The AI Productivity Boom Hasn't Started Yet

AI is getting dramatically better. Businesses are spending heavily on it. Employees are using it every day. Controlled studies are showing substantial productivity improvements on many tasks.

Yet if you look at national productivity statistics, there is little sign of an AI-driven boom.

In Australia, GDP per hour worked was actually 0.2% lower over the year to June 2026.

So where is the AI productivity revolution?

I think the answer is relatively simple: at the level of individual tasks, the AI productivity boom has already started. At the level of businesses and the broader economy, it has barely begun.

We are confusing adoption with transformation.

The productivity gains are real

There is now good evidence that AI can make people substantially more productive when applied to the right work.

A well-known study of more than 5,000 customer-support workers found that access to a generative AI assistant increased issues resolved per hour by 14% overall, with gains of 34% for novice and lower-skilled workers.

Another experiment involving 758 consultants found that, on tasks within AI's capability frontier, people using AI completed 12% more tasks and did them around 25% faster.

These aren't marginal improvements.

But they also don't mean that AI makes every worker 20% or 30% more productive. The results vary enormously by task, experience and how well the AI can perform the work.

There is evidence of the opposite as well. In one study, experienced software developers working on mature codebases actually took longer when using AI.

That distinction matters.

AI doesn't have a single productivity effect. It has a rapidly expanding productivity frontier. Inside that frontier, the gains can be enormous. Outside it, AI can still create more work than it saves.

As the models improve, that frontier keeps expanding.

But even large improvements at the task level don't automatically translate into higher productivity across an organisation.

Saving time isn't the same as transforming a business

One of the most interesting studies involved more than 7,000 knowledge workers across 66 companies.

Workers using integrated generative AI spent around two fewer hours a week on email and worked less outside normal hours.

That's a real benefit.

But researchers couldn't detect a significant change in the overall composition of their work.

This gets to the heart of the current AI productivity debate.

Imagine AI saves an employee 30 minutes every day.

What happens to those 30 minutes?

Perhaps they produce more. Perhaps they improve the quality of something else. Perhaps they answer more email. Perhaps they attend another meeting. Perhaps some of the saving is consumed checking AI output.

Giving thousands of employees an AI assistant can create thousands of small pockets of saved time without fundamentally changing how the company operates.

Individual productivity tools create fragmented savings. Process transformation captures them.

And most organisations haven't reached that second stage yet.

Enterprise AI is still a toddler

This is where I think much of the discussion about AI adoption becomes misleading.

AI appears to be everywhere.

But measuring how many organisations have bought AI licences, enabled Copilot or have employees using ChatGPT tells us very little about how deeply AI has changed those businesses.

Australian evidence illustrates this well. The Reserve Bank found that around two-thirds of surveyed medium-to-large firms had adopted AI in some form. But nearly 40% described their use as minimal, while only around 30% had made more substantive progress.

Recent New York Federal Reserve research tells a similar story. AI use has spread rapidly across businesses, but investment remains predominantly minimal or modest.

Gartner recently reported that only 22% of organisations it surveyed had successfully scaled AI across multiple business units or adopted an AI-first approach.

So we are mistaking ubiquity for maturity.

AI is already everywhere, but enterprise AI is still a toddler.

Most organisations have taught it to help employees perform existing tasks.

Very few have rebuilt the way the organisation works around what AI can now do.

The next stage is process transformation

The first phase of enterprise AI has largely involved questions like:

How can AI help this employee write this document faster?

How can it summarise this meeting?

How can it help a developer write code?

How can it help customer service find an answer?

Those are useful questions, and there is still enormous value to capture from them.

But the much more interesting question is:

If AI existed when we designed this process, would we have designed the process this way at all?

Take customer service.

The first generation of AI might help an employee search a knowledge base and draft a response.

An AI-native process could look very different.

The AI could understand the customer's request, identify the customer and product involved, retrieve information from multiple sources, reason across that information, interact with other systems, execute permitted actions, generate the response and escalate only the cases requiring human judgement.

At that point we aren't making the existing customer-service process 20% faster.

