Indy Nagpal
Writing Speaking About

Kinarey: Notes from the Edge

24 June 2026 ·

You're getting this because you signed up at kinarey.com. If someone forwarded it to you, welcome — you can subscribe at the bottom.

I've been writing on kinarey.com for a while now — essays on AI, sovereignty, infrastructure, the quiet shifts in how organisations work. Starting now, I'm planning a short email once or twice a month — a few things I've published, a few things I've found worth reading, and occasionally a thought that didn't make it into a full essay. No filler. If you have thoughts on any of it, or want to see more (or less) of something, just hit reply — I read every one.

If you find any of this useful, forward it to someone who'd appreciate it.

Recent writing

As organisations scale their use of AI, the structural questions get harder. Who controls access to the models you depend on? Who bears responsibility when autonomous systems cause harm? Where does confidence tip into blind trust? These four pieces explore what it takes to build with AI seriously — not whether to, but how to do it without losing sight of what matters.

The Control Plane

A developer in Melbourne wakes up to find her AI agent returning 403 on every call. A US government directive killed the model overnight. Her data never left Australia. The model that reasoned over it was American. That was the only fact that mattered.

Read →

Nobody Did It

When an algorithm denies care, kills a pedestrian, or shapes a child before birth — and nobody can be found who made the decision. On AI and the responsibility vacuum.

Read →

The Hallucination Architecture

AI is a disembodied prediction engine optimised for compliance, not truth. On hallucination, sycophancy, and what we lose when friction is engineered away.

Read →

The Confidence Trap

AI is reshaping what technical leaders are for. Six patterns of response — each delegating the thinking somewhere different. The trap is not AI. It is the certainty.

Read →
Five things worth your time
1. Europe built sovereign clouds — then forgot about the processors — The Register

Europe spent billions on sovereign cloud infrastructure. But the chips inside those servers still have hidden management engines that US law can compel access to. The sovereignty stops at the silicon.

2. Why AI bills rise as costs fall — Exponential View

Token prices have dropped dramatically. But company AI budgets are going up, not down. Cheaper AI means people use more of it — more agents, more calls, more invisible background work that nobody tracks.

3. Anthropic blocks international users as US labels top AI models a national security risk — Interesting Engineering

The US now treats its most capable AI models as controlled technology. Anthropic cut off international access overnight. If your business depends on frontier AI, your supply chain just became a foreign policy question.

4. 60% of US consumers say 'AI' in brand messaging is a turnoff — WP VIP

Companies are racing to put AI in their marketing. Consumers are racing to avoid it. A survey of what happens when the industry's enthusiasm meets the public's actual feelings.

5. PsychAdapter: personality in LLM output via trait-language patterns, not prompts — Stanford HumanLab · GitHub

Stanford researchers found they could make an AI write like a specific personality type — not by telling it to, but by rewiring part of the model itself. They mapped how real people with different personality traits actually use language, then built those patterns into the architecture.


If someone forwarded this, get your own copy.
View in browser →

Read in 50 countries by the people who decide how AI gets deployed.

Subscribe for new essays — a few times a month, no spam.

← The Control Plane The Efficient Frontier →

Joined by founders, CTOs, and thinkers from 50 countries.

Subscribe for new essays — a few times a month, no spam.

Read in 50 countries

↗ LinkedIn

© Indy Nagpal · Kinarey Ltd · 2026