opinion
AI Is Making Traditional Automation Faster and Cheaper
Faster, lower-cost software development is making automation projects viable that businesses once struggled to justify.
What Changes When Software Becomes Faster to Build?
Faster development can bring the cost of automating a workflow within reach of the benefit it will deliver.
AI-assisted development changes that calculation. Our experience with AI-Augmented Software Development is that improved development speed has made it possible to automate processes that previously did not justify the investment. The benefit comes from the software that can now be built and put to work.
There are two places to look for that benefit:
- Building the Software
The effort required to develop it. - Running the Process
The effort required to operate it afterwards.
They are distinct measures. A quicker build does not establish a successful automation unless the finished process delivers a useful result.
What Changed in Our Own Payroll Process?
Adaca One allowed us to automate our payroll calculations for the Philippines, moving from three people allocated to calculating payroll to a process with one person providing human oversight.
We built the product around our own operational processes and the requirements of our Philippines payroll. Our need was a system that fitted that combination. Improved development speed helped make building that product for ourselves a viable undertaking.
The minimum viable product (MVP) was the first working version with the core functions. It took three months to build, and the system was in live use five months after development began. The payroll calculations follow fixed rules. No large language model (LLM), the type of AI trained to process and generate language, performs them.
The manual payroll calculation work was automated, and fewer people are now needed to maintain the process. The remaining person:
- Cross-checks a subset of calculations.
- Checks that staff have entered their leave appropriately.
- Checks that human resources (HR) has approved the leave.
These figures reflect our internal payroll process as of September 2026. The staffing comparison describes involvement in the process, not employees leaving the business or full-time roles saved. Without hours, costs and a comparable previous build estimate, it does not establish a percentage productivity gain, development speedup or financial return.
Does the Finished Process Need an AI Agent?
An AI agent, in this context, is software that uses an AI model to choose actions and use tools towards a task. A workflow based on explicit rules and calculations may be better served by software that executes those rules directly.
In the opportunities we are seeing, larger-scale process automation is often a higher priority than putting agents into production. Much of the work follows defined business rules and quantitative processes. There is no inherent requirement for an LLM to be involved in executing those rules.
The distinction matters when a business starts with a desire to deploy AI and then looks for somewhere to put it. The more useful starting point is the workflow: what happens, what decisions are required and which parts can be specified clearly enough to automate.
Our view is that relatively few of the workflows we encounter require an agent. That is a position drawn from our work, rather than a measured estimate of demand across the economy. It leaves room for agents where the task warrants them, while giving conventional automation proper consideration.
What Do the National Figures Tell Us?
The national figures describe a weak productivity result, without isolating AI's contribution or explaining the cause.
Gross domestic product (GDP) measures the value of goods and services produced. In the June quarter of 2026, Australian GDP per hour worked was flat at the published precision and 0.2% lower than a year earlier. GDP per capita was also flat in the quarter, but 0.7% higher over the year. These measure different things: output per hour worked and output per person. ABS, June 2026 national accounts.
Those results provide context for the pressure to improve business processes. They cannot show that AI has delivered no benefit inside individual organisations, or establish that deploying more agents would reverse the national result.
A business can assess a specific workflow more directly. It can establish what the work required before automation and what it requires afterwards, while accounting for the cost of building and maintaining the system. That assessment provides a basis for an investment decision without requiring a claim about the whole economy.
How Should a Business Choose What to Automate?
We recommend comparing workflows by the impact of automating them, the cost of building the solution and the risks involved. The business also needs to decide who will own the software and lead its delivery.
The assessment should begin across the business, giving different opportunities a common basis for comparison:
- Scope the Workflow
Describe the work, the people involved, the systems it depends on and the decisions required to complete it. - Quantify the Impact
Establish the current effort and the change automation could deliver. Record the assumptions behind that estimate so the result can be checked later. - Estimate the Cost
Assess the build and the ongoing work needed to operate and maintain the automation, including human oversight. - Assess the Risk
Consider what could go wrong, the consequences and the controls needed before the process is put into use. - Rank the Opportunities
Compare the impact, cost and risk together, then proceed with the strongest opportunities. The scoring should reflect the business's priorities rather than reward a particular technology choice.
An agent may emerge as the appropriate solution for some workflows. Others can be handled through software that follows defined rules. The choice follows the work and the evidence supporting the investment.
The useful next question is which process has become worth automating now that building the software takes less effort.