Introduction

I wanted to talk a little bit about the future of skills. And I updated this presentation. This is a presentation I did about six months ago, seven months ago. And I'm going to go through three things.

The Importance of Discussing Skills

One, why are we talking about skills? I know this is a meetup about AI. But I think skills are an extremely important thing we should all think about.

Impact of Automation on Jobs

How skills are changing, the skills that are important. And then three, what we can all do to try and stay ahead because a lot of this stuff is getting automated around us. And so we have two options is lean back and wait for everything we're doing to get automated and then we're all out of a job.

Or we can figure out what are the next things that are actually valuable so that we can use these tools to actually produce more awesome stuff and hopefully really flourish from it. So if you're here,

Recent Trends in Job Automation

I hope you have heard a lot about this, but in 2023, IBM put pause on about 7,800 jobs because they expected that all of it was going to get automated. 4,000 jobs were eliminated from CNET at the time. Sorry, in May alone, actually, that was a US jobs report, and then CNET had started fully pushing AI-produced stories, so no staff involvement whatsoever.

On the back of two key reports, the World Economic Forum last year said that about 42% of business tasks operated today will be fully automated by 2027. And McKinsey pushes this to the fact that Today's technology, so without any further evolvement, can automate about 60% to 70% of employees' time today. That is without this technology evolving.

Two weeks ago, three weeks ago, there was this big report from Klarna. I don't know if somebody looked at this. So Klarna had a big decline in valuation. They basically had to raise another round, had to cut a bunch of staff.

Well, the entire job was automated by AI. Now you have 100 people doing the job of 800 before. because customer support is basically fully automated. And they've run proper tests on this.

Speaker Background

So a little bit of context. I said, I am Josh. I'm the CEO at Mindstone, co-founder.

Previously, I built a company called Super Awesome. That became the biggest kids technology company in the world. It was acquired by Epic Games, the creators of Fortnite that you might all be aware of.

I am also a limited partner and a venture partner with Emerge Education, which is the most active early stage venture capital in education in Europe. And I run these meetups, which at this point is the biggest practical AI community in the world.

Looking for Parallels in Skills Development

And so I was trying to figure out if there's anywhere we can look towards that helps us understand what is actually going on. Is there a parallel that we can take that helps us understand what are the skills that actually are relevant tomorrow rather than trying to speculate too much. And I think there is.

Now, I'm biased because I am a software engineer by background. And so somehow everything comes back to software engineering. But I will try and convince you of this particular fact, which is that when I started with software engineering, and it's actually more than 15 years ago now. Shit, it is... It is 25 years ago. OK, I have not actually done that calculation on stage before. This is frightening.

So 25 years ago, 15 years ago in this case, when you were in software engineering, your ability to code was a differentiator on its own. The fact you knew PHP, you knew Java, you knew C++, whatever the thing was that you were expert at, That alone would get you a job, because very few people knew how to build stuff at the time. And you actually had to go down to the code in order to go and build a lot of stuff. And this is changing very rapidly.

Today, Node.js, or NPM in this case, which is one of the more popular execution environments around at the moment. NPM is basically a package management system, which is the equivalent of Lego blocks for engineers. So you can take a whole bunch of code that's been written by other people, put it together, and then figure out how you build software.

1Today, when you're a software engineer, your job is entirely different. It is mostly taking those Lego blocks and figuring out how data goes into one and figure out what spits out from the first one, put it in, maybe transform it a little bit, put it into a second and a third and a fourth. And often you're looking at thousands, if not tens of thousands or hundreds of thousands of these Lego blocks that altogether end up building a solution to the problem that you're trying to solve. What I'm saying here is that the job of a software engineer is much less today the actual ability to code. It's still somewhat there. But it is much more putting these LEGO bricks together into a solution to the problem that you are trying to solve in the first place. And so the differentiator for a good software engineer is your ability to solve a problem. And I think that is really important.

AI's Role in Task Automation

Because Sam Altman, earlier in the year, said something very similar to this, which is that these systems, and when he talks about these systems, he's talking about large language models in this case, are much better at doing tasks than jobs. give them an entire job, they're not quite there. They can't quite do that.

And there's a whole discussion about, will they ever get there from a large language model perspective? Will we need other kind of technological changes before an AI can actually get there? But what is clear is that right now, these larger language models are getting better at very specific tasks. And they are giving people tools to do their work faster and at better quality.

The Evolution of Software Engineering

This is a non-exhaustive map of the amount of AI tools that have been built in the last few years. And what I'm arguing here is that what is happening with large language models and with AI is similar to what has happened in software engineering before, which is that we are building a whole bunch of Lego blocks that we now, in different industries, have to start putting together into solutions to the problems that we're trying to solve.

Just like a software engineer has gone from the ability to code with a differentiator to the ability to put the Lego blocks together and create or build a solution to a problem.