We are designing a different process.

The same applies across finance, HR, legal, software development, procurement, operations and countless other functions.

The first wave of enterprise AI is making people faster. The next wave will redesign the work itself.

That's when the productivity numbers should start getting interesting.

AI also changes what is economical to do

There is another part of the productivity story that I think gets overlooked.

AI isn't only going to help organisations do today's work with fewer resources.

It makes work economical that previously wasn't worth doing.

A company might manually review 100 customer interactions each month for quality. AI could potentially review every interaction.

An analyst might investigate five hypotheses because investigating fifty would take too long. AI can dramatically change that constraint.

A manufacturer might have thousands of products for which creating and maintaining detailed knowledge has historically been too expensive.

A software team might have hundreds of small automation opportunities that were never worth allocating developers to.

AI changes the economics of all of these activities.

So productivity doesn't necessarily initially look like:

10 people → 7 people

It can look like:

10 people → dramatically more work gets done

Better analysis. Better documentation. Better customer service. More automation. More experimentation. More software. More decisions supported by data.

Some of the largest economic effects of AI may come from increasing the amount of economically viable cognitive work rather than simply reducing the labour required for today's workload.

Then organisations themselves change

Process transformation eventually leads somewhere even more significant.

If AI and agents perform increasing amounts of cognitive work, organisations will eventually start changing around that capability.

Roles change.

Management spans change.

Hiring changes.

Outsourcing changes.

Software changes.

Teams change.

Processes disappear.

Entirely new products and services become viable.

Very few organisations are genuinely at this stage today.

That's why I find it difficult to look at today's productivity statistics and conclude that they tell us much about AI's eventual economic impact.

We are measuring the outcome before most organisations have actually undertaken the transformation.

We've seen this movie before — but this time could move faster

There is historical precedent for this.

General-purpose technologies rarely translate immediately into economy-wide productivity.

Computers required new software, skills and business processes. The internet required organisations to rethink distribution, communication and eventually entire business models. Cloud computing wasn't transformative simply because a company moved a server from its own data centre into somebody else's.

The bigger gains came as organisations redesigned themselves around the new capability.

Economists sometimes describe this through the Productivity J-Curve: organisations invest in technology and complementary capabilities before the resulting productivity becomes fully visible.

AI requires exactly this kind of complementary investment.

Enterprise data needs to be accessible. Systems need integration. Processes need redesign. Employees need new skills. AI needs evaluation and governance. Permissions and security models need to change.

All of that takes time.

But there is an important difference with AI.

The underlying capability is improving extraordinarily quickly and is delivered largely through software running on infrastructure businesses already have access to.

AI adoption therefore doesn't require decades of factories, telecommunications infrastructure or physical equipment to be installed.

The technology can diffuse incredibly quickly.

Access to AI is moving much faster than organisations can currently redesign themselves around it.

That gap won't necessarily remain this large.

The boom is still ahead of us

None of this guarantees enormous productivity growth.

AI has real constraints. Models make mistakes. Verification has a cost. Enterprise data is messy. Security and regulation matter. Some processes are difficult to automate. Organisations are notoriously difficult to change.

And saving an employee 20 minutes doesn't automatically create 20 minutes of additional economic output.

Those are important reasons to be cautious about some of the more spectacular forecasts.

But I think it is equally wrong to look at weak productivity statistics in 2026 and conclude that AI has failed to deliver.

We have strong evidence that AI can already create large productivity improvements on suitable tasks.

We also have strong evidence that enterprise adoption remains shallow and that most organisations have barely begun redesigning their processes around the technology.

The AI productivity paradox is only really a paradox if we assume making a task faster should immediately make an economy more productive.

It doesn't work that way.

First individuals learn the tools.

Then processes change.

Then organisations change.

Only then do millions of local productivity improvements begin to show up clearly in company results and eventually national productivity.

We have strong evidence that the first stage is already delivering. Most enterprises have barely begun the next two.

That's why I don't think today's disappointing productivity numbers tell us the AI boom has failed.

I think they tell us it hasn't really started yet.