AI: removing jobs or enabling change?

In the past, technologies have taken five or even 10 years to become ubiquitous. That timeframe is now much shorter.

Today, organisations know that you bought an item, and because of that item, maybe you're interested in another item. It's predicting your personal preferences. We see this every day and we don't think about it anymore, because it's commonplace. A similar timeframe of acceptance also applies to generative agents, because the economics behind it are so great, as I’ll explain, but it’s potentially very bad news for the current roles of humans within organisations such as supply chains. 

"I don't think it's going to be an evolution. I really do mean that it’s going to be a revolution"

The changing reality of entry-level jobs

If you could look into the future, you could say that by 2028 possibly 40% of business activities will be done by a generative agent. A generative agent works 24/7 and doesn't lobby for a salary increase; they don't get sick, they don't get healthcare benefits, and they don't get paid vacations.

In addition, humans can make errors. A generative agent can also make errors, but the errors of a generative agent will be consistent. You can go back into the previous decisions and correct them, if necessary, while with humans, you don't know how many times that error occurred in their activities or those of their associates. 

Because of this, many entry-level jobs in procurement and supply chain management currently handled by humans are soon going to go away.

What this means is that the new challenge is going to be how we develop humans to work within the supply chain collaboratively with generative agents. Generative agents are going to take most of the process-driven jobs that entry-level people would normally take. Typically, when you leave university, you leave with certain skills which prepare you for an entry-level job, usually, by performing some well-defined process that is much more mechanical than it is creative. Those won’t exist in the future.

The question is how are young entrants into the supply chain function going to get experience when we’ve implemented these tools much more widely? The answer is that education and training are going to change, and all forms of training, whether at a professional or university level, will be radically restructured.

Emphasis will be on data and organisational knowledge identification, capture, retention, classifying and reuse.

Data and agent management: the new entry-level skills

The question of how you manage an agent becomes a much more interesting paradigm and shows that the real value young people are going to bring is that they can learn about data quality and data governance. Those are the things that feed generative agents and make them relevant and trustworthy. If we don't trust our data, we're not going to trust the output of these generative agents, and we’re not going to take advantage of them.  

This is why the first question I always ask organisations is, ‘Who is in charge of your data?’ and usually, everybody looks at each other. They then think for a bit and ultimately say, ‘Information technology’. 

Well, Information Technology is not in charge of your data. They collect it, they store it, and they put safeguards around it as to who can have access to it, but that’s it. They’re not saying it’s relevant, accurate, or even complete. If you ask that question of any supply chain organisation, I'm willing to bet that 80% of the time, if not more, nobody is in charge of their data, and yet they use it to make decisions. 

It’s often a summary of another summary of some data that's years old. It was used to make a different decision, and now they're trying to use it to make a completely unrelated one. It's probably not even relevant, but it's the best they have. The lineage of data is very important, and more people need to ask: where did it come from? 

The answers to that can be alarming. Studies will show that most organisations have the same basic data stored in at least 10 to 12 different places and often have variations of that data because it is used for different purposes. The problem is that if it's wrong, you don't know where to go to correct it, because it's on people's laptops, it's in a file, or it's on a server somewhere. 

We're more out of control than people want to realise or want to admit, yet we feel good because we’ve got tons and tons of data. But is it worth it? And that’s what companies are starting to think about. Data is the building block of generative AI, and the technology is going to get better, but data has not progressed at the same rate.

Moving on to the end stage

The reason why it matters is that we want to move on to the third stage of this. In the first, we said, ‘We’ve got loads of data, isn’t it fantastic?’ In the second, we said, ‘We have to interrogate this data so it is meaningful,’ and in the third, we will be using it to make decisions on our behalf and to address risks far in advance.

Imagine a continuum. Data goes to information, information goes to knowledge, and knowledge goes to wisdom. Before, you'd have a spreadsheet, and it would give you some data, and then a human would have to interpret that spreadsheet and say, ‘OK, given what I know in the environment and that business, I can make a decision.’

