← richmorgan.co.uk

On LinkedIn

Notes on AI-native tooling, knowledge graphs, and what delegation actually means — plus the occasional hamster. An index of what I've posted; the live feed is on LinkedIn.

Connect on LinkedIn — in/richardmorgan1972

Featured

Pinned · 27 reactions · 5 comments

I built a browser for my LinkedIn data export

Step one of updating my profile was downloading LinkedIn's data-export zip — about 30 pleasingly structured CSVs. That looked like it could fit a browsable UI, so I asked Claude to build one: a Flutter web app that parses the zip entirely client-side. Nothing leaves the tab — the hosting provider only ever sees "someone loaded a static site". It has an Advisor tab that exports a clean Markdown dossier you can paste into any LLM. I still haven't updated my profile.

Try LinkedOut!

Pinned · 21 reactions · 4 comments

79% of the UK curriculum can be delivered through AI-powered adaptive learning

Not an opinion — the result of an analysis. It started watching my son use IXL and thinking this isn't very good. So I ETL'd the entire UK National Curriculum into a knowledge graph: 1,351 concepts across 55 subjects, Reception to Year 11, each classified by the mix of AI, human facilitation and specialist expertise it needs. Maths: 100% AI-reachable. Science: 99%. The bottleneck to personalised education isn't the curriculum — it's the infrastructure to deliver it.

UK Curriculum Explorer

Pinned · 110 reactions · 19 comments

Photos → medical instructions → calendar

Preparing for a colonoscopy is surprisingly involved — a 9-page instruction sheet, started days in advance, with a strict timetable of no-seeds, no-fibre, fasting and laxatives. I struggle to turn boring instructions into action, so I photographed every page and asked Claude Code to turn them into calendar events I could actually follow. By the time I arrived at the hospital, for the first time in my life, I was no longer full of it.

Pinned · 137 reactions · 20 comments

The lift looks like engineering. The plank is engineering.

Organisations are biased toward rewarding visible complexity; quiet simplification registers as a non-event. The lift doesn't just solve the problem — it creates cost, delay, maintenance, ownership, politics. Technical debt becomes headcount; headcount becomes ownership; ownership becomes self-justifying. Pre-AI you might install ten lifts a year. Now you can install ten a day. Will AI save us from our bias to overengineer — or will we just throw electricity at inefficiency and call it progress?

Pinned · 1,080 reactions · 108 comments

In a world of AI, education has never mattered more

Mass public education is a modern invention — Prussia formalised one of the early state-schooling models in the 18th century, aimed at producing literate, compliant, economically useful citizens. 260 years on, people act as if AI has made education obsolete because it retrieves facts faster. That misses the point. Education was never just about outputs — it builds the person who can judge, question, combine, reject, imagine, and decide what matters.

Recent posts

21 May 2026

AI found a 27-year-old bug in OpenBSD reply to Niels Provos

Niels Provos wrote a signed-integer bug into OpenBSD's TCP SACK implementation in 1998, and has spent the thirty years since at the top of vulnerability research. OpenBSD's whole reputation rests on nothing getting past its reviewers. The bug shipped for 27 years anyway. Anthropic's Mythos found it in a single run for about $50. If careful human review caught bugs, it would have caught this one — the more useful data point is the one about what review will and will not catch.

21 May 2026 · ~5 reactions · 4 comments

A spring clean for Gmail — ~86,000 emails deleted in an afternoon

Written in two voices — mine and Claude's — about a fifteen-year inbox purge. The point Claude kept underlining: I didn't ask it to "clean the inbox", I handed it the outcome and let it choose the method. Most people over-specify — they hand over the steps instead of the goal. Delegation isn't handing over the work; it's handing over the work you shouldn't be doing, to something that'll iterate on it without getting bored.

Read the full write-up

~15 May 2026

We can now see the private thoughts of AI

A short note on interpretability research as a genuine AI-safety breakthrough — translating a model's internal reasoning into language. I'm less bothered by AI taking human jobs than by extinction events, and this kind of work is reassuring.

~1 May 2026 · image post

Terry Pratchett, RIP

A tribute image.

~1 May 2026 · image post

The hungry caterpillar becomes a beautiful agentic butterfly

A nine-panel image riff on the agentic-AI hype cycle.

~1 May 2026 · image post

The very hungry LLM — will it eat the world?

An image post playing on Eric Carle by way of large language models.

~24 April 2026

Is this good or bad for Anthropic?

A reading of Anthropic's enterprise-share chart, written in the imagined voice of "a friend who works in an Anthropic data centre". The red line is real — Anthropic is winning the enterprise fight on merit — but predicting your position on an exponential 12–18 months out is genuinely impossible, and that's the horizon you order silicon against. On paper the best chart they've ever had; in practice probably the tightest operational window they've faced.

~24 April 2026 · sharing Oren Eini

If you're overwhelmed with AI slop, you're using it wrong

Shared alongside Oren Eini's account of a RavenDB experiment — a feature scoped at three months landing in about a week. The work doesn't go away, it moves: into reading the code, understanding it, and deciding whether it should exist in that form at all. AI amplifies whatever engineering discipline you already have.

~24 April 2026 · ~41 reactions · 20 comments

Cheap professional portraits

I needed a headshot. Instead of booking a photographer, I took a handful of dodgy selfies and fed them to ChatGPT for a synthetic photoshoot — faster, cheaper, and free of being bullied by a photographer. To Blue Steel or not to Blue Steel?

~24 April 2026

CatSnax is a deluded hamster. He is not a god.

We looked after a hamster while a friend was away — and we have cats. One day I spotted our ginger cat wandering the garden with the hamster in his mouth. I rescued him, named him CatSnax, and invented the story that got him out of his cage and onto the "garden tour". It became sample data for pushing image tools — this time ChatGPT's latest generator, which produced a genuinely decent comic page.