Who I rate

Andrej Karpathy

Tesla to OpenAI to Anthropic, and still the clearest build-from-first-principles voice in AI - his overnight autoresearch loop beating years of his own manual tuning is the note I keep returning to, and what my whole autoresearch system is built on.

Start with: Auto research proves recursive self-improvement already works ↗

Benedict Evans

The calm analyst register this site borrows from - one idea per page, no shouting.

Start with: AI is still at its 1997 moment ↗

Ethan Mollick

The best writer on actually working with AI rather than talking about it.

Start with: Detecting the secret cyborgs ↗

Simon Willison

Builds in public at a pace nobody matches, and writes it all down.

Start with: LLMs in 2024 ↗

Dario Amodei

Anthropic's CEO and the clearest picture of where the frontier is heading from someone building it - a risk-aware optimist, not a doomer. The essays are long, dense and worth every minute.

Garry Tan

YC's president treats personal AI as an operating system, not a chat window - skills, a huge context brain, a compounding loop where every fix bakes back in. The best map of what an individual's AI stack should look like.

Diana Hu

The 'tokenmaxx, not headcountmaxx' doctrine - spend on tokens not heads, charge per outcome not per seat. The org-chart version of the AI-native company, and a thesis my notes keep returning to.

Pete Koomen

His 'horseless carriages' essay reframed AI product design for me - most AI apps fail because they bolt AI onto old interfaces; the fix is letting people write their own system prompts.

Start with: Horseless carriages ↗

Boris Cherny

The creator of Claude Code - on why coding is already solved for him, why loops and routines are the future, and why software gets democratised like literacy did after the printing press. Straight from the person who built the tool I use daily.

Tom Blomfield

Monzo's co-founder, now at Anthropic - clearest on rebuilding a company as recursive self-improving loops rather than bolting agents onto old processes.

Cody Schneider

Growth marketer building AI marketing agents - sharpest on the thesis that distribution and brand are the only moats left once intelligence goes to zero, and that bottom-up adoption beats top-down mandates.

Dwarkesh Patel

The best signal-to-noise interviewer in the space - long-form conversations with the people actually building this. If you follow one podcast, this is it.

Start with: The Dwarkesh Podcast ↗

swyx

Runs Latent Space and coined 'AI Engineer' - one of the fastest ways to keep a finger on what practitioners are actually shipping.

Start with: Latent Space ↗

Steve Yegge

The most useful frame I have read on AI literacy - three cohorts defined by daily token spend, and five hours of training reliably moves a team up a cohort. Practical, not hype.

Start with: The flat curve society ↗

Daniel Miessler

Security researcher who built Kai, a modular personal AI - and the best I have read on high-entropy content: the idea that the only writing worth making now is the dense, surprising kind an AI could not have predicted.

Greg Isenberg

Practical on AI, startups and building online - strong on vertical AI agents as cash-flowing businesses and founder-agent fit as the new founder-market fit.

Jack Dorsey

Running Block as a live experiment in replacing the org chart with an AI intelligence layer - the most ambitious internal AI bet I have read.

Start with: From hierarchy to intelligence ↗

Vas Vasuman

The barbell - every AI rollout splits into a power-user core and a silent majority. The sharpest reason adoption metrics lie, and it matched what I saw inside a PE firm this summer.

Start with: Enterprise AI adoption is a myth ↗

Pedro Franceschi

AI-first is not bolting agents onto old processes, it is refounding the company. His 'the CEO must be Chief AI Officer' is the clearest version of that