Yes, one AI subscription can do the work you would otherwise give a virtual assistant, with limits I will name. And the reason so many owners get the decision wrong is not the technology. It is where they file the cost.
An AI plan lands in the same mental column as the streaming service and the accounting software. A subscription. Judged by its price.
An employee is never judged that way. You weigh what they produce against what they cost. Nobody looks at a salary in isolation and calls it expensive.
Put the plan in the wrong column and it looks pricey. Put it in the right one and the decision makes itself.
Measure your AI subscription in hours saved, not as a subscription cost.
Here is the honest arithmetic, what a plan actually replaces, what it does not, and where to start without overspending.
Key takeaways
- An AI subscription belongs in the staffing column, not the software column. Judge it by what it produces.
- One premium plan costs roughly a thirty-sixth of employing one person in Australia. The gap is not close.
- It replaces the administration around your work, not the substance of the work itself.
- The top-tier plans are for people who code. Most owners can start one tier down.
- The jobs at the bottom of the list get done too, not only the urgent ones.
Why does an AI subscription feel expensive when it is not?
Because it is priced like software and used like staff, and you are reading the price tag from the wrong shelf.
Every other subscription you own is a tool. It does one thing. You would never expect your invoicing software to answer your emails. So when a plan costs USD $200 a month, the instinct is to compare it with the last tool you bought, and by that measure it looks steep.
But the work it does is not tool work. It is the kind of work you hire a person for. Calendar and commitments. Drafting. Filing. Chasing. Matching one list against another at the end of the month.
Compare the plan with the output it produces and it is, in my experience, a no-brainer. The trouble is that almost nobody makes that comparison. They see the monthly charge, file it under subscriptions, and quietly decide it is a luxury.
What does employing someone actually cost compared with an AI plan?
Roughly thirty-six times more, and the arithmetic is public.
Start with the person. The Australian Bureau of Statistics puts average weekly ordinary time earnings for a full-time adult at $2,083.70, as of May 2026. Over a year that is about $108,350. Add superannuation at the 12 per cent guarantee rate and you are at roughly $121,350 a year before a single other on-cost. Call it $10,100 a month.
Now the plan. The two I pay for are Claude Max at the 20x tier and ChatGPT Pro. Both are USD $200 a month. At the Reserve Bank's rate on 1 September 2026, that is about AUD $279 each. Roughly AUD $3,350 a year, per plan.
One person, $121,350. One plan, $3,350. That is the thirty-six to one.
My own spend is two plans, so about AUD $558 a month. That is my number, not a benchmark, and I will come back to why it is two.
One person costs about thirty-six times what one premium plan costs. The question is what the plan actually does for the money.
Two peer-reviewed studies say what it does is real and measurable. In Science in 2023, Noy and Zhang gave writing tasks to 453 professionals. Task time fell 40 per cent and quality rose 18 per cent. In the Quarterly Journal of Economics in 2025, Brynjolfsson, Li and Raymond tracked 5,179 support agents. Output rose 14 per cent on average, and 34 per cent for the newest and least skilled.
Those are task-level gains, not a person replaced. I want to be exact about that, because the honest claim is narrower than the headline and it is still a very good claim.
And the local baseline: the ABS reported in June 2026 that 12 per cent of Australian businesses used AI in 2024-25. Small and micro businesses sat at about 11 per cent. Most of the businesses you compete with have not started.
What does one AI plan actually replace, and what does it not?
It replaces the administration around the work. It does not replace the substance of the work. Knowing where that line sits is the whole skill.
This is Level 2 of the four levels I teach, the one I call Delegate. You are not asking the AI questions any more. You are handing over outcomes, with a checkpoint, and reviewing what comes back.
What fell away. In my publishing business, at various points, I had four virtual assistants working on different parts of the book production process. I had graphic designers. I had ghostwriters. All of that has fallen away.
What runs in its place is a system that covers, in effect, four roles. An executive assistant managing my calendar and commitments. A social media manager. A copywriter producing search-optimised blog content, social posts, planning and strategy. And operations, including the end-of-month finance work, where it is very efficient at matching and reconciling.
There is a fifth job nobody expects. It is very good at historical data cleanup, auditing and reorganisation. The jobs that always got pushed to the bottom of the list, the ones you were never going to get to, get done as well as the urgent ones. That is not time saved on today's work. That is capacity for work that would otherwise never happen.
