They promised us
flying cars.
We got bank feeds.

In 1926 Henry Ford told reporters that within two years ordinary Americans would fly to work. He had led the automobile revolution and thought soon aviation would do the same. The 1939 World's Fair handed out pamphlets describing a future of three dimensional travel with no congestion. By 1962, The Jetsons made it a cartoon fact.

None of it happened.

The pattern is familiar. Every major wave of technology made the same promise: machines would take the drudgery so humans could focus on judgment, direction and decisions. In manufacturing, logistics and communications, that promise was largely kept. Entire categories of repetitive work disappeared.

In accounting, the promise was not kept. What accounting got was bank feeds.

Right now, somewhere in Australia, a small business owner is squinting at their computer trying to work out whether they can make payroll.

They are checking unpaid invoices. They are looking at a bank feed full of uncoded transactions. They are trying to remember whether that supplier payment was materials, software, insurance, freight, repairs or something else entirely. They're exporting a P&L as an antiquated CSV because the reporting still doesn't show what they actually need.

They are doing this in software that looks and behaves much the same as it did a decade ago. Except now it costs more. A lot more.

Most of accounting is still manual. Bank transactions need reconciling. Bills need coding. Cash flow needs interpreting. The owner still has to log in, click through and decide what matters.

This isn't because the technology doesn't exist. Machine learning, natural language processing and large-scale pattern matching have been viable since the early 2010s. The knowledge is there. What's missing is the incentive.

When a platform has millions of locked in customers and high switching costs, the incentive to remove work is weak. The incentive to meter the work and charge more for access to it is much stronger. Why eliminate the problem when you can charge every year to make it slightly more tolerable?

There is a pattern that appears in mature software markets. A company wins the category. The product becomes embedded. Customers build processes around it.

Then the roadmap slows. The software gets more expensive without getting meaningfully better. Small improvements are announced as major releases. Customers grumble but they don't leave: because switching costs are real and alternatives are thin.

Xero shows the pattern clearly. It has moved from customer-led innovation toward rent extraction.

The most visible sign is the metronomic price increases. In 2016, the plan most growing Australian businesses needed cost $60 per month. From July 2026, the comparable plan is $107 - a 78% increase. The product has not improved by anything like that margin.

It shows up in other ways too. For years, small customer requests sat on the public roadmap with hundreds of votes - some for five years, some for ten, one for fourteen. Basic improvements arrived late. More recently, the company began charging partners for API access, monetising the integration layer instead of improving the core product.

Then there is executive compensation. In 2024, the CEO's target compensation was $22 million. The CEO is not the founder. They joined in 2022, long after Xero had become entrenched. That revenue came from small business owners with limited alternatives and high switching costs.

A rent-extractive model can look rational when you have market position. The business takes more from customers while giving them less in return. Over time, that asymmetry becomes corrosive. Customers tolerate it while switching is hard. They remember when it becomes possible to leave.

There is a fundamental difference between adding AI features to accounting software and building an accounting system on AI from the start.

When AI is bolted onto old software, it inherits the old assumptions. The human is still expected to do the work - log in, check the feed, review the coding, chase the missing bill. The AI assists within that workflow. The fundamental model stays the same: human labour, mediated by software.

AI native software is built on a different assumption. The human sets the goals. The AI does the work. AI does not just automate tasks - it reasons, holds context, and learns the shape of a business over time. The past is software you operate. The future is a system that does the work and asks for your judgment when it matters.

Applied to accounting, that is not an incremental improvement. It is the difference between a faster horse and a car.

Marbles was built from first principles for that world. Not refactored. Not retrofitted. And we went one step further: instead of selling you smarter software to operate, we combined the software and the bookkeeping into one service.

Routine bookkeeping is automated wherever it can be, with our Australian team reviewing the work and handling exceptions. Bank transactions are reconciled and coded. Receipts and invoices are matched. Cash flow stays current. BAS is prepared as the books are maintained, not scrambled together at quarter end.

There is still a human in the loop - accounting needs control, audit trails and review. But control should not mean doing every low value task yourself. It should mean knowing what happened, why it happened, and when your judgment is required.

That is what Marbles delivers: bookkeeping as a service, with the software built in. The books get done - every month, in the background - and Australian small business owners get their time back.