27 April 2026 to 3 May 2026
This week's developments highlight the practical friction in AI adoption, from regulators flagging governance gaps for autonomous agents to businesses grappling with the unglamorous reality of data preparation. Meanwhile, the economics of AI tools are shifting, with a move towards usage-based pricing that could complicate costs for developers.
Regulators eye agentic AI's governance blind spots
Australian financial regulators are flagging concerns about how firms are managing the risks associated with AI agents. A review by the Australian Prudential Regulation Authority (APRA) found that many financial institutions, as they increasingly integrate AI into everything from internal operations to customer service, lack robust governance frameworks. This oversight gap is particularly worrying given the sensitive nature of financial data and AI's growing influence on decision-making.
The push for AI adoption in finance, while promising efficiency, appears to be outpacing the development of adequate controls. The regulator's findings suggest a need for improved oversight, but whether firms will treat these warnings as more than just regulatory noise remains to be seen. The potential implications for consumer trust and compliance are significant.
The unglamorous reality of enterprise AI: data first
While executives are keen on AI, many companies are discovering that the real hurdle isn't the technology itself, but their own data. Unlike the slick, consumer-facing AI applications, deploying AI effectively within an organisation demands a far less glamorous, yet critical, focus on data management. Without a solid data foundation, the promise of AI adoption often remains just that - a promise.
The drive for scaled AI initiatives is exposing fundamental weaknesses in data quality, accessibility, and integration across enterprises. This reality is forcing businesses to confront the need for a comprehensive overhaul of their data infrastructure, a task that prioritises accuracy and availability over flashy new tools. The path to meaningful AI adoption, it seems, is paved with data reform.
Shifting economics: GitHub Copilot moves to per-token
The pricing models for AI tools are evolving, and GitHub Copilot's upcoming shift from a flat-rate subscription to a per-token charging system, effective June 1st, 2026, exemplifies this trend. This move replaces a straightforward model with one that could lead to unpredictable costs for developers, depending heavily on their usage. While potentially beneficial for high-volume users, the complexity might deter those who preferred a fixed monthly fee.
This change reflects a broader industry movement towards usage-based pricing. However, it also risks alienating users who value pricing transparency. The long-term impact on user retention and satisfaction in a competitive coding assistant market is yet to be determined.
New models and the ongoing debate around AI capabilities
April saw the release of Anthropic's Opus 4.7 and OpenAI's GPT-5.5, marking another step in the ongoing race between major AI labs. While both companies are touting enhancements, the true significance of these updates will likely emerge through real-world applications rather than benchmark scores. The competitive timing suggests an accelerated development cycle, but the claims of superiority should be met with a degree of skepticism.
Elsewhere, discussions around agentic AI continue, focusing on systems designed for autonomous operation. This evolution raises questions about control and oversight, particularly as these systems are tasked with complex objectives. Separately, the distinction between 'Agent Skills' and 'MCP' is being clarified, highlighting that these are fundamentally different technologies serving distinct purposes, rather than competing alternatives. The discourse around such advancements often leans towards spectacle, underscoring the need for a discerning approach.