The way businesses and professionals interact with artificial intelligence has changed. Over the past two months, we have observed a shift across our client base and the broader regulatory and legal community that goes beyond curiosity or experimentation. Professionals who were previously sceptical are now actively engaging with AI tools. Those who were already experimenting are finding that the tools have become materially more capable than they were even six months ago.
The cause is straightforward: AI models have increased in capacity and usefulness at a pace that has caught many off guard. The implications for energy retailers, compliance professionals, and every individual working in a regulated industry are worth examining carefully.
AI capability is advancing rapidly and unevenly. The organisations and individuals who experiment safely and strategically now will be best placed to take advantage of the opportunities that emerge.
An Unprecedented Pace of Change
Since the release of ChatGPT in November 2022, we have witnessed intense competition amongst the leading AI providers, including Google, Anthropic, and OpenAI. The pace of development has been remarkable. At the leading AI companies, the vast majority of code is now AI-generated. An Anthropic spokesperson told Fortune that between 70% and 90% of the company’s code is now written by AI, whilst individual engineers at both Anthropic and OpenAI have reported that figure as 100%. Microsoft’s CEO Satya Nadella noted in April 2025 that AI was generating about 30% of Microsoft’s code. The gap between the frontier AI labs and other technology companies is notable, but the direction of travel is clear.
When you compare the output of ChatGPT at its initial release to the output of current models, the difference is striking. However, it is important to understand that AI capability has advanced unevenly across different domains. Researchers at Harvard Business School, in collaboration with Boston Consulting Group, have described this as the ‘jagged technological frontier’. One AI tool may be excellent at solving logical problems but produce unreliable text. Another may write fluently but struggle with numerical reasoning. The frontier of AI capability is not a smooth line; it is uneven and sometimes surprising.
Despite this unevenness, the direction of advancement is unambiguous. We are now seeing large-scale workforce impacts directly attributed to AI. In February 2026, Australian logistics software company WiseTech Global announced plans to cut approximately 2,000 jobs (roughly 29% of its global workforce) over two years as it integrates AI across its operations. WiseTech’s CEO Zubin Appoo stated that ‘the era of manually writing code as the core act of engineering is over.’ In January 2026, Amazon announced 16,000 redundancies as part of a broader restructuring linked to AI adoption and efficiency gains, following 14,000 cuts in October 2025. These are not isolated examples. According to tracking by Layoffs.fyi, more than 123,000 technology workers lost their jobs globally in 2025, with AI-driven automation a contributing factor in a material proportion of those redundancies.
The World Economic Forum’s 2025 survey found that more than 41% of companies worldwide expect to reduce headcount over the next five years as AI integration deepens. This is not a trend limited to the technology sector.
Company Policies Are Being Ignored
There are varying degrees of AI adoption across organisations. In many businesses, particularly in regulated industries, the policy is that either generative AI cannot be used at all, or its use is limited to a single approved product such as Microsoft’s Copilot. The reasons for these restrictions are understandable: concerns about data sovereignty, confidentiality, and the risk of hallucinated or fabricated outputs.
The trade-off, however, is that many of these organisations are not using state-of-the-art tools. And the reality on the ground is different from what the policies dictate. Individuals will, whenever possible, look to more advanced tools to make their work more efficient. It is likely that most people reading this article have already used AI tools in their personal lives or to assist with aspects of their work, regardless of what their employer’s policy says.
Consider the compliance officer at an energy retailer who is restricted to using Copilot for drafting internal documents, but privately uses a more capable model to help structure a regulatory response. The individual benefits from the better tool, but the organisation has no visibility over the quality controls, data handling, or accuracy of that work. This gap between policy and practice creates risk, and it is a problem that organisations need to address directly rather than pretend does not exist.
The Individual Opportunity
For individuals, the most important first step is experimentation. This does not mean uploading confidential customer data into a public AI tool. It means developing a deliberate approach to testing AI capabilities using anonymous or synthetic data, in contexts where you can evaluate the output based on your own professional expertise and judgment.
What you will likely find, as we have found, is that AI excels in a number of areas but still hallucinates and makes mistakes. Sometimes those mistakes are obvious. In other instances, the errors are subtle yet persistent, and they are exactly the kind of errors that would go unnoticed by someone without deep domain expertise.
When it comes to regulatory compliance, for example, asking an AI model to interpret regulatory obligations will often produce answers that appear correct but are in fact fabricated. The model may confidently cite provisions that do not exist or misstate the effect of a provision that does. For a business operating in the energy sector, relying on the raw output of an AI model to answer a regulatory question could lead to non-conformance, regulatory penalties, or enforcement action by bodies such as the Australian Energy Regulator (AER) or the Essential Services Commission (ESC).
