Why the world will need more code, not less

AI can write code. That doesn't mean we will need less of it.

One of the more common predictions surrounding generative AI is that the need for software development will decline.

If machines can generate software, surely we need fewer people writing it.

It sounds logical.

We think it may misunderstand what happens when the cost of creating something falls dramatically.

The important question isn't whether AI can produce code.

It can.

The more interesting question is: what happens when millions more organisations can afford to turn ideas, processes and problems into software?

Our view is that the world is likely to need more code, not less.

Software demand has never really been satisfied

Look inside almost any established organisation and you will find a backlog.

A customer portal that needs rebuilding.

A spreadsheet that has quietly become a business-critical system.

A legacy platform everyone knows needs replacing.

A workflow that still requires six people to copy information between systems.

An integration that was postponed because it was too expensive.

A mobile experience customers dislike.

A reporting process that takes three days every month.

A new product the business wants but technology cannot prioritise yet.

The constraint has rarely been a shortage of ideas.

It has been the cost and complexity of turning those ideas into reliable software.

AI changes that constraint.

Productivity creates new demand

We have seen this pattern before.

Higher-level programming languages did not eliminate software development.

Open-source software did not eliminate it.

Cloud computing did not eliminate infrastructure.

Low-code platforms did not eliminate applications.

Each reduced the effort required to accomplish certain tasks.

And each contributed to an expansion of what businesses attempted to digitise.

AI-assisted engineering is likely to do the same.

If something that previously required twelve months can eventually be delivered in four, organisations don't necessarily spend eight months doing nothing.

They attempt more.

The long tail of software is enormous

Large enterprises can already justify building highly specialised systems.

Smaller organisations often cannot.

That could change.

AI-assisted engineering may make software economically viable for increasingly narrow use cases.

A logistics company may build software around a specialised operational process.

A manufacturer may digitise workflows previously managed through spreadsheets.

A retailer may build customer experiences that previously required enterprise-scale budgets.

A government department may finally modernise a twenty-year-old internal platform.

The potential software backlog across the global economy is enormous.

Lowering the cost of engineering unlocks that backlog.

More software also creates more complexity

There is another side to this.

Every new system creates something that must eventually be operated, secured, integrated, monitored, updated and potentially replaced.

Software creates dependencies.

Dependencies create complexity.

The engineering challenge therefore shifts.

Writing every line manually becomes less important.

Understanding the system becomes more important.

Architecture becomes more important.

Security becomes more important.

Testing becomes more important.

Observability becomes more important.

Data governance becomes more important.

Engineering judgement becomes more important.

AI changes what engineers spend their time doing

The engineer of the future may spend less time writing repetitive boilerplate code.

That is a good thing.

More time can be spent understanding requirements, designing systems, reviewing architecture, validating AI-generated output, solving difficult problems and thinking about how technology affects the business around it.

Engineering becomes more leveraged.

One good engineer equipped with powerful tools can accomplish considerably more.

But greater leverage does not eliminate engineering.

It increases the consequences of engineering decisions.

Legacy systems alone represent decades of work

Then there is the software we already have.

Across banking, insurance, government, telecommunications, manufacturing and almost every major industry are systems that have accumulated over decades.

Much of that technology will need to be modernised.

Some should be replaced.

Some should be re-engineered.

Some should remain exactly where it is.

AI-enabled analysis can help organisations understand these environments more quickly.

But the scale of the modernisation opportunity remains enormous.

We have barely begun.

More code means more infrastructure

Software doesn't exist independently.

More applications mean more APIs.

More APIs mean more traffic.

More AI means more computation.

More computation means more processors, storage, networking and power.

The software explosion and infrastructure explosion are therefore connected.

The future digital economy will require both.

The question isn't whether we can generate code

That question has largely been answered.

The harder questions are:

What should we build?

Why should we build it?

How should it fit into everything else?

How do we secure it?

How do we operate it?

How do we know it is correct?

And how do we ensure it produces an outcome worth having?

AI makes producing code easier.

Engineering determines whether that code becomes something valuable.

And as the barriers to software creation continue to fall, we expect the amount of software surrounding us to increase dramatically.

The future needs more code.

It also needs better judgement about what that code should do.