So the same is gonna happen in a whole bunch of different industries where maybe your ability to write a very fine-tuned contract used to be your differentiator or your ability to really understand someone's problem from a sales perspective in order to position your product in the right way was your specific skill or maybe your skill was to adapt your style to the individual that was in front of you. All of those are very specific skills all of those one by one are going to get automated in different ways.

And our job is going to be to take all of those different building blocks, putting them together, and figuring out what is the problem that I have, what is the right set of LEGO blocks that I can use in order to build a solution to that particular problem. Now, this is happening today.

Case Studies in Automation Tools

So how many of you are familiar with DocuSign Analyzer? No, not that many, but few. Few, OK, good.

Well, so basically, it analyzes the contract for you and gives you some high-level detail. Right now, I actually use ChatGPT for contract analysis. And maybe I should, maybe I shouldn't, but I definitely send much less stuff to my lawyer today than I did a year ago.

How many of you have seen Devon? Yes. Well, that has really spread.

I was talking about how the ability to code is no longer a differentiator. That, Devon, is getting really, really far to making that almost obsolete. And when I say obsolete, obviously there's always going to be a actual use case where people have to go very much to the detail. But I'm saying that for the majority of people, the ability to write code is no longer going to be a differentiator.

And Zoom Virtual Agent does what it says. Basically, it's an enhancement for Zoom Virtual Agent specifically for call centers. And it's an enhancement for call center operatives. So basically, when you're sitting in a call center and you have to answer a bunch of questions, it listens to the call. It starts to surface potential answers to whatever questions coming up before you, as an operative, have even queried your own database. So it's basically starting to take all of that task away from you as well.

Now, you don't just have to trust me on this.

Insights from a CTO Perspective

I am also part of one of the biggest CTO groups in the world. I was the Chief Technology Officer at Super Awesome when we were building that out.

So I reached out to a group of a few 100 CTOs and I asked them, what are the key skills that you are looking for when you're hiring software engineers today? Because if the thesis holds that software engineering has been living in the future in a way, basically they've been using building blocks to go and build solutions to problems, If we're going to get there, then there's a good analogy where what software or what CTOs are looking for today in software engineers might be similar to what everyone else is going to be looking for across different industries very soon.

And what they came back with was the ability to collaborate, the ability to learn, and the ability to solve problems. Now, you will notice that the ability to code is not part of this top three, and it was on the list of options that was offered to CTOs that were answering on this. It doesn't come in the top three. It even didn't come in the top five, by the way. Yeah, I think it was number six on the list.

And for those of you that are still not clear about if coding is going to disappear or not, 40% of code on GitHub in May last year, of all new code committed to GitHub, 40% of it was entirely AI generated without having been touched. So that gives you an idea of where we are.

Now, on top of that, Google and Pearson got together to do a study on the future of skills. And they tried to figure out what might be relevant in, I think their horizon was 2030. No, 2025. And the ability to collaborate, the ability to learn, and the ability to solve a problem, achievement focus here, came in the five skills that they came up with as well. So I think there's a decent amount of evidence and logic as to why this would be the case.

Building Future Skills

What can we do to try and build them?

How many of you have read a book called Range? Yes. OK. Really, really good book.

How many of you have read a book called Outliers? There we go. OK. So interesting room.

Range and outliers are on two spectrums, but basically the thesis of range is that there are two different types of skills that we can learn. Skills that live in a kind environment and skills that live in a wicked environment.

A kind environment is basically a skill that you can build where the feedback loop is extremely fast. So whenever you try something, It either goes right or it goes wrong. But you instantly get feedback on if you got it right or wrong. And if you got it right or wrong, you have a fairly easy way in understanding what went right and what went wrong.

A wicked environment is the entire opposite, which is that you never quite know if you got something right or wrong. And even when you know that you got it right or wrong, you kind of don't really know why you got it right or why you got it wrong.

So can anyone give me an example of a kind environment? Sorry? Chess could be a good kind of environment indeed. And so the reason for chess is that basically you make a move. Well, sometimes it takes you a few moves to realize that you were really wrong with this one. But basically, you can analyze it very easily. And you can understand that that move was bad. And there's a whole bunch of rules around why it was. And you have a direct feedback mechanism.

The typical example of a kind of environment is golf. You hit a ball, it goes left, it goes right, you instantly understand, okay, I did something wrong, and you instantly get to try again, and again, and again, and just until you get it right.

What would be an example of a good wicket environment? Yes. That is the example that gets taken a lot. The reason that venture capital is such a wicked environment is exactly for the rules I outlined, which is you really don't understand, or it takes a very long time to understand if you're right or wrong. Often 10 years or so from the moment you invest to the moment you might see a return. And when you see the return, you still don't know if it was just because you were entirely lucky, you were at the right time at the right place, or because you actually knew something that other people didn't. And you don't have a way of figuring that out, which makes it extremely hard to build a skill of venture capitalist.