One of the patterns we have noticed is that as the so-called ‘cost of intelligence’ moves towards zero, the value of genuine expertise increases rather than decreases. Our direct experience interacting with regulatory bodies including the AER and ESC, our understanding of the broader market context, and our knowledge of the priorities, concerns, and constraints of individual clients all become more valuable in an environment where surface-level analysis can be generated instantly by anyone with access to an AI tool. The differentiator is no longer access to information; it is the ability to exercise professional judgment about what that information means in context.
The Organisational Opportunity
At the organisational level, businesses need to examine the realities on the ground. Many businesses are not allowing individuals to access state-of-the-art tools, and even those that prohibit their use are, in practice, ignoring the fact that individual employees are using them anyway.
The organisational opportunity presented by generative AI is substantial. However, to realise that opportunity, more than simply copying and pasting outputs from AI interfaces is required. At Compliance Quarter, we have completed considerable work developing agent-based systems with specific skills, verification methods, and defined processes. Most importantly, these systems include humans in the loop to guide outputs, to review results, and to make the judgment calls that AI cannot reliably make on its own.
Our experience has been informed by early access to OpenAI models well before they became publicly available, through direct engagement with OpenAI. That early exposure gave us the opportunity to understand both the capabilities and limitations of these tools in the context of regulatory compliance work, and to develop approaches that account for both.
Broadly, businesses tend to fall into one of three categories.
The Lumbering business is the large incumbent, slow to adopt any new technology, who risks falling behind more nimble competitors. New entrants who are unencumbered by legacy systems and institutional inertia will be more able to adopt and take advantage of AI tools.
The Sprinting business is the organisation that can adopt quickly, consolidate market power, and leverage its institutional knowledge and competitive advantages to make the most of generative AI. These businesses treat AI not as a threat but as an accelerant.
The Sleeping business is the organisation that does not have a sufficient understanding of what is happening and what it means. By the time these businesses recognise the shift, the competitive landscape may have already moved past them.
At this stage, experimentation within safe boundaries is the most important step any organisation can take. If you would like to discuss how your business can conduct safe experiments and explore what is possible with AI today, please get in touch.
Be Wary of Over-Hyped AI Solutions
Businesses should also exercise caution with vendor claims. We expect that over the coming 6 to 12 months, a number of system providers, including large customer management system vendors, will claim that their AI capabilities can solve various operational pain points. A number of those claims will not withstand scrutiny.
We have already seen examples of this in the compliance space. Providers have promised automated compliance review capabilities which, upon inspection by actual experts, proved to be non-compliant with the very regulatory frameworks they claimed to address. The underlying AI may produce outputs that look professional and well-structured, but the substance may be wrong in ways that only a domain expert would recognise. Before committing to any AI-powered compliance or operational tool, insist on independent expert review of the outputs. Test the system against scenarios where you already know the correct answer. Do not rely on vendor demonstrations alone.
What Comes Next
The pace of change is not slowing down. In February 2026 alone, Anthropic, Google, and OpenAI each released major updates to their flagship models, coding tools, and agent platforms. The five largest US cloud and AI infrastructure providers have collectively committed to spending more than US$650 billion on capital expenditure in 2026, nearly doubling 2025 levels.
For energy retailers and regulated businesses more broadly, the question is no longer whether AI will affect your operations. It already is, whether through the tools your employees are using independently, the AI-driven capabilities being embedded in the systems you rely on, or the competitive dynamics being reshaped by organisations that are moving faster.
The organisations that will thrive are those that embrace uncertainty, experiment deliberately, invest in genuine expertise, and maintain the human judgment that AI cannot yet replicate.
We’d welcome your thoughts on this topic, including:
- what you’re observing
- what tools you are finding useful
- what risks have materialised
- where you think this is heading
For assistance navigating the intersection of AI and regulatory compliance in the energy sector, contact our team. To discuss the legal implications of AI adoption for your business, reach out to Law Quarter.
References
Dell’Acqua, F. et al. (2023). Navigating the Jagged Technological Frontier. Harvard Business School Working Paper No. 24-013.
Fortune (29 January 2026). Top engineers at Anthropic, OpenAI say AI now writes 100% of their code.
Reuters (25 February 2026). AI disruption prompts Australia’s WiseTech to cut a third of global workforce.
CNBC (28 January 2026). Amazon layoffs: 16,000 jobs to be cut in latest restructuring.
MIT Technology Review (15 December 2025). AI coding is now everywhere. But not everyone is convinced.
SWE-bench Verified Leaderboard. swebench.com; Epoch AI evaluations.
