CBA CEO Matt Comyn used the phrase “waste work” to describe the low-quality AI output that now flows through corporate workflows, as AI costs billed with tokens increase with task complexity.
Matt Comyn, chief executive of the Commonwealth Bank of Australia, used a speech on Monday to highlight two AI adoption issues that large corporate buyers have been quietly working on for several months.
The first is that the cost of running generative AI within corporate workflows is increasing substantially faster than most companies budget as task complexity increases.
The second is what Comyn called “waste work,” the low-quality AI-generated text, code, and analysis that flows through a company’s internal systems when employees use AI without sufficient quality control.
The cost formulation is the part that will resonate with the audience of corporate IT buyers. Token-based pricing, the character-based billing model that Core Model Labs use to charge enterprise customers, has escalated over the past 18 months from a modest line item to a significant operational expense category.
Comyn’s point is that the cost compounds faster than expected because token consumption per task increases nonlinearly with task complexity: a simple summary task may consume 1,000 tokens, but a multi-step reasoning task with tool usage may consume more than 100,000 tokens for the same output value. Companies that priced their AI releases based on simple tasks are now seeing bills climbing the complex tasks curve.
This problem is not specific to ACB. Morgan Stanley doubled its forecast for job losses in European banking and AI last week, in part because of evidence that AI’s cost-benefit ratios are tightening at exactly the time big institutions expected them to loosen. The token cost escalation problem Comyn described is the underlying mechanic: the same AI implementation that worked at pilot stage volumes can produce 10 to 100 times the costs at production stage complexity.
The result is the squeeze on corporate AI acquisitions that Comyn predicted will intensify through 2026: Companies intensifying scrutiny of AI-related spending as pressure increases to demonstrate return on investment.
The “sloppy work” frame is the more colorful but equally substantive half of the speech. The category Comyn was naming, low-quality AI-generated output that nominally completes a task but actually degrades the subsequent workflow, is the analog in corporate knowledge work of the social media “overwhelming AI” problem that emerged in 2024 with imaging tools.
The banking version looks like this: an employee uses ChatGPT to compose a customer email, the email is technically grammatical but objectively inaccurate, the recipient takes the inaccuracy as a compromise, and the bank resolves the resulting complaint three weeks later at a substantially higher cost than the original work would have generated without help.
The specific context of CBA is significant. The bank announced 90 job cuts earlier this year and another 120 cuts in May explicitly attributed to AI-driven productivity gains, along with a AU$90 million AI workforce reskilling commitment.
Comyn’s comments therefore fall within a CBA strategy that has visibly committed to AI substitution at scale: the “useless work” framework is not a defensive critique of AI by a bank that has rejected the technology, but a sharper internal reading of AI deployment by one of Australia’s current largest adopters.
It’s also worth noting the broader context of Australian banks. Sam Altman has been arguing Over the past month, an AI jobs apocalypse is unlikely at the macro level, and jobs data through March 2026 has so far supported the conservative reading.
Comyn’s comments complicate that picture: Macro labor data does not yet show a large-scale shift, but operating margin data within large corporations is beginning to show the trade-offs between AI, cost and quality that the CBA now explicitly names.
The substantial implication is that the AI cost narrative for 2024-2025, that token prices were falling so rapidly that the issue of implementation economics would resolve itself, has been structurally inverted.
The drop in per-token prices has been outweighed by rising per-task token consumption as companies move from pilot deployments to production use cases. The phase of procurement discipline that Comyn forecasts until 2026 is, on this evidence, the predictable consequence.
