Working notes — Editorial Intelligence

Your AI work needs a save point

Claude and ChatGPT each remembered fragments of Editorial Intelligence, but neither held a trustworthy project state. GitHub became the shared, inspectable memory that allows the work to survive either provider.

Read →

Where editorial value is moving 11 min

The writing world I wanted to join disappeared before AI

I began in journalism and thought I might end up working for magazines. Watching The Devil Wears Prada 2 reminded me that the institutions which supported that craft were being dismantled long before generative AI arrived.

Where editorial value is moving 10 min

AI didn't create generic content. It exposed it

Generic content didn't start with generative AI — we were producing interchangeable listicles under deadline pressure years before ChatGPT existed. What AI has changed is that producing plausible copy is now free, which means the writing that used to get by on adequate prose has nowhere left to hide.

By theme

The running arguments

Most of the writing belongs to one of three running arguments. Follow whichever is closest to your problem.

Where editorial value is moving 36

Evidence and organisational knowledge 31

AI, judgement and practice 39

New to the idea?

Start here

Three pieces, in order: what is changing in the market, what I think the response is, and why I would know.

The editorial jobs are not disappearing. They are moving upstream.

Job adverts from Mintago, Nscale, Paddle and Deepgram are converging on the same shape: as AI makes production cheaper, organisations are hiring for evidence, judgement, narrative and orchestration instead. That's the market signal behind Editorial Intelligence.

9 min read 02

What is Editorial Intelligence?

Most organisations do not lack insight. They lack a system for remembering it. Editorial Intelligence is the discipline of turning organisational evidence into reusable intelligence that compounds over time — and in an era of agentic AI, that discipline has never mattered more.

10 min read 03

From journalist to editorial architect

I have had five job titles. They look like a career progression. They are actually a single argument, developed over twenty years.

8 min read

Sizing me up for a role?

Five pieces that show the working

Alongside the three in Start here, above — these show judgement under real pressure rather than opinions about AI, and two are checkable against one real employer.

I've stopped optimising for the ladder. I'm optimising for leverage. A case for careers narrowing with age is easy to make: fewer senior roles, real ageism, tenure that counts against you unless you become an executive, founder or recognised expert. I think the sharper problem is that employers are bad at recognising accumulated experience — which is a different fix.

21 Aug 2026 9 min

Prompting is becoming routing For a while the whole skill of professional AI use was asking one model well. Once the work is spread across several models, tools and price tiers, the more useful skill is deciding where a given piece of work should go — without turning the routing itself into work.

24 Aug 2026 16 min

Raising the floor is not building a ladder AI can raise the standard of early-career output while removing the work that used to build judgement. That creates two different problems: fewer entry-level roles, and credible work that no longer proves the person producing it is learning. Updated with market evidence and MIT's review of AI-era assessment.

15 Aug 2026 14 min

Give every piece a job Most content plans are lists of things to publish. A portfolio is different: fewer pieces, each with a defined role — reach, conversion, practical utility or narrative development — and no piece measured against a job it was never given.

7 Aug 2026 8 min

When the best AI model stops being worth it Anthropic's most capable model is a small share of what businesses actually buy from it. That isn't disappointment with AI — it's evidence that for a lot of professional work, another increment of raw intelligence has stopped changing the outcome, and the advantage has moved to whatever surrounds the model instead.

23 Aug 2026 17 min

Evidence in motion