Open the network tab on almost any 'AI-powered' SaaS tool and you will usually find the same thing: a form, a prompt template baked into the backend, and a single API call to a foundation model. The company selling it did real work building the UI and the workflow around it. But the AI itself, the part doing the actual thinking, often costs them a few cents per use. What they charge you is a different number entirely.
What an AI wrapper actually is
A wrapper product takes a general-purpose model, in many cases the cheapest tier available, wraps it in a form and a save button, and sells access to the combination. There is nothing wrong with that on its own. Good UI and a well-tuned prompt are real value. The problem is pricing that assumes you cannot see behind the curtain.
A typical Australian small business signs up for three or four of these tools over a year: one for social captions, one for meeting notes, one for email drafts, one for a customer-facing chatbot. Each charges $39 to $99 a month per seat. Add them up across a 12-person team and you are at $15,000 to $35,000 a year for four thin layers over the same handful of model calls.
The maths a $49 tool is hiding
A single Claude API call for a typical business task, drafting a 500-word email or summarising a meeting transcript, costs a few cents in raw model usage. Even a heavy user running 300 tasks a month is chewing through perhaps $10 to $20 in actual compute. The rest of a $49 subscription is UI, hosting, support, and margin. That is a legitimate business, but it means the same workflow built directly on the Claude API, or inside Claude Cowork with a saved skill, can cost a fraction of the price once you are past the first few weeks of setup.
Check whether the tool's marketing mentions a specific model or just says 'AI-powered' with no detail.
Time a single task in the tool, then estimate what the same prompt would cost run directly against Claude's API pricing.
Count how many of your subscriptions solve a problem that is really one well-written prompt plus a template.
Ask whether the tool stores your data to retrain a model you don't own, which is a cost you are paying in a different currency.
When a wrapper is still worth the money
Not every wrapper is worth replacing. If the tool does real orchestration, multiple model calls, database lookups, third-party integrations, and reliable error handling, the markup is buying engineering you would otherwise have to build and maintain yourself. A $200-a-month tool that reconciles bank feeds against five different POS systems earns its price. A $49-a-month tool that turns a bullet list into a LinkedIn post, using one model call and a system prompt, usually does not.
The test is simple: could a competent person rebuild the core workflow in an afternoon using Claude directly, or Claude Cowork with a saved skill and a scheduled task? If yes, you are renting something you could own for the cost of that afternoon. A rough rule that holds up across the Australian SMBs we have audited: if the tool's own help centre describes the feature in one sentence, it is probably one prompt.
What the audit actually looks like
The audit itself does not need to be complicated. Pull your last twelve months of card statements and list every line item with 'AI' in the merchant description. For each one, write down what task it does, how often your team actually uses it, and what a comparable Claude Cowork skill would need to replace it, roughly how many hours to set up once. Most Australian businesses running this exercise for the first time find at least two tools nobody has opened in a month, which is a cost saving before you have even touched the wrapper question.
What to do instead
Start by listing every AI subscription your business pays for and what each one actually does under the hood. For any tool costing more than $50 a month per seat where the workflow is a single prompt and a template, price out building it as a Claude skill instead. A Sydney bookkeeping practice we worked with replaced three separate $79-a-month 'AI assistant' tools with one Claude Cowork setup and two saved skills, cutting the annual bill from roughly $8,500 to under $2,000 in direct API and subscription cost, including the setup time.
This is not an argument against every AI SaaS tool. It is an argument for knowing which of your subscriptions are genuine engineering and which are a markup on a prompt you could write yourself. If you want a second opinion on your current AI stack before you renew anything, that is a conversation worth having before the next billing cycle, not after.
