CBA's CEO Just Dropped Two Uncomfortable Truths About Corporate AI: Surging Costs and 'Work Slop'

"In the early days, you want people to start using the tools – and they didn't really cost a lot [...] now the way the models work, the amount of context that you can put into it, your costs don't scale linearly," Comyn explained . This creates what he labels a "key emerging management challenge": CFOs and CTOs are locked into annual budget cycles that cannot account for cost-per-task variability that can swing 10x to 50x depending on complexity .

This dynamic is not abstract. GitHub's June 1, 2026, transition to usage-based billing for Copilot made it painfully concrete. Previously, a flat subscription fee covered all requests; now, every token of input, output, and cached context is metered via GitHub AI Credits, where one credit equals $0.01 . Heavy users running agentic coding sessions are immediately seeing their bills spike, with some reporting projected increases of 10x to 50x . A single complex agent session against a top-tier model can now burn through an entire monthly credit allowance in a single run .

'Work Slop' Defined: The Hidden Cost of Valuedless AI Output

Comyn's second major warning is qualitative. He used the phrase "work slop" to describe the proliferation of AI-generated text, code, and analysis that appears productive but adds negligible real value . This output does not simply fail to help — it actively creates hidden costs: every piece of AI-generated content that enters a workflow must be reviewed, fact-checked, edited, or discarded by human employees .

This is the enterprise counterpart to the consumer-web phenomenon of "AI slop" — search-engine spam, formulaic social media posts, and auto-generated filler. In a corporate setting, the risk is higher. "Work slop" can silently degrade internal decision-making, compliance documentation, and even customer-facing products if it passes through inadequate quality gates . The more companies deploy AI across every function without rigorous validation, the larger the pile of valueless work they must pay humans to clean up.

CBA's A$2.4 Billion Tech Bet and the Workforce Equation

Comyn's warning is not that of a disinterested observer. CBA invests roughly A$2.4 billion annually in technology, more than any other major Australian bank by at least $500 million . That spending is framed as a strategic bet on AI-led productivity, but Comyn's comments acknowledge that this budget line is itself exposed to the same unpredictable inflation he is warning others about .

At the same time, CBA is demonstrating the workforce impact of AI productivity gains in real time. The bank cut around 300 jobs in early 2026, on top of 90 support-staff roles previously replaced by an AI chatbot and another 120 positions eliminated in April . Comyn has been explicit: AI "will take away jobs at businesses across the economy" and firms must help staff prepare for that future rather than pretending otherwise .

Yet CBA has also committed A$90 million over three years to its Future Workforce Program, a significant reskilling initiative for its 30,000-plus employees . The program includes a new internal career platform called Grow Your Career, AI-focused training, and skills mapping designed to make internal mobility transparent . This dual posture — cutting jobs while reskilling—is an honest reflection of Comyn's view that the disruption is already here, and companies need a strategy for the workers who stay as much as for those who leave .

Strategic Implications for Every Enterprise

The convergence of Comyn's warning, CBA's own actions, and the immediate trigger of GitHub Copilot's billing change points to three strategic imperatives for any organization running AI in production:

1. Budgeting Must Become Dynamic and Metered. The era of fixed annual AI budgets is over. Token-based billing models turn AI into a variable-cost service, comparable to cloud computing. Companies need real-time cost monitoring, per-team credit allocation, usage caps, and the ability to adjust budgets mid-cycle — disciplines that cloud-native companies learned a decade ago but that many traditional enterprises have yet to adopt for AI .

2. Quality Control Is Not Optional. "Work slop" creates a direct link between quality failure and cost overrun. Every unvetted AI output that enters a workflow demands downstream human review. Enterprises must impose quality gates, human-in-the-loop validation, and output auditing. Without these systems, the cost line will rise while the value line stays flat .

3. Workforce Strategy Must Plan for Simultaneous Cutting and Upskilling. CBA's model is instructive: AI reduces headcount in some teams while the reskilling investment creates new career paths for others. The A$90 million program signals that the alternative to layoffs is not job protection but job transformation — and that companies bear responsibility for guiding their workforces through that transition .

Comyn's core message is that corporate AI has entered a sharply harder phase. The easy gains are harvested; what remains is complex, expensive, and demanding of a discipline most organizations lack. The bill for that lack of discipline is now arriving in the form of both unpredictable token costs and a growing pile of "work slop" someone must clean up .