Agents save time, but do employees have time to build them?
On paper, spending three hours to save 15 minutes each working day looks like a good investment. The arithmetic misses an important question: where will the employee find the first three hours?
“What? I spent three hours building this agent, and it only saves me 15 minutes a day!”
This is a comment I hear quite often when conducting AI workshops.
On paper, spending three hours to save 15 minutes each working day looks like a good investment. The employee recovers those three hours after using the agent for 12 working days.
If the saving continues across 240 working days, it could add up to approximately 60 hours in a year. Even after deducting the initial three hours, the potential return appears worthwhile.
But the arithmetic misses an important question:
Where will the employee find the first three hours?
For someone operating in firefighting mode, three uninterrupted hours may be difficult to protect. There are customers to respond to, reports to complete, problems to resolve and deadlines that cannot be moved.
The agent may save time in the future. Building it adds work today.
The employee may not be resisting AI
When an employee says, “I do not have time to build an agent,” a manager may hear short-term thinking. Invest three hours now, recover that time within a few weeks and let the savings accumulate.
From the employee’s perspective, however, the situation looks different. The existing work has not disappeared. The employee is expected to meet the same deadlines while learning a new tool, examining a workflow, building an agent and testing whether it works. Spending several hours to improve one small task can feel disproportionate. Continuing with the familiar method may seem more practical, even if it remains slower.
Future time savings do not create capacity in an already overloaded week.
If leaders interpret every concern as resistance to change, they may overlook a practical barrier. Employees can understand the value of an agent and still lack the time to develop one properly. The organisation may be expecting transformation work to happen on top of business-as-usual work.
Building an agent is more than configuring a tool
The visible part of building an agent may look simple: write some instructions, connect information and ask it to perform a task. The more demanding work happens before and around the tool.
Someone must understand how the work is currently performed:
- What starts the process?
- What information is required?
- Which steps follow consistent rules?
- Which decisions require human judgement?
- What exceptions occur?
- Who needs to approve the result?
- How will the team ensure that the agent produces consistent output?
- When should the agent stop and return the work to a person?
- How will the team decide whether the output is reliable?
Experienced employees may carry many of these answers in their heads. They know which requests need special treatment, which information requires checking and which apparent shortcuts create problems later.
Turning that knowledge into an agent requires employees to make an informal workflow explicit. Some may find it difficult to articulate what they have learned through experience. Others may be cautious about sharing their knowledge if they are uncertain how the agent will affect their role.
This is not merely tool configuration. It requires process-analysis and workflow-design skills that employees may not yet have had the opportunity to develop.
A template agent can only take us so far
Organisations may try to reduce the effort by providing ready-made agents or templates. This can be a useful starting point.
A standard agent might summarise meeting notes, prepare a recurring report or classify incoming requests. Employees do not need to begin with a blank page, and common instructions can be tested once and reused. However, a template does not automatically understand how every functional team works.
A finance team, human resources team and operations team may all prepare reports, but they use different information, rules and approvals. Even two teams performing similar work may handle exceptions differently.
The closer an agent gets to performing meaningful work, the more it needs to reflect the real workflow. Organisations therefore need both reusable foundations and local knowledge. A template can provide structure, but employees must still explain where it fits, where it does not and what should happen when the normal process breaks down.
Leaders need to create the conditions for adoption
If a company expects employees to use AI, it should not leave each person to solve the adoption problem alone.
Providing access to an AI tool is a start. It is not the same as enabling people to redesign their work with it.
Singapore’s Ministry of Manpower reported in April 2026 that only 3.8% of firms surveyed were integrating AI into core processes. Among firms already using AI, 70.7% reported improved worker productivity, but implementation cost and lack of internal expertise remained major barriers. Ministry of Manpower
The productivity opportunity appears to be real, but organisations must invest before they can capture it.
Leaders can make that investment practical by:
- Selecting workflows that are worth improving.
- Allocating protected time for employees to map, build and test.