Then we started summarising the data by graphically depicting it. That was good. We started to understand massive amounts of data through visualisation, but that was a transitory technology. It helped us see patterns. But that's not the end state. We don't want to just visualise things. What we want to do is use the data to help us act.

A human can't hope to monitor these variables in real time. While supply chain folks typically do a very good job at putting due diligence in place and writing the contracts, there are also people to consider, logistics, regulatory issues, changing technology, variable currencies, and so on.

AI as a decision-maker

Some data doesn’t need to be perfect to make tactical decisions. You can train these agents on partially incomplete data to make basic decisions. There is not a lot of risk, and little cost. And you’re not going to lose business if you make a bad decision.

If you’re going to build a new facility somewhere, or you're going to expand and you’re buying new equipment for millions, that is a more critical decision. Your data needs a different level of credibility, completeness, accuracy, and relevance. 

This will start to shift the agents toward making lower-level decisions, which is why entry-level jobs will change. Soon, these agents will do everything. They’re collecting massive amounts of data, but they'll be able to interpret it, to make recommendations and take action. 

When you think about agents right now, we're still talking about singular ones. We’re already moving to agent teams. We have an agent who knows accounts payable, an agent who knows quality inspection, and an agent who knows logistics. And right now, we're working on languages that let these independent agents communicate with each other as a team. 

The question will be how much authority do we want to give agents? As we get to the end state, and we have absolute confidence in an agent to monitor our contract terms and a supplier starts to fail, we can have the agent notify the supplier and provide written legal notice. There is already legislation in the United States that recognises an agent’s ability to act on behalf of a human third party and cancel a contract.

It’s impossible for humans to manage this much variability, and it changes the paradigm completely. Most people advance through this profession based on their ability to do firefighting. They got where they are because a crisis happened and they solved the problem. What you don't see are the many people advancing their careers based on their ability to prevent problems from happening. That's where we want to be. 

Yet for many people, access to data or control over it can be their power base. That’s what makes them important. Now we’re trying to make information more widely available, and these systems can make decisions, there's a big resistance at middle-management level to embrace this because it’s threatening to them as individuals. 

If someone were to create a tool for supply chain managers who are not well-versed in financial analysis, an analyst might think of building a hard-coded system that takes and analyses financial reports and other data.

This expert might start to question whether he would be needed in the future, or if he is building something that would put him out of work? As a result, he might start to become protective of sharing his knowledge. This is why it has been known in some businesses that a financial reward, or benefit, would be put in place, to drive the expert to make the best tool possible and keep upgrading it. For example, the bigger the bonus if more people were using the tool.

Conclusion: ‘lights out’ by 2028?

I'm not sure if we'll ever replicate the human mind. When you think of a child growing up, they may see a dog 100 times in their first year, and then after that, an adult may try to signal that the child has seen a dog whenever one appears. We must take millions of images to train these generative agent systems to recognise a dog. Advances in programming and processing speeds are getting better, and use less time, but it’s still difficult.

That’s not what we’re trying to do. We’re creating tools that go from data to information to knowledge, and more advanced systems are achieving a level considered to be wisdom (wisdom is developed by integrating situational awareness). These tools can take in data on weather, data on political disruption, or labour issues. They can give you their financial analysis of the supplier, their history, their quality, and how they compare to other suppliers.

They can even pull in unstructured data such as information from emails, videos and news reports, things that aren’t necessarily recorded anywhere in our traditional systems. This means our profession is going to face an interesting challenge. As we move forward, jobs, education and training will be different. They’re not going to be about the skills needed to complete basic, mundane tasks. They’re not even going to be about managing people. They’ll be about managing and orchestrating teams of generative agents. This is the right place and the right time for generative agents to have a major impact on the profession.

If generative agents are adopted, we could have a supply chain function operating in a ‘lights out’ way - no humans - by 2028.

"This is the right place and the right time for generative agents to have a major impact on the profession"