What AI can’t do: The new skills every developer needs in 2025 and beyond

Artificial intelligence can now write, refactor, and deploy code faster than any human ever could. What it can't do is understand why it's doing it. And that matters because without context, AI can produce outputs, but not the right outcomes - code that fits into a live system, aligns with regulation, and behaves predictably within a web of dependencies and business logic.

That gap between what AI can do and what it should do is shaping the next generation of software development. AI isn't a replacement for developers, it's an amplifier. The craft is evolving from syntax to semantics, from writing code to curating meaning. Our role is no longer to make the machine work, but to make it work with us.

From Coder to Curator
When I began my career, the path for a software developer was clearly defined. You learned a programming language, built experience, earned seniority, and eventually moved into architecture or leadership. It was a steady climb up a familiar ladder.

That model no longer fits. Today, AI can read more code in an afternoon than I have in a lifetime. It can understand syntax perfectly, but it has no awareness of systems, business rules, or the nuances that drive human decisions.

That's why the developer's role has shifted. We are no longer producers in the traditional sense. We've become curators and orchestrators that shape how intelligence is applied, ensuring what's produced fits within the wider system and the organisational reality it serves.

The Three Core Skills Every Developer Needs
At bet365, our Platform Innovation Team has spent the last two years exploring what this shift really means for engineering. We've distilled it into three core skills that I believe every developer will need in the age of AI.

1. Prompt Engineering
Prompt engineering isn't about typing clever instructions into a chatbot, it's about learning how to communicate with a machine, so it truly understands what you mean.

AI responds differently depending on how you frame a question, how much context you give, and how clearly you describe the outcome you want. The more precise and relevant the context, the better the result.

When I work with AI coding agents, I think of it as giving directions to a highly capable but very literal assistant. If I'm vague, it will still act, just not necessarily in the way I intended. If I'm specific, it can produce something outstanding.

That's the real skill; knowing what information matters, and how to express it clearly. The best developers will treat prompting as a form of critical thinking and communication. Not writing commands, but structuring ideas in natural language so the machine can do its best work.

2. Risk-Aware Decision-Making
AI has no sense of consequence. It will always give you an answer, even when the right answer is to wait. Humans, on the other hand, have intuition, judgment, and moral context. They can weigh up what should happen, not just what can happen.

I've seen this distinction play out in code generation. A model might confidently produce a function that technically works but violates an upcoming regulation or exposes a performance risk under load. A human developer knows when to pause, to look beyond the parameters in front of them, to make a decision informed by experience and awareness of the wider system.

As developers, our role is to manage that risk. It is to bring ethical and contextual reasoning to a process that otherwise lacks it. AI can calculate probabilities; it can't interpret implications.

3. Data Fluency
AI lives off data. It doesn't just consume it, it depends on it. That means understanding not only where data comes from, but how it's structured, what it represents, and why it matters.

At bet365, we've learned that the shape of data defines the shape of intelligence. Our work with GraphRAG - a graph-based retrieval approach that gives AI access to context - taught us that the way we model relationships between entities determines the quality of insight we get back. A graph doesn't just connect tables or fields; it encodes meaning.

In practical terms, this means giving the AI a map of your world, how your systems interconnect, why they're designed that way, and what dependencies matter. That's how you make AI think like your organisation, not just like the internet.

Teaching Machines to Work with Us
We've spent 25 years teaching humans to adapt to machines. GenAI flips that. Now we must teach machines to work with us.

That shift is both liberating and challenging. For the first time, I don't need to learn another programming language to be effective. I need to learn how to communicate with intelligence - to give it purpose and context.

We're also moving into a multimodal world, where interaction isn't limited to text. Machines can speak, see, and respond in real time. That changes everything. Developers will need to decide the right communication model for each problem. Not every interaction belongs on a webpage or in a form. Sometimes the best interface will be voice, gesture, or something we haven't even invented yet.

Beyond Code Litreacy
There's a growing debate about whether developers still need to "read" code. My view is simple: you still need to understand how systems fit together, especially when working with legacy infrastructure. But you don't need to handcraft every line to add value.

AI can now do much of the mechanical work of refactoring, testing, and documenting. Our job is to ensure it's solving the right problem. The skill isn't in writing syntax; it's in defining outcomes.
That requires systems thinking. You need to know where the code fits, what it interacts with, and why it exists. The machine can produce a perfect function, but only a human can decide whether it belongs.

From Automation to Amplification
This isn't a story about replacement. It's about amplification. We're not removing developers from the loop; we're putting them at the centre of it.

At bet365, the engineers who've adopted AI most effectively are the ones who use it to handle the repetitive work, freeing their time for design, innovation, and problem-solving. They've learned to think of AI as a grandmaster at chess who doesn't know the current game. Once you teach it the board - the context - it plays brilliantly.

That's the new partnership model: the human defines the rules, the AI accelerates the play.

A More Human Future
The irony of all this is that as AI becomes more capable, software development is becoming more human. The skills that matter most now - clarity, empathy, judgment, creativity - are ones the machine can't replicate.

We're entering a phase where communication and context outweigh code volume, where understanding intent is as important as writing logic. The future of development belongs to those who can teach intelligence, not just use it. Because, while AI can write the code, only we can make it mean something.