- Providing training that connects the AI tool to actual work, supported by workflow or technical expertise.
- Supplying reusable templates where the work has common elements.
- Allowing other priorities to be adjusted during the development period.
- Deciding who will maintain the agent after it is introduced.
This does not mean giving every employee several hours to build any agent they can imagine. The proposed workflow should have a clear problem, a plausible benefit and an owner.
An approved AI initiative should be treated as real work, not an extracurricular activity undertaken after everything else.
Employees also need to make the work visible
Employees understand details of the workflow that managers, technology teams and external vendors may not see.
Their participation is essential, but it should mean contributing knowledge and judgement, not carrying the entire adoption burden.
Employees can help by identifying repetitive work that is genuinely worth improving. They can explain where time is being spent, which cases are straightforward and which exceptions require experience. Managers should create a safe environment in which employees can say:
“I believe this agent could save 15 minutes each day. I need three hours to map the workflow, build it and test it. Can we agree on which work should be postponed so that I can do this properly?”
That turns a general complaint about having no time into a proposal that the manager can evaluate.
Employees should also define what success means. Saving time is only useful if the output remains accurate, usable and safe. The employee’s role is not simply to build the agent. It is to help the organisation understand what dependable performance looks like.
Training should create space to work on the workflow
An AI workshop can introduce the concepts and give employees supervised practice. It cannot automatically provide the time needed to turn every exercise into an operational workflow.
An agent created during training may demonstrate what is possible. Bringing it into daily use still requires testing, access to the right information and agreement on who will review its output. Without follow-through, the workshop can create another unfinished experiment.
Companies should therefore connect training with an implementation period. This could involve a small number of supported pilots rather than asking every participant to build an agent immediately.
A useful pilot should answer:
- Does the agent solve a real problem?
- How much time does it save in practice?
- What checking remains necessary?
- What happens when it encounters an exception?
- Who will maintain it when the work changes?
The result may be an agent that is ready to use. It may also be a decision that the workflow is not suitable for automation. Both outcomes can be valuable. Deciding not to automate an unsuitable workflow can prevent wasted investment and avoid weakening confidence in future AI initiatives.
Measure the complete time investment
The three-hour setup and 15-minute daily saving provide a useful starting point, but they are not the complete calculation. The organisation should also consider workflow documentation, training, testing, human review, maintenance, actual frequency of use and the consequences of an incorrect result.
A simple time-investment payback period can show when the initial effort might be recovered:
Time-investment payback period = total setup time ÷ time saved per use
For the example in this article:
Three hours ÷ 15 minutes per working day = 12 working days
This does not prove that the agent is worthwhile. It gives the organisation a testable expectation.
After the pilot begins, the team can compare that expectation with actual experience. If the agent saves less time than expected or requires frequent correction, the payback period becomes longer. If checking and correction take as much time as the agent saves, the expected return may disappear entirely.
If it can be reused by several employees, the return may arrive sooner.
The calculation should support a decision, not pressure an employee into finding time that does not exist.
Make time to create time
Microsoft’s 2026 Work Trend Index found that organisational conditions such as culture, manager support and talent practices were more strongly associated with reported AI impact than individual mindset and behaviour. The study was conducted across ten markets and was not Singapore-specific, but its underlying message is relevant: employee capability alone is not enough if the organisation does not create the conditions in which it can be applied. Microsoft Work Trend Index
Leaders should not simply tell employees to use AI more.
Employees should not be expected to redesign workflows quietly between urgent assignments.
Leaders need to provide direction, time, training and support. Employees need to contribute their knowledge of the work, surface the exceptions and help test whether the new workflow is dependable.
An agent may eventually return many more hours than it takes to build. But the initial investment still needs to come from somewhere.
If AI-supported work is important to the organisation, then the time required to design it must also be treated as part of the work.
Preparing your employees to redesign work with AI? Explore LIVO’s approach to solution design and team capability development.