Now, what I find interesting, so programming is a kind environment skill. Sorry, I was going to map that, actually. So yeah, programming is one of the examples where you make a mistake. At this point, you actually get extremely detailed feedback on what you got wrong. And now you can even copy-paste that, whatever error you got out, put that back into ChatGPT and try again. And you have a very interesting loop coming back.

Collaboration, problem solving, and learning, which are the three skills that are going to be key in the future, are in the wicked environment problem. If I get good results around collaborating, I don't know if that's because of me or because my team is much better than usual, or if I solve problems more often than others. Actually, solving problems is maybe a little bit more in the middle, but the learning bit is often a very long timeline. You can think you're doing all the right things and realize 10 years on, oh shit, I was wrong on all the, at least that happens a lot to me. Things that I thought that I learned 10 years ago, it's like, yes, I've got them, and actually it's like, shit, I've been wrong all this time. Happens over and over. A very wicked environment because you're constantly reassessing and you constantly get much more data, but it's hard to get that feedback loop.

1Now, this is where I think AI can help a lot because that feedback loop is about to get shorter for a lot of skills that live in this wicked environment. So from a venture capital perspective, a very simple prompt where you can now say, imagine you're the world's best startup investor and you are, I have got this investment thesis that I'm thinking about. give me one, the ideas of where my reasoning might be wrong or right, but also give me actual examples of previous investments, previous history or parts of history that have followed a similar trajectory. This will allow me, not in a perfect way, obviously this is not an actual feedback loop, I'm not directly seeing if my investment was right or not, but it gets me a perspective that I would never have had before, which allows you to have a simulated, a virtual feedback loop for so many different skills.

Learning is another environment where Everything I do now I can put into ChatGPT. I can get a first, a second, a third look that I normally would never have gone out to because very few people have one-on-one tutoring. And so you're able to get to a feedback loop that allows me to hone whatever skill I'm building in a way that makes a wicked environment look more like a kind environment, which means we should all be in a position now to build these wicked environment skills in a way that we were never able to do before, which is where all the value is going to lie.

A Tipping Point in Skill Assessment

Now, we combine that with what I think is a once in a life tipping point. Because at the same time of these skills being different, we are starting to lose confidence in how we are assessing and signaling the value in the skills that have been relevant to date.

The world around us is moving faster and faster and the type of skills that we need keep changing at a pace that is faster today than it was five years ago. The average lifetime all right the average time any particular skill stays relevant today is about five years that's down from eight years uh was eight years 20 years ago so going down soon it's going to drop under the time it takes to get a degree in the first place which means that by definition the moment you get into university by the time you're out whatever you learned is on average already obsolete now

Employers' Shift to Skill-Based Hiring

The world of work is realizing this, which is why 76% or so of employers are already using skill-based hiring. They're starting to move away from degrees and you're starting to see a bigger move towards micro-credentials, which is the idea that, okay, we just need to know if you have this one particular skill so that you can do the job today.

But I would argue that That is trying to take an existing type of solution, which is degrees and certificates, and putting it into a world that is now fundamentally different. The difference being that the entire system that we built was built for a world of information scarcity.

The best way to try and pass on the knowledge, figure out how to educate everyone else was to find the one person that had the knowledge or could read when you go back hundreds of years, and then spread that knowledge to 100, 200, 300, as many people as possible in the same, in one go. But that's why you have a lecture format. One person standing in front of a maximum amount of people to distribute the knowledge. Because information scarcity was our problem.

It was hard to get to that information. We now have the opposite problem. We are overloaded with information. And actually, the skill is to try and figure out what information do I pay attention to? And there are entirely different models.

New Models of Skill Evaluation

So again, if we move towards engineering and we try and figure out how does a CTO hire today, a great GitHub profile or a great Stack Overflow profile will get you more interviews than a degree from MIT today. So how many of you are, well, I guess most people know the name GitHub. No? Yeah, quite a few. How many of you are familiar with Stack Overflow? OK, well, we've got a fairly engineering-heavy room, which is good.

Basically, the entire Stack Overflow model is based on great questions and great answers. And the idea here is that there are different ways, micro signals, that we can bundle together into a profile, which then allows us to evaluate if someone is competent at a particular skill or if they are not. And actually, at Mindstone, we are working at something exactly like that, which is can we go and find these micro signals of individuals across non-engineering disciplines, so consulting, customer support, human resources, lawyers, whatever you are. And can you take those micro signals, transform them into a profile so that that profile becomes or is a clear representation of the skills that you have, which employers can then hire from.

Now, this is a live or die situation for companies today.

42% of business tasks will be automated by 2027, and 60% to 70% of skills or time spent of employees today is already fully automatable with the technology we have here, which means that the ability for companies to shift the way they hire, to figure out how their employees stay relevant rather than getting rid of them in different ways is going to be extremely important as we go to the next step.

The Future of Skills and Lessons from Engineering

And as paraphrasing Gibson's, the future of skills is already here. It is just unevenly distributed.

Engineers have been living in the future all this time, and we can learn a lot from it. So I hope that was useful.