Generative AI

This narrative documents the first stage of the AI editorial progression developed at Sage — the generative AI phase, roughly 2023. It is the starting point of a sequence: Generative AI → Agentic AI → Practical AI → Trusted AI.

The problem

When generative AI arrived in public consciousness — with ChatGPT’s release in late 2022 and the months of coverage that followed — most organisations faced the same editorial challenge.

They needed to say something. But what they said mattered. Say too little and you look behind. Overclaim and you lose credibility when the technology doesn’t deliver. Explain the wrong thing and you create confusion rather than confidence.

The market question was: what can AI create?

The answer was demonstrably yes — faster, cheaper and more fluently than expected. But that answer immediately raised harder questions. Could it be trusted? Where did human judgement still matter? What did this mean for finance teams, accountants, CFOs who had spent careers building expertise that AI seemed to replicate in seconds?

The editorial challenge was not to explain the technology. It was to help organisations think clearly about what to do with it.

The narrative thesis

Generative AI required an augmentation argument, not a replacement argument.

The organisations that communicated well in this phase were those that named the genuine capability — productivity gains of 30–40%, faster drafting, better summarisation, improved scenario modelling — while being honest about the limits. Hallucination was real. Domain-specific accuracy was uneven. The gap between what general-purpose AI could generate and what a finance professional could trust was significant.

The right argument in this phase had three parts:

AI augments, it does not replace. The human remains the expert. The AI is a capable collaborator that removes friction from known tasks — drafting, summarising, translating, formatting. The judgment, the accountability and the final call remain with the person.

General-purpose tools are not enough. For finance, accounting and compliance-critical work, the risk of a hallucinated figure or a plausible-sounding but incorrect answer is not a minor inconvenience. Domain-specific models, trained on accounting standards and compliance data, were the credible path — not ChatGPT applied to financial statements.

Start with the customer, not the capability. The strongest communications in this phase were those that began with the problem the customer was trying to solve, then showed how AI helped solve it — not the other way around. Capability-led content aged badly. Customer-need-led content remained useful.

The content that came out of it

The generative AI phase produced a specific kind of editorial output: practical guidance aimed at reducing anxiety and increasing confidence.

“Generative AI in 7 Easy Steps” was a deliberate editorial choice — a crawl-walk-run framework that gave finance and accounting professionals a structured way to think about adoption. It was not a technology explainer. It was a decision framework.

“Beyond ChatGPT: Unlock the Financial Potential of Generative AI” made the domain-specificity argument directly. Generic AI could produce plausible text. Finance-grade AI needed to produce accurate, compliant, auditable outputs. The editorial position was clear: general-purpose tools are a starting point, not a destination.

“AI, Finance, ChatGPT and Generative AI for CFOs” addressed a specific audience with specific concerns. CFOs were not asking whether AI was real. They were asking whether they could trust it, what their liability was, and what their boards would ask. The editorial task was to answer those questions directly.

The consistent thread across all of it: calm the hype, name the real capability, give people something actionable.

Signals from this phase

The productivity number was credible, the replacement fear was not The 30–40% productivity gain estimate landed well because it felt realistic rather than utopian. The replacement narrative did not land — professionals who understood their work knew that the complexity of client relationships, regulatory judgement and professional accountability was not something a language model was going to take over.

Hallucination was the central trust problem Every conversation about generative AI in finance eventually reached the same point: but what if it’s wrong? The answer — human review, domain-specific training, accuracy safeguards — was necessary but not sufficient. It set up the trusted AI phase that followed.

The audience moved faster than expected By mid-2023, accounting professionals were already experimenting with AI tools personally, even when their firms had not yet adopted a position. The editorial challenge shifted from explaining AI is coming to helping people who are already using it use it better.

What this phase set up

The generative AI narrative created the foundation for what followed.

It established augmentation as the frame — AI as a capable collaborator, not a replacement. It named the domain-specificity problem — generic tools were not enough for high-stakes professional work. And it created an audience that was curious, experimentally active, and increasingly ready to ask the next question.

That next question was practical: how do businesses actually embed this into their workflows?

See Agentic AI for how the narrative evolved from here — the vision argument that followed the generative moment.

Why this narrative matters

The generative AI narrative matters because it documents what responsible editorial work looks like when a genuinely new technology arrives — and the pressure to say something fast is high.

The instinct in most organisations is to be first: first to comment, first to claim expertise, first to publish. The editorial challenge with generative AI was different. The technology was real, the implications were significant, and the audiences — finance professionals, accountants, CFOs — had legitimate concerns about accuracy, liability and professional judgement. Being fast without being grounded would have damaged credibility rather than built it.

The approach that worked was to slow the argument down relative to the hype: to acknowledge what AI could do, to be specific about what it could not, and to give people a practical frame rather than a capability claim. That is harder editorial work than producing another explainer. But it is the kind of work that ages well — and that audiences return to when the noise has subsided.

The broader point is about the relationship between editorial judgement and technology adoption. Every significant technology shift produces the same pressure: say something confident, say it now, align yourself with progress. Editorial Intelligence, applied well, does the opposite — it finds the argument the evidence actually supports, even when that argument is quieter than the moment demands.

Articles from this phase

Published at Sage during the generative AI phase — the editorial work that built the argument in public.