71% of people get their AI level wrong

One of the interviews I recorded in Davos was with Stanford AI expert Kian Katanforoosh.

On the drive home, I was already on the phone with my team: get ready, we're changing things.

Here's what happened.

Kian is the CEO of Workera (an AI skills intelligence company), adjunct lecturer at Stanford, and co-creator of the Deep Learning specialization on Coursera, which has been completed by millions of learners.

With his team he recently assessed over 22,000 people on their actual AI proficiency. Turns out, 71% of people seriously misjudge their own AI level. Some wildly overestimate it. Others underestimate.

And it's not random: go-to-market people tend to think they're better than they are. Engineers who are actually really good tend to say they're not.

I'm definitely in the first camp 😂

Kian made a clear distinction: adoption is not the same as proficiency.

Adoption is: I use AI every day. I have Claude and ChatGPT open at all times. I've built my whole writing system around AI.

Proficiency is: How complex are my prompts? Am I using chain-of-thought? Am I building prompt chains where one output feeds into another? Have I set up retrieval systems where AI accesses my actual company documents?

By adoption standards, I'm probably in the top 5% of AI users. I've built a 35-person team where AI is embedded in almost everything we do: we use it to write and edit texts, research guests, build content strategies, and prep for every interview I record. I've even trained AI to sound like me when it writes anything.

But ffter listening to Kian describe how his company actually uses AI, I realized that by proficiency standards I'm maybe at level 2 out of 5.

Kian described how his team runs on Claude. Not just "we use Claude for writing." Here's what they actually do:

They have coded documents they call "skills" — files that contain company's brand guidelines, colour palettes, copywriting rules, hiring processes, the way they speak as a company. Every single one is accessible to their AI.

So when an engineer builds a website, they don't need to call the marketing team to check if the font is right or if the messaging matches brand voice. The engineer just asks Claude.

His assistant? Built their own AI workflow that pulls from Kian's calendar, his past conversations, and his meeting history. Every morning he gets a Slack briefing: here's where you need to be and here's what you need to know. Automatically.

His teams are getting smaller and flatter. They went from 8-engineer squads to 2-engineer squads that own more and move faster. One of their top AI leaders voluntarily moved from management back to individual contributor because he's more productive closer to the tools.

I sat there listening to all of this and thinking: I thought I had it all pretty much figured out. Turns out I'm still at beginner level!

I came back from that conversation and we started rebuilding.

Every direction in my company now has its own Claude project with its own knowledge base. I type "I want to invite this person to my podcast" and I get a full breakdown: should I invite them or not, what views to expect, which topic angle works best.

All backed by actual data from our previous episodes (my team pulled the analytics into Notion and connected it to our Claude).

This is an absurdly good strategist for $20/month, right?*

_{*If you have a lot of data like me, you'll probably need a Max account. I kept hitting limits at first.}

Adopting AI first actually doesn't mean as much as you'd think. The people who are actually ahead are the ones who never stopped learning: five minutes a day reading the right people on X or in the right newsletters. It won't feel like much after a week, but after a year you're in the top 1% on that thing.

Our 35-minute conversation with Kian will help. It's one of the most useful interviews I've done this year because it's practical: Kian gives you a 90-day plan, names the specific skills that are worth the most right now, and explains why 95% of AI agents fail in production (and what the 5% that work actually look like).

Here's the link to the episode:

Kian also highlighted that almost every prediction since ChatGPT launched about jobs disappearing has been wrong. The radiologists and drivers were supposed to be replaced, translators were supposed to become obsolete.

None of it happened on the timeline AI buzz told it would. Even self-driving has been in development for 11 years and it's still not everywhere!

Things take longer than the headlines suggest. But they do happen. Which means the window to get proficient (actually proficient, not just adopted) is still open. Not going to stay open forever though.

— Marina