, work through problems, recover from mistakes, and keep going without you constantly telling them what to do next.
How is this supposed to work in the real world? Instead of asking AI to write a report, you give it responsibility for preparing the report. It gathers the information, figures out what’s missing, works through the steps, and comes back when it needs your judgment.
That’s a very different relationship with technology. And Meta isn’t alone. OpenAI, Anthropic, Google, and now Meta are racing toward the same destination: AI that does the work, not just talks about it.
Why does Karpathy say almost nobody gets it?
Andrej Karpathy is one of the most respected voices in AI. He was a founding member of OpenAI, led AI at Tesla, and has spent years working at the frontier of this technology. When Karpathy talks, the AI world listens.
He recently shared an observation I think every business owner needs to understand: the gap between what AI can actually do and what most people think it can do is getting wider.
He broke it into four groups:
- 6 billion people have barely used AI.
- 1–2 billion use it casually, like a better Google.
- 20 million use advanced AI professionally to complete substantial projects.
- Just 5,000 have access to the most advanced systems, where thousands of agents can work together on enormous problems.
These are his rough estimates. But look at the difference. Most people are still asking AI questions. A tiny fraction are giving AI entire projects.
The technology isn’t the dividing line anymore. It’s how people use it.
Which raises an important question: if AI is becoming this capable, why aren’t more businesses seeing the results?
What is “human glue” and why does it break?
After helping more than 400 businesses install AI, I’ve noticed something. Most companies don’t actually have clean processes. They have people who know how to get things done.
Ask how something works, and you’ll hear things like:
- “Kim usually handles that.”
- “It depends on the customer and what they want.”
- “We usually jump on a call and figure it out.”
- “If something goes wrong, send it to me.”
Sound familiar? That’s what I call human glue: people filling in the cracks of messy, undocumented processes.
And humans are AMAZING at this. We remember exceptions. We make judgment calls. We chase down missing information. We figure things out on the fly. We’ve been doing it for so long that we barely notice.
But when the business grows, all that human glue starts to break.
So what do we do? Hire another person. Then another. Then a manager to manage those people. Pretty soon, you’ve got a bigger team, more meetings, more complexity, and somehow you’re STILL the bottleneck.
I spent much of my career growing businesses this way. Until I sold an AI company serving 300 customers with just four employees. That changed how I think about growth.
AI is making this even more obvious. AI isn’t exposing a technology problem. It’s exposing a business problem. You can’t reliably give an AI agent responsibility for a process nobody can explain.
- Who’s responsible?
- What triggers the work?
- What happens next?
- What does a good result look like?
- Which decisions require a human?
If the answer is “Kim just knows,” you’ve found the problem. And probably a pretty big opportunity.
Which jobs can AI replace? (Wrong question)
Here’s the mistake I see entrepreneurs making. They look at their team and ask, “What jobs can AI replace?”
Wrong question.
A job isn’t one thing. It’s a collection of different kinds of work. Some of that work is repetitive. Some requires judgment. Some requires creativity, relationships, and experience. And some of it shouldn’t be happening at all.
Before you give work to AI, you have to understand the work you’re giving it. That means breaking the work apart. I call it unbundling.
Pick one person on your team who’s overloaded. Look at everything they did last week. Don’t look at their job title. Look at the actual work. What could AI handle? What requires human judgment? And what shouldn’t be happening at all?
You might discover you don’t need another employee. You need a better way to get the work done.
What are the four kinds of work in your business?
Once you unbundle the work, you can see very different opportunities:
- Work AI should do. Research, organizing information, preparing reports, following up, moving data between systems.
- Work humans should do. Vision, relationships, creative judgment, important decisions.
- Work humans and AI should do together. AI gathers the evidence, analyzes the options, and makes recommendations. A human makes the call.
- Work nobody should be doing. The meeting nobody needs, the report nobody reads, or the approval step that adds nothing.
Don’t automate that fourth category. Delete it. Because automating crap just gives you more crap, faster.
How does the Find, Filter, Free framework work?
I’ve been using a simple three-step framework to help business owners grow without automatically adding more people.
- Find the constraint holding your business back and map how the work actually gets done.
- Filter the work into what belongs with AI, what stays human, and what should be shared.
- Free your people from repetitive work so they can focus on the decisions, relationships, and creative work that actually grow the business.
That’s it. And it starts with a different question. Not “Who do I need to hire?” Or “What AI tool should I buy?” But:
“What work actually needs to get done, and who (or what) should be doing it?”
I walk through the full Find, Filter, Free framework in Before You Hire Another Employee, Do This, including how we helped a CEO get two full days back every week without adding more staff.
The way forward
The AI models will keep getting smarter. Meta caught up. Someone else will leap ahead. And six months from now, we’ll be arguing about another leaderboard.
But none of that fixes a broken business.
The advantage won’t belong to the company with the smartest AI. It’ll belong to the company that understands its work well enough to give AI the right jobs.
That’s the way forward.
— Joe
Want help putting this to work?
- Binge my best training on using AI to grow revenue without hiring more people on YouTube.
- I’ve installed AI in 400+ businesses. This video breaks down why most people get stuck and how to actually save time and drive revenue.
- Want my team to install AI agents in your business that save you at least 15 hours of your time? Learn more.
