Refactoring Your Skillset for the Intelligence-Native Era
By Wortya
- article
- Synthetic Data
- Artificial Intelligence
- AI
- AI Agents
- Insight Transformation
Summarise with AI
In the first article in this series, Refactoring the Insights Industry, I described the Intelligence-Native Organisation – a company designed so that intelligence, human and artificial, becomes its primary coordination mechanism – and I ended with a simple idea: before we refactor the organisation, we need to refactor ourselves.
The Question I Had to Ask Myself
There is a question I kept coming back to in almost every conversation about AI in our industry: if machines can already analyse, synthesise, simulate and draft, what exactly is left for a researcher to do?
It is not a particularly comfortable question, and I know, because I had to ask it myself. After 25 years building companies at the intersection of human intelligence and technology, I assumed that most of what I had learned would translate naturally into this new world, but it didn’t. Some of what I thought was my unique expertise turned out to be much easier for a machine to reproduce than I wanted to admit.
For years I had been rewarded for being able to make sense of complexity: structure the analysis, connect the dots, synthesise the findings, turn them into a compelling story. Those are still valuable skills for us researchers, but they are no longer enough.
So I had to become a beginner again. I went back to school in my fifties, started building things myself, and I had to get comfortable with something that was surprisingly difficult: not being the person in the room who already knew how things worked.
Four things I would recommend to all of you, based on my own experience of intensive dedication to AI learning, practice, and testing:
1. Build
Don’t just use AI. Understand enough to interrogate it.
Before the current AI wave, researchers researched, strategists designed, technologists built. If you were on the “business” side, you wrote the brief and waited for someone to turn it into something real.
AI has changed that relationship. Today, I can have an idea in the morning and have a rough version of it running that evening. It will probably be terrible, but that’s not the point. The point is that I can now learn by building.
Some of the things I work with today started exactly this way: me, late at night, experimenting with an AI assistant, breaking things, fixing them, and discovering that something I had spent weeks thinking about could be tested in a few hours. That was a surprisingly important shift for me. I stopped thinking of AI as something I needed to understand before I could use it, and started using it as a way to understand what was possible.
That changes your understanding of the technology in a way no webinar can. You discover what it is brilliant at, where it fails, and, just as importantly, you start developing a feel for the difference. Then you stop saying “AI could transform how we deliver insights”, and start saying “Here is the thing I built last week. Here is what it did, and here is where it failed.”
My suggestion as a first move: pick one recurring piece of your work (a report, a coding exercise, a competitive scan, a piece of analysis) and build something that does part of it – not because the industry needs another piece of software, but because you need to understand the machine.
2. Understand
Don’t just use AI. Understand enough to interrogate it.
There is a difference between using AI fluently and understanding what you are using. I think that difference is going to matter enormously, and for me, that meant going back to school and learning how these systems are actually designed and built.
I used to be the person who understood enough about the technology to explain it to everyone else. Suddenly, I was the student again. But there was another reason I wanted to go deeper: I didn’t want to take the promises of this technology on faith. When people started telling me that synthetic respondents could replace human respondents, I had a very specific reaction: I didn’t want another opinion. After 25 years in insights, I knew how easy it is to build a convincing argument around a small amount of evidence, so I wanted to test the claim.
So I started testing them. Thousands of synthetic interviews, with multiple models, different countries, different prompts, different ways of representing people. And some of the results surprised me! The systems are remarkable at some things and surprisingly unreliable at others, and more importantly, the boundary between the two is not random. That was the part that really caught my attention.
It can be studied, tested, and, in many cases, understood. That is where I think a significant part of our professional value is moving, and this is where people in the insights industry have an advantage. Our training has always been about questioning data quality, detecting bias, challenging assumptions, and asking whether what we are measuring is actually what we think we are measuring.
We don’t need to become engineers. We need to bring that discipline to AI. The sceptic’s toolkit isn’t obsolete – it just found a much more important dataset.
The move: go one level deeper than you think you need to. Understand, at least conceptually, how these systems are trained, why they hallucinate, why context matters, why models behave differently, and why the same prompt can produce different answers. Then test something you know extremely well, and trust what survives your own scrutiny.
3. Release
Don’t spend your expertise producing what machines can produce.
This one is harder, because it is not really about technology.
Look at your calendar and be honest about how much of your week is spent moving information around: formatting findings, summarising meetings, cleaning up documents, pulling numbers from different places, building the first version of a deck, rewriting the same content for three different audiences, chasing status. We have spent years calling this work “being thorough”, and sometimes it is, but sometimes we are just being busy.
And there is a difficult truth hiding in that calendar: if a machine can do something in ten minutes that takes you two hours, doing it yourself is no longer proof of value. I don’t mean that we should automate everything. I mean that we should become far more deliberate about what actually deserves our time.
The machine can produce the summary, but it cannot decide which question is worth asking. It can identify the pattern, but it cannot walk into the room and understand why nobody wants to hear what that pattern means. It can give you ten possible interpretations, and you still have to decide which one matters. That is judgement, and that is where I want to spend my time.
The move: take your ten most repetitive tasks and ask a simple question about each one: does this require my judgement, or just my effort? If the honest answer is effort, start handing it over. And when the hours come back to you, resist the temptation to fill them with more production. Spend them on the questions, the connections, and the conversations that actually require you.
4. Show
Don’t claim transformation. Demonstrate it.
There is one more change I would recommend, particularly if you lead others: do all of this in public.
Not because the industry needs more AI influencers. Quite the opposite: there is already plenty of AI theatre out there, with clever prompts, demos that never become products, and confident opinions about technologies the speaker has barely used. I don’t believe credibility will come from any of that. I believe it will come from showing the work: what you built, what didn’t work, the experiment that produced a result you didn’t expect, the assumption you got wrong.
I am trying to do more of this myself, and I have noticed something curious: the most interesting conversations rarely start from the things that worked perfectly. They start from the things that broke.
When you have actually built something, tested it, and put it in the hands of another person, your relationship with the technology changes. You stop having opinions about AI and start having experience with it, and experience is becoming a rare currency in a world where everyone is suddenly an expert.
From Yourself to Your Leadership
These four moves – build, understand, release, show – are not an AI skills checklist. They are a different way of working, and none of them requires becoming technical. They require something harder: the willingness to be a beginner again.
That has probably been the biggest lesson of this journey for me. After decades of experience, the natural instinct is to protect your expertise, but this technology is moving too quickly for that. At some point you have to be able to say “I don’t know how this works yet“, and then go and find out. Build something. Break it. Test it. Change your mind, and do it again. Because the credibility to lead in the intelligence-native era will not come from knowing all the answers; it will come from having done the work yourself.
That is where the next part of this series begins. In the next article, I’ll move from the individual to the leadership team: how to identify the people already experimenting on their own, how to bring an entire leadership group through the same journey, and how scattered individual experiments become a shared organisational capability. Because the transformation still isn’t really about technology. It is about people – one layer further out.
