The conversation around AI in finance tends to live at two extremes – the panic that automation will hollow out the profession, or the dismissiveness that insists nothing will really change. Neither is particularly useful. What happened at our recent Amsterdam event with Light‘s Jonathan Sanders was something more grounded, and more honest, than either of those positions.
Sanders opened with a provocation he called “The 500-Year Lag” – the idea that double-entry bookkeeping, born in Renaissance Florence, still forms the conceptual backbone of most enterprise finance software today. The interfaces have changed. The underlying logic hasn’t moved much.
The question worth sitting with isn’t whether AI changes the skills equation for finance. That part is settled. But where it raises the bar, and where it quietly lowers the floor, is where the real conversation starts. The distinction makes a huge difference when you’re deciding who to hire, how to develop your team, and what kind of finance function you’re actually trying to build.
The tools are moving faster than the teams
The numbers bear this out. A 2025 survey by AICPA and CIMA found that 88% of senior finance and accounting leaders expect AI to be the single most transformative technology in their field within the next 12 to 24 months. Yet only 8% said their organisation is very well prepared for it. A 2025 Citi report went further, finding that 54% of financial sector jobs have high potential for automation by 2030.
What Light articulated in Amsterdam was the operational reality behind those statistics. AI is already absorbing the execution layer of finance work: transaction reconciliation, invoice processing, variance reporting, audit trail documentation. These aren’t future scenarios. For teams running modern tooling, this is the current reality.
Sanders framed the vision behind Light’s “organic software” concept: rather than AI sitting on top of existing workflows as a feature, it sits underneath them, learning your business logic over time, catching errors before they compound, adapting when your revenue model or entity structure changes, without needing a developer or a consultant to make it happen.
The goal, as he put it, is to move finance leaders from historians to architects. Most finance teams spend the majority of their time accounting for what already happened – closing the books, reconciling, overriding. Eliminating that administrative burden is what finally creates the capacity to shape strategy rather than just report on it.
What AI can’t do
At Gapstars, this is where we think the more important conversation starts, about what AI in finance can’t change, and why that matters for how teams are built.
- AI does not understand your business. It works on data.
It doesn’t know that your Q3 always looks unusual because of a seasonal contract, or that a supplier’s payment pattern is strategic rather than a cash flow problem. That kind of knowledge lives in people, built through years of context and accumulated experience. It cannot simply be fed into a model. - AI can also be confidently wrong and that’s the more practically dangerous failure mode.
Unlike a spreadsheet error that leaves a trail, an AI-generated forecast can look entirely clean, reference the right time periods, and still contain a consequential mistake that nobody catches until it matters. Finance teams that adopt AI without building the capacity to question and validate its outputs aren’t running leaner. They’re running riskier. - And AI has no relationships.
In finance, relationships are not a soft benefit sitting on the side of the real work. They are often the work itself. When a company is navigating a difficult conversation with its bank about covenant headroom, or a board is losing confidence in a forecast, those moments aren’t resolved by a model. They’re resolved by someone who has built enough trust, over enough time, to be believed when they say: we understand the problem, and here is how we are thinking about it.
The skills that are becoming more valuable
What this means practically is that the skills gap in finance is shifting – not shrinking.
- The ability to ask the right question matters more than the ability to produce the output.
AI is very good at answering questions. What it isn’t good at is deciding which questions matter. In finance, the ability to look at a business situation and ask the right question, such as “Why is our gross margin compressing despite revenue growth?” or “Which assumption in this model is most sensitive to market changes?” is where human judgement still leads.
It is the ability to frame a problem before you solve it. The parts of finance work that require a human to understand what the numbers mean for this organisation, in this moment, with this set of people. - Interpreting numbers in context.
A model can tell you that cash runway is 7 months. It can’t tell you whether the situation is signalling a crisis or is manageable, given the specific business, its investor relationships, its upcoming receivables, and the mood of the market. Knowing when a forecast is plausible versus merely technically correct, translating financial insight into language that a board or a founder can act on. These are harder to automate and, as a result, increasingly more valuable. - Stakeholder Communication
The most underrated skills when discussing skills that are irreplaceable by AI are business partnering and the beauty of networking. AI cannot sense when a board is losing confidence. It cannot know which number to lead with in a difficult conversation. It cannot build the relationship between a finance director and a CEO that makes the hard calls possible. - Working With AI Tools (Not Just Alongside Them)
At the same time, AI literacy is becoming a baseline expectation. Knowing how to structure a prompt, when to trust an output and when to verify it, and how to build workflows where AI handles the right parts. These are learnable skills, and the professionals developing them early are already seeing a real productivity advantage. A model that generates a credible-looking forecast with a subtle but consequential mistake is more dangerous than one that fails visibly. Without someone capable of questioning and validating those outputs, a single unnoticed error can ripple through an entire organisation. - Judgement and Risk Intuition
AI can flag anomalies, but a seasoned finance professional often senses when something is off even before it is formally detected. Regulatory environments, counterparty risk, fraud patterns, and unusual transactions all rely on the ability to recognise subtle signals that only emerge through years of exposure to complex, real-world situations.
At Gapstars Finance, this is the profile we increasingly look for: professionals who combine strong financial fundamentals with the curiosity and judgment required to work effectively alongside AI.
What we’re thinking about at Gapstars
For us, the Amsterdam event reinforced something we think about in the context of our own clients and teams: the finance functions that come out of this transition strongest won’t be the ones that panicked, and won’t be the ones that waited. They’ll be the ones that were honest about what was changing, and built toward it deliberately.
That means hiring for curiosity and reasoning ability alongside technical competence. It means investing in the people who can take a complex financial position and communicate it with confidence to a non-finance audience. And it means treating AI not as a threat looming in the shadows, but as a new team member, one that handles the processing, so the humans can focus on the judgement.
The question is no longer whether AI will enter the finance function. The question is whether organisations are building teams that know how to use it responsibly.