Editorial Intelligence and Agentic AI
What knowledge should an AI agent be built on?
Generative AI made content creation faster. Agentic AI made the vision clear — this was where autonomous execution was heading. Practical AI brought implementation into focus. Trusted AI made confidence and governance the central argument.
Each stage answered a different market question. And each stage raised the stakes for the knowledge underneath.
This framework is grounded in a body of editorial work developed at Sage from 2023 onwards — moving through each of these stages in real time, building the narrative as the market moved.
How the narrative evolved
The move from generative to agentic AI is not just a technology story. It is a record of how market questions mature as adoption deepens — and how editorial work needs to move with them.
Generative AI
What can AI create?
The market question was possibility. Could AI write, summarise, translate, generate? The answer was yes — faster and cheaper than expected. The editorial task was to calm the hype and give organisations practical guidance.
Agentic AI
Can AI complete meaningful work on our behalf?
The vision argument came early — before the technology fully arrived. Where was generative AI heading? Towards autonomous execution: agents that monitor, decide and act without being prompted at each step.
Practical AI
How can businesses use it today?
With the vision established, the market needed implementation. Domain-specific models, embedded workflows, real adoption evidence. For finance and accounting, 25,000 professionals using Copilot daily and MTD as structural catalyst.
Trusted AI
Can we trust it with data, decisions and workflows?
As adoption deepened, governance became the central question. Authentic Intelligence — domain-specific AI with transparent training, human accountability and professional standards — became the differentiating argument.
Each stage required a different editorial argument. In the generative phase, the question was possibility — can AI write, summarise, analyse? The answer was yes, but the practical guidance needed to calm the hype: a seven-step adoption framework, a crawl-walk-run methodology, emphasis on augmentation over replacement.
In the practical phase, the argument shifted to application — how do businesses embed AI into real workflows? For finance and accounting, this meant domain-specific models rather than general tools. ChatGPT could hallucinate; a finance-trained model trained on accounting standards and compliance data could not afford to. By the time 25,000 UK accountants and bookkeepers had adopted Sage Copilot, the market question had moved again.
In the trusted phase, governance became central. The Sage Trust Label — disclosing training methods, accuracy measures and bias safeguards — was an editorial and product argument simultaneously. Authentic Intelligence was the concept that bridged trusted and agentic: AI capability combined with human judgement, built for the specific domain, not borrowed from general-purpose tools.
The progression mattered because technology narratives do not move at the same speed as technology. The market needed different arguments at each stage — and editorial work that anticipated the next question before audiences had fully formed it.
What agentic AI changes
Earlier AI tools produced outputs for a human to assess and use. Agentic AI completes sequences of tasks with some autonomy — researching, drafting, synthesising, routing, deciding — often without a human reviewing every step.
In finance and accounting, the shift is from hindsight to foresight. Traditional practice explained what happened. Agentic AI helps navigate what is ahead.
The practical evidence is already visible. AI agents scanning 3.2 billion transactions have flagged and corrected 190 million errors. Implementations have removed up to 90% of manual data entry. Financial period close processes have moved from weeks to two or three days. And Making Tax Digital — which shifted accounting from annual to continuous — created the structural conditions in which agentic AI becomes not just useful but necessary.
As Sage CTO Aaron Harris put it: “The future of AI in finance won’t be about faster answers. It’ll be about encouraging smarter, more autonomous action.”
That changes the quality requirement for the knowledge underneath. When a human edits an AI draft, they can correct gaps in context, fill in missing organisational logic, and override outputs that feel wrong. When an agent runs a workflow autonomously, those corrections don’t happen in real time.
The agent’s output quality depends almost entirely on what it has been given to work from.
The Editorial Intelligence stack
There is a progression in how organisations build on AI, from one-off instructions to operating systems that guide consistent, autonomous work.
A one-off instruction to an AI model. Useful for discrete tasks. Does not accumulate knowledge or improve over time.
A repeatable sequence of prompts and steps for a known task. More consistent than ad-hoc prompting. Still depends on the quality of the inputs at each stage.
A reusable capability — creating an editorial brief, analysing a customer signal, mapping event insight. A skill encodes a method, not just a task. It can be called by agents and applied across contexts.
A system that combines multiple skills to complete a task with some autonomy. Can research, decide, draft, route and iterate — but only as well as its underlying knowledge allows.
