AI-native publishing system
Turn your publication into a living body of knowledge.
I help founders, research teams and B2B editorial organisations build publishing systems that can use AI across the whole corpus — not only to draft the next article.
The result is a structured knowledge and publishing layer that makes previous arguments, evidence, revisions and editorial rules easier to retrieve, maintain and reuse while keeping human editorial authority intact.
The shift
From publishing pages one by one to stewarding a corpus that can remember, revise and learn.
The proof so far
The Editorial Intelligence website is the first implementation. Its public corpus is versioned and structured; its underlying method and working knowledge live separately; substantial new ideas are checked against the existing body of work; AI instructions define source-of-truth and revision rules; and material updates can develop live arguments without erasing their publication history.
Editorial Intelligence Core — method, evidence, working knowledge
↓ human judgement + AI-supported routing
Editorial Intelligence website — public articles, frameworks, pages
That is real implementation evidence, not independent validation. The model has not yet been proven inside a multi-author client organisation, which is why this is positioned as a founding engagement rather than a mature productised service.
The problem is no longer only producing content
Once a publication becomes substantial, the expensive part is often coordination. Teams repeat arguments, lose research, forget why decisions were made and create new pages because returning to old work is harder than starting again.
AI can make that worse if it is used only as a drafting layer. Every model can produce more plausible content while the organisation becomes less certain about what it already knows, which source is authoritative and whether a new idea actually adds anything.
Context is fragmented
Research, decisions, rejected ideas and previous editorial judgement live across chats, documents and individual memory.
The CMS stores pages, not reasoning
A normal publishing system can tell you that an article exists. It does not usually tell you whether today's idea should extend, correct, merge, split or create.
AI has no stable operating model
Teams repeatedly re-prompt the same source-of-truth, revision, evidence and review rules instead of making them part of the environment.
Published work becomes disposable
Material that could develop over time goes stale because revisiting, checking and republishing it carries too much friction.
What I can build
The exact implementation follows your existing stack. The aim is not to copy the Editorial Intelligence website. It is to reproduce the operating principles that make the system useful.
Knowledge and source-of-truth architecture
Separate working knowledge, evidence, decisions and private context from the public delivery layer, with a clear owner for each type of information.
Structured publishing model
Make important content machine-readable through stable slugs, publication states, revision dates, controlled metadata and explicit relationships where they add value.
AI operating instructions
Document where truth lives, how approved models should retrieve context, what they may change and when human review is required.
Continuous publishing route
Introduce a practical decision before substantial new work: extend, correct, merge, split or create rather than defaulting every signal into another page.
Review, provenance and version history
Make consequential AI-assisted changes inspectable, attributable and reversible through Git, CMS revisions or another suitable mechanism.
Implementation and handover
Adapt the existing CMS where it can support the model, or build a lightweight repository-driven publishing layer when the current platform is the constraint. Leave the team with documented workflows it can own.
How a founding engagement works
Audit the existing corpus and workflow
Identify what is already published, where the underlying evidence and working knowledge live, how AI is currently used and where repetition or maintenance is breaking down.
Design the smallest useful architecture
Define the source-of-truth layers, content structure, access boundaries, AI instructions and one publishing route worth testing first.
Build and test it on real work
Implement the agreed changes and run a live editorial case through the system so the operating model is tested against actual content rather than a demo.
Handover the rules, not just the software
Document what the team owns, what AI can do, what requires a human gate, how revisions work and how the system should improve from future use.
Initial founding builds are intended to be fixed-scope pilots, likely around three to six weeks depending on the existing stack and corpus. Scope and fee are agreed before work begins.
What this is — and what it is not
It is
- an editorial and knowledge-system build;
- a way to make AI useful across an existing body of thought;
- a model-agnostic operating layer;
- a practical route for keeping published work alive;
- a system with explicit human editorial authority.
It is not
- a high-volume AI content factory;
- a generic website redesign;
- autonomous publishing without approval;
- a requirement to use GitHub, Astro, ChatGPT or Claude;
- a replacement for subject expertise or editorial judgement.
Who this is for
The strongest early fit is an organisation that already has enough published and working knowledge for corpus-level problems to be real.
- Founder- or expert-led B2B publications with a growing archive.
- Research and thought-leadership teams that repeatedly reuse evidence across formats.
- Specialist consultancies and professional-services firms whose expertise is scattered across people and documents.
- Small editorial teams already using AI seriously but relying on chats and individual memory to hold the system together.
- B2B SaaS and fintech teams with substantial research, customer evidence and editorial history.
A brand-new website with very little content probably does not need this yet. It is more likely to need positioning, evidence and an ordinary publishing system first.
Founding engagement
If your publication knows more than your publishing system can use, this is the problem I want to work on.
Send me the site, the workflow and the part that keeps breaking. I can help work out whether the answer is a lighter operating layer on top of what you already have or a more substantial rebuild.
Discuss a founding build