For Jaci Badenhorst CA, the lightbulb moment came in January 2023. “I was listening to a podcast explaining ChatGPT and I will never forget it – suddenly realising the implications of this technology and what it could potentially do. My life’s never quite been the same since,” he says.

Badenhorst became an AI evangelist overnight. “I spent the next few months telling people in finance and accounting: you need to find out about this; you need to pay attention – this is going to change everything.”

He’s not wrong. Wolters Kluwer surveyed 100 UK accountants in March this year, revealing that 91% use AI or plan to use it in the near future, with a 29% higher adoption rate among 18 to 24-years-olds than those aged 55 or above. Looking ahead, 38% believe the technology would unlock efficiency and time savings, while 37% say it would help them to offer better advisory services. 

This is why last year ICAS formed its AI working group, which reports directly to ICAS Council. Badenhorst was one of several people instrumental in spawning the idea and he now has a strong team around him. “It’s good to have a variety of people who’ve moved into technology and data, like me, but are accountants by background, as well as others who are involved in the space and can see the impact of it,” he says.

The working group exists to provide member support, training and ethical frameworks to ensure responsible AI implementation, as well as to help to shape the profession more broadly. The group met recently to discuss strategic objectives and how it can best support members, making sure they have the materials and information they need.

“We’re not quite there yet, we need to do more,” says Badenhorst. “For example, how do we support people who are sole practitioners and in smaller businesses? Because I think it’s a big burden to try and implement technology into these firms, and it can be difficult to keep up with all the latest changes with this technology, especially if you are tight on time.”

But Badenhorst says self-learning will also be key, given the rapid speed of transition: “Change is happening around us in ways we don’t necessarily realise, and members need to be willing not just to use AI but to teach themselves – it sounds harsh but those that don’t put in the effort will be left behind.”

Agentic autonomy

The speed of progress within AI is both alarming and exciting, and the latest step in this evolution is agentic AI. Unlike simple automation that follows fixed rules, an agent can decide which task to tackle next, extract information from multiple systems and compile text that explains its findings — before a human has even opened the file.

Agents’ capabilities can span everything from autonomously researching complex issues and reviewing financial records to identifying anomalies and preparing initial audit reports.

Badenhorst says: “Where these large language models (LLMs) and agentic systems become interesting is that they can leverage reasoning and understanding to execute judgement. Usually, you would analyse the information and, based on your analysis, execute several actions and use specific tools; these agentic systems are trying to replicate that.”

In May, Deloitte launched its agentic network to enable businesses to seamlessly integrate AI agents, allowing it to deliver AI solutions that boost client productivity and efficiency. KPMG, EY and PwC have also announced similar plans to integrate agentic into their operations this year.

While the opportunities are clear, Badenhorst also offers a word of caution: agentic systems aren’t a quick fix for a flawed process. He cites a hypothetical example of a plane that flies from Edinburgh to London to Glasgow just to get to Glasgow. “Don’t put a fancy thing on it to make that flight more efficient and quicker but still fly via London,” he says. “You’ll probably be opening yourself up to all sorts of other unintended consequences and problems. Once the process is optimal, then apply an agentic system.”

“The responsibility remains with the professional at the start and end of the process. You still need to maintain an element of oversight and control – just because the car is steering itself, essentially you’re still driving it”

Jaci Badenhorst CA, NatWest and ICAS

In April, Google introduced an “agent interoperability protocol”, allowing AI agents to communicate and operate across different company apps, clouds and data stacks without friction. The protocol means AI agents can work across cloud environments and automate repetitive and low-value workflows, such as admin and basic analysis, integrating those agents into existing tools. It is supported by Accenture, Deloitte and KPMG.

But practitioners must remain vigilant about agent-to-agent systems. “These systems require updates and upgrades, and the way they behave after these events are significantly different,” says Badenhorst. “If you end up with an ecosystem of multiple agents running processes that you rely on and something happens with the technology, that could potentially destroy what you’re trying to do.”

In short, accountants remain accountable. “The responsibility remains with the professional at the start and end of the process. You still need to maintain an element of oversight and control – just because the car is steering itself, essentially you’re still driving it.”

Good vibrations

The need for humans to retain ultimate control and responsibility is also pertinent to the recent shift towards “vibe coding”, which IBM describes as “a fresh take in coding where users express their intention using plain speech and the AI transforms that thinking into executable code”. This allows people without coding skills to describe what they want to achieve, while an LLM such as ChatGPT writes the code.

As Badenhorst says, “The danger of vibe coding is that it gives the impression that you can now be an expert coder without understanding a single line of code. But if you don’t understand what the code is doing and how it’s been built, you won’t have the knowledge to fix any problems that arise.”

“If you’re good at something, agentic AI can
accentuate that. If you’re not good at it, the tech
might give you a plausible answer but it won’t necessarily be the right answer”

He says he has started to see a trend for “vibe accounting”, which follows a similar vein: “It’s people saying, ‘Oh, I don’t need accountants, I can just feel my way through using AI.’ I think having the domain expertise and understanding is going to become absolutely critical.”

Badenhorst adds, “If you’re good at something, agentic AI can accentuate that. If you’re not good at it, the tech might give you a plausible answer but it won’t necessarily be the right answer.”

Black and white

As AI continues its rapid evolution, the accountancy sector has arguably been slower than other industries to embrace the technology. Badenhorst believes this is due to the risk of “hallucinations” – the false or misleading outputs generated by AI – in an industry which is relatively binary. 

“In accounting, there are certain things where there is some kind of judgement, but it’s generally right or wrong – there’s no grey area. I think that has caused adoption to be slightly slower, which has been a saving grace; we’ve not wasted time or money on experimenting, and now the systems have really improved.”

As AI grows in sophistication, the industry will have to adapt the way it works. “When Excel first came out, the way accountants worked changed almost overnight. This will demand a similar shift,” says Badenhorst.

While he is clear about the need for caution, Badenhorst remains super-excited about the potential of AI within the accountancy sector. “For me, it’s probably one of the most exciting times of my career.”