What it does not replace. The actual content. The substance still comes from the people, and in my experience the majority of the real ideas do. The administration around content is handled. The thinking is not.
Visual assets are the other limit. Left to itself, the output leans into the same look every time. Tools like Canva still matter, and real photographs of real people remain a priority.
Beyond those two, there has not been much I have handed over that it could not handle.
One condition sits under all of it. The plan replaces a virtual assistant only when there is a system around it: a short context file so it knows your business, and a checkpoint so you see the work before it leaves. Without those, you are back at Level 1, asking questions and doing the work yourself. With them, an ordinary owner on a lower tier can run a well set up operating system, and the plan starts earning its place.
It replaces the administration around the work, not the work.
How do you choose your first AI plan without overspending?
Start one tier lower than you think, and pick one plan rather than two.
- Do not start at the top tier. I pay for the USD $200 plans because I write code, and the higher usage limits matter for that. If you do not code, you are paying for headroom you will not use. Start a tier down and move up only when you hit the limits.
- Pick one. I would still recommend Claude, but I am genuinely on the fence, and I use both. Both have a desktop app. ChatGPT is better value: you get more for the subscription. Either one works. What does not work is paying for both before you have made one of them earn its keep.
- Write down three tasks before you pay. The three things you personally did twice last week that you want off your desk. Not your highest-stakes client work. Something ordinary and repeated.
- Run one of them this week. Hand it over properly, with what done looks like and when you want to see it. Then price the hours it gave back against the monthly charge.
- Revisit the tier in a month, not a day. If you are hitting usage limits on real work, move up. If you are not, you have your answer.
Run that, and by the end of the month the subscription line will have stopped looking like a subscription.
Frequently asked questions
Is ChatGPT or Claude better value for a small business? ChatGPT gives you more for the subscription, and I still lean towards Claude, so I am honestly friends with both. They each have a desktop app, and most of what you will build works the same way in either. Pick the one you already pay for and make it earn its place before you add the other.
Do I need the USD $200 a month plan? No. Claude sells its Max plan in two tiers, 5x at USD $100 and 20x at USD $200, and ChatGPT Pro is USD $200. I pay for the top tiers because I code and the usage limits matter. An owner running admin, content and finance through a well set up system can start lower and only move up when the limits actually bite.
What can an AI subscription not replace? The substance of your content, and distinctive visuals. The ideas still come from people, and the images still lean towards sameness without a real photo or a designer's hand. Everything around those two, the drafting, scheduling, filing, matching and chasing, I have been able to hand over.
How do I work out if an AI plan is worth it? Price the hours it returns, not the monthly charge. Take one recurring task, hand it over for a week, and count the hours you did not spend. Compare that with the plan's cost, converted to Australian dollars so you are comparing like with like. If the hours cost you more than the plan, you have your answer.
Is AI cheaper than hiring a virtual assistant? On the arithmetic, one premium plan is roughly a thirty-sixth of one full-time salary with super. On the work, it is comparable to a virtual assistant for the administration around your business, based on what I see produced. It is not comparable for the parts that need a person's judgement, and it will not be for a while.
The bottom line
You would never judge a staff member by their salary alone. Stop judging the plan by its price.
The plan is not a subscription. It is capacity. Some of that capacity goes on today's work, and some of it goes on the work that was never going to happen otherwise. That second part is the one that surprised me.
If you have been sitting on the decision because USD $200 a month felt like a lot for software, run the comparison in the right column. One person, about $121,350 a year. One plan, about $3,350.
Then hand over one task this week and count the hours.
If you want the four levels behind this, the post that lays them out is here. And if the deeper pattern is familiar, that your business runs on you rather than on systems, that is where all of this started.
Sources
The wage figure is the Australian Bureau of Statistics, Average Weekly Earnings, May 2026. The superannuation rate is the Australian Taxation Office's guarantee rate from 1 July 2025. Plan prices are from Anthropic's and OpenAI's own pricing pages, and the exchange rate is the Reserve Bank of Australia's published AUD/USD rate for 1 September 2026. Adoption figures are the ABS Business Characteristics Survey, published June 2026. The two productivity studies are Noy and Zhang, Science, 2023, and Brynjolfsson, Li and Raymond, Quarterly Journal of Economics, 2025.
