If your admin load has grown to the point where you are weighing up a virtual assistant, it is worth putting Claude Cowork on the other side of the scale first. This is not a hit piece on assistants. A good assistant is genuinely valuable, and there is work Cowork simply should not do. It is an honest division-of-labour analysis with the Australian dollar figures attached, so you can see where each option earns its keep.
The costs side by side
Start with the raw numbers, because they frame everything else.
Offshore VA: roughly $15 to $30 AUD an hour, which lands around $1,500 to $4,500 a month for part-time hours.
Local admin hire: $65,000 to $85,000 a year, plus superannuation and the usual overheads of an employee.
Claude Cowork: $30 to $45 AUD per user per month in licensing, plus a one-off implementation typically from around $8,000 if done properly, with near-zero marginal cost for each additional workflow.
The shape of those costs differs as much as the size. The assistant is a recurring cost that scales with hours. Cowork is a small fixed licence plus an upfront build, after which adding another workflow costs almost nothing. That difference is the whole story when your workload is repetitive and growing.
What Cowork does better
On a particular slice of work, the software wins clearly.
Document production: reports, decks, and spreadsheets from templates, around the clock and without a queue.
Research and summarisation at volume, where a human would need hours that Cowork compresses to minutes.
Availability: never sick, never resigns, and scales to a busy week without a renegotiation.
Consistency: the same workflow runs the same way every time, which matters for anything quality-sensitive.
The common thread is volume and repetition. Anything you would describe as the same job done over and over, on documents and data rather than people, is where Cowork pulls ahead. It does not get bored on the fortieth report of the week, and the fortieth is as clean as the first. For a small business drowning in routine output, that consistency at near-zero marginal cost is the part that changes the maths.
What a human assistant does better
And on another slice, the person wins just as clearly, which is why this is a division of labour and not a replacement.
Phone calls and the relationship judgement that goes with them.
Ambiguous, novel situations with no precedent to work from.
Physical-world tasks and the wrangling of suppliers and couriers.
Reading the room, which no amount of automation replicates.
The common thread here is judgement under uncertainty. The moment a task needs someone to weigh competing interests, soothe an annoyed client, or improvise when the plan breaks, you want a person. Trying to push that work onto software produces confident answers that are subtly wrong, which is worse than slow. Knowing where this line sits is most of the skill in setting the two up to work together.
The realistic hybrid
In practice, most businesses do not choose one or the other. They land on a split where Cowork absorbs the repetitive 60 to 70 per cent of the assistant workload, while a human keeps the judgement work, often at fewer hours or redeployed to higher-value tasks. Here is a worked example. A business spending $3,500 a month on VA hours moves to about $1,200 of human time plus $100 of licensing. After implementation costs, that saves around $26,000 a year in the first year alone, and more once the upfront build is paid off.
The saving is real, but the better outcome is usually the redeployment. The hours you free up from formatting reports are hours a capable assistant can spend on clients, suppliers, and the judgement calls that actually need a person. You are not shrinking the role so much as upgrading it. For most owners, that is the difference between an assistant who is always behind and one who finally has room to think.
When not to switch
To keep this honest, here is when you should keep the human model and leave Cowork out of it. If your processes are messy and undocumented, automating them just makes the mess faster. If there is no internal owner to maintain the workflows, they will rot. And if the workload is dominated by phone calls and relationships rather than document work, a person is simply the right tool. Automating the wrong thing is more expensive than not automating at all.
We run Cowork implementations with guardrails for businesses across Sydney and beyond. If you are not sure which side of the line your workload sits on, book a discovery call and we will map it with you before you spend a dollar.