The shared knowledge, narratives, frameworks, rules and examples that guide the agent's work. This is what Editorial Intelligence builds. Without it, agents produce generic outputs. With it, they produce work that reflects the organisation's point of view, argument and evidence.
Most organisations are comfortable with prompts and beginning to build workflows. Fewer have developed reusable skills. Fewer still have the editorial operating system that would make an agent’s work genuinely reliable and on-brand.
The accounting sector is ahead of many because MTD forced it. Annual workflows became continuous ones. Client relationships shifted from transactional to advisory. The infrastructure for agentic AI — real-time data, continuous monitoring, structured decision rules — was built out of operational necessity before the agents arrived.
Without Editorial Intelligence
When agents operate without a structured knowledge base, the results are predictable.
- Every task starts from approximately the same base as the last
- Context is fragmented — the agent knows facts but not the argument
- Outputs are generic — they reflect AI’s general knowledge, not the organisation’s perspective
- Judgement is inconsistent — without rules and frameworks, the agent improvises
- The organisation’s point of view disappears under the pressure of fluency
The agent may produce content that is coherent and grammatically correct. But it does not sound like the organisation, does not build on previous work, and does not compound in value over time.
With Editorial Intelligence
When agents operate from a structured editorial knowledge base, the picture changes.
- Narratives are already defined — the agent knows the argument, not just the facts
- Frameworks guide decisions — the agent applies organisational logic, not just general logic
- Evidence is connected — the agent can draw on a body of proof rather than reconstructing from scratch
- Outputs are consistent with the organisation’s point of view
- Work compounds — each task adds to a growing library of reusable intelligence
This is what Authentic Intelligence means in practice. It is not just AI with guardrails. It is AI that operates from a structured understanding of the domain, the organisation’s position within it, and the evidence that supports that position.
Why this matters for Editorial Intelligence
Generative AI made content creation cheaper. That created pressure on volume and speed, and accelerated the argument that more content was always better.
Trusted AI made confidence and governance central. Organisations had to think seriously about what AI should and should not do with their data, decisions and customer relationships.
Agentic AI makes execution possible. And that raises a harder question: what knowledge, judgement and narrative logic should those agents be built on?
Editorial Intelligence is the answer to that question. The organisations that have invested in structured knowledge, clear narratives, reusable frameworks and connected evidence are the organisations whose agents will produce the most consistent and valuable work.
The Accountex argument made this explicit: AI won’t replace you. But accountants who use AI will replace those who don’t. The same logic applies to organisations. Those with better editorial operating systems will get more from their AI investments than those running the same agents on fragmented, unstructured knowledge.
What this looks like in practice
The frameworks and experiments on this site are early examples of what an editorial operating system for AI looks like — grounded in actual editorial work rather than hypothetical architecture.
The AI Maturity Curve maps the progression from AI adoption to commercial transformation, and identifies the capacity gap that agentic AI is partly designed to close. It was developed from the same practitioner research and Sage survey data that informed the narrative at each stage of the progression above.
The Editorial Brief Generator shows how a structured briefing framework can guide AI output — agents that work from clear briefs produce better drafts than agents that improvise.
The Event Intelligence workflow shows how signals captured at events can be processed, structured and fed back into a knowledge system — the kind of evidence base an agent can draw on consistently.
The Customer Signal Framework shows how customer insight can be structured into reusable intelligence rather than one-off anecdotes — exactly the kind of organised evidence that improves agent outputs.
The Product Marketing Framework shows how frameworks guide consistent positioning — when an agent has a positioning framework to work from, its outputs align with the organisation’s argument rather than defaulting to generic description.
And Editorial Intelligence and AI Search addresses a related challenge: as AI systems increasingly mediate between organisations and their audiences, the quality and structure of the underlying knowledge base determines what gets surfaced.
The consistent thread
Across all of these, the consistent argument is the same.
AI does not reduce the need for structured organisational knowledge. It increases it. Each stage of the AI narrative — generative, practical, trusted, agentic — has raised the stakes for the quality of the knowledge underneath.
Agentic AI raises them further because the consequences of poor knowledge quality are now built into automated workflows rather than visible in individual outputs a human might catch and correct.
Editorial Intelligence is the practice of building the knowledge system that makes those workflows worth running.