The Future Belongs to Systems Thinkers in the AI Era

Your lane just became five lanes

A few years ago, building a product followed a clear relay. The product manager wrote the spec, the designer designed it, the developer built it, the data team analysed it, and marketing sold it. Everyone stayed in their lane.

AI is blurring those lanes. A product manager can build a working prototype before an engineer touches it. A designer can generate starter code. An engineer can produce decent UI, write requirements and analyse user data. Someone in finance can interrogate a dataset without waiting days for an analyst. That does not mean everyone becomes equally good at everything. But the cost of crossing functions has collapsed.

Netflix's product and technology chief recently described exactly this: people inside Netflix can now move much further through product development before handing over to another function, and the company is deliberately hiring more people who can think across systems rather than within a narrow domain. Meanwhile, most education still prepares people as if the walls between departments never moved.

What systems thinking actually means

It sounds like a fluffy LinkedIn phrase, but it is simple. Normal thinking asks: what task have I been given? Systems thinking asks: what larger outcome is this task part of?

Say you are an engineer asked to build a login page. A task thinker works out how to build a login page and builds it. A systems thinker asks why users are logging in, what information is stored, where identity lives, what happens when someone forgets their password, how this affects security, what happens on refresh, and whether every product in the company needs authentication, in which case it should be solved once instead of rebuilt by every team. Same task, completely different level of thinking.

Or take marketing. Someone asks you for a thousand leads. You could simply become brilliant at Facebook ads. A systems thinker asks where the leads go, what the close rate is, who actually buys, what the customer acquisition cost and lifetime value are, whether sales follows up fast enough, and whether these are even the right leads. Now marketing is not ads. It is the whole acquisition system, from first hearing about you to trusting you to paying you.

Why stronger AI makes this more valuable, not less

You might assume that as AI improves, humans need to understand less. I think the opposite happens. Imagine I give you 20 incredibly fast employees who can write code, analyse spreadsheets, build landing pages, research competitors, write emails and design graphics almost instantly.

You still have to tell them what they are trying to achieve. You have to stop them contradicting each other, stop them adding too much, give them shared guidelines, decide what information each can access, set quality standards and spot when something they produce is wrong. You need someone looking at the whole machine. That is exactly the situation with AI agents. They massively increase output, and output without coordination creates more problems, not fewer.

Netflix made this point specifically about agents: once many agents work across systems, companies need shared infrastructure, reliable data, security rules, common design standards and people who can reason across all of it. The bottleneck moves from "can you build this?" to "do you even know what should be built, and can you make all these moving parts work coherently for users?" That is a much harder problem. If agents are new to you, start with AI agents explained like you're 10.

This is not an argument for being a generalist

It would be easy to hear all this and conclude you should not go deep on anything. That is probably worse. AI makes mediocre generalists extremely easy to create. A bit of marketing, a bit of design, a bit of code, and ChatGPT for the rest is no longer rare. Being a generalist is now table stakes.

What is still rare is depth. Great engineers, designers, salespeople and builders are still rare. Netflix drew the same distinction: specialist knowledge has not stopped mattering, but the person who understands one narrow tool and refuses to look beyond it becomes much less useful.

So the generalist versus specialist debate is the wrong argument. The winning profile is T-shaped: depth somewhere, breadth around it. Your craft becomes your home base, not your prison. An engineer who understands product, data, users and business models. A marketer who understands statistics, pricing, retention and unit economics. A designer who understands what technology makes possible and why the business is building what it builds.

Stop naming yourself after your tools

Many young people identify with their tools: React developer, Python developer, Facebook ads specialist, SEO writer. But tools change ridiculously fast. LinkedIn estimates 70% of the skills used in most jobs could change by 2030. If your identity is attached to one tool, you are building a house on moving ground.

Aim to be excellent at a durable category of problem instead. Not "I build websites with React" but "I know how to turn ideas into software people can use". React is one tool for that. AI coding agents are another. Whatever comes next will be another. Not "I run Facebook ads" but "I know how to acquire customers profitably". Facebook could disappear tomorrow and you would still understand the problem.

This is the part people miss about becoming AI-proof. You do not need a magical skill AI can never touch. I do not think one exists. You need to work one abstraction layer above the tool. We explore which work holds up in jobs AI can't replace.

How to practise systems thinking

The simplest method: whenever you solve something, zoom out one level. Just one. Not ten levels until you are philosophising about capitalism.

What this means for learning to code

I run an AI-native coding school, so this hits close to home. If all we taught was syntax, we would have shut down two years ago. AI generates syntax faster than almost any engineer. But that does not make software engineering useless. It changes what we teach.

Can you understand the code that was generated? Spot what is wrong? Debug it? Do you understand the database, frontend, backend, authentication, security and how information moves through the system? Can you explain why you built it one way and not another? When I train students or hire engineers, I no longer care whether someone memorised how React hooks work. Documentation existed before AI. I care whether you can look at a broken product and work out what is actually happening.

There is a huge difference between someone who can ask AI to build an app and someone who understands the app AI just built. AI may well generate most of the world's code, but humans still have to understand the system, review the work, test it, diagnose it and take responsibility when it breaks. AI does not take accountability. That is the craft, and craft plus systems thinking is very hard to replace. It is the exact combination our AI-Native Software Development Programme trains.

The three things I would build if I were 21

If I were 18 or 21 or 25 and preparing for the next decade, I would not obsess over finding the perfect AI-proof profession. I would build three things. First, a real craft: be genuinely good at something, with or without AI, whether engineering, sales, design or finance. Second, AI fluency, which is not prompt engineering. It is knowing how to delegate work to AI, what it does well, where it fails and how to check it, and using it to move beyond the boundaries of your role. Third, systems thinking: how your work connects to everything around it.

The higher you go in any organisation, the more your job already looks like this. A junior owns tasks, a manager owns a team, a founder owns a company, and a CEO spends the day asking how capital, people, product, customers and technology interact. AI simply pushes people towards that work much earlier.

So a better career question than "what one skill should I specialise in?" is: what system do I want to understand extremely well? Build expertise inside it, use AI to move around it faster, and on every problem ask what sits above it, beside it, before it and after it. If you want a craft that gives you a front-row view of how systems fit together, software is a very good one to start with. The free trial lets you try it without committing.