Marina here.
Job postings that ask for AI skills now pay $18,000 more per year. According to the World Economic Forum, 39% of today's skills can be outdated.
So before I picked the skills I'm going to spend my own time on, I went straight to the people building the AI tools the rest of us use. Amjad Masad at Replit. Mike Krieger at Anthropic. Reid Hoffman, who built LinkedIn. Mustafa Suleyman at Microsoft AI.
7 skills came up across all the conversations. None of them are "learn to code." None of them are "become a prompt engineer." They're more boring than that, and that's exactly why most people are missing them.
1. Problem framing
Knowing what to actually ask AI to do.
Most people fail before they even type — they have a vague problem in their head, paste it into ChatGPT, and get a vague answer back. Frame the problem clearly first, then talk to the model. This is the skill that separates 10x users from people who think AI "doesn't work."
2. Prompting. Not magic words.
And not jailbreaks. The ability to describe what you want clearly enough that a stranger could deliver it. Replit's CEO Amjad Masad told me:
Prompt engineering and prompting is not that different than programming. We just take away the syntax from it. You don't have to understand the underlying details, but you still have to be very precise.
If you can write a good brief for a freelancer, you can write a good prompt. I've been calling people "bad at AI" when really they're bad at delegating.
3. Workflow orchestration
The next gap is between people using AI like a search engine and people setting up small teams of agents that do work while they sleep. Mike Krieger, who runs product at Anthropic and co-founded Instagram, opened our conversation with a line I haven't stopped thinking about:
In the AI age, anyone can become a hundred-million-dollar entrepreneur. Build a company solo.
The solo founders he's seeing aren't hiring teams anymore. They're orchestrating them.
4. Verification
AI hallucinates, lies confidently and even invents sources. The skill nobody trains for is sniffing out when an output is wrong before you ship it. This matters more, not less, as the outputs get more polished.
5. Creative thinking
Reid Hoffman, who co-founded LinkedIn, gave me the cleanest version of why this matters. We were talking about whether AI can replace original thinking, and he said:
Say everyone is going to GPT-4 and say, give me an idea for a lemonade stand business. Well, it'll give them all the same idea.
That's the whole game. AI is incredible at remixing what exists. It's bad at deciding what should exist. Sitting in front of a blank page and choosing what to make is the one move AI can't borrow from you. Most people skip this step their entire career.
6. Repurposing
Taking one piece of work and turning it into 10. A single podcast becomes a newsletter, a Reel, a thread, a YouTube short, a LinkedIn post. AI makes this possible in an afternoon now. The people who treat content as a single output are losing to people who treat it as a system.
7. Continious learning
Mustafa Suleyman, the CEO of Microsoft AI, isn't teaching his kids how to use AI tools. He's teaching them how to learn. Because the tools will keep changing. As he put it:
You just have an expert teacher in your pocket on any subject.
The pattern across all 7: they're not technical. They're about how you think, how you delegate, and how fast you can learn the next thing.
I started spending more time getting better at these points. The output is completely different. Try it this week.
Talk soon,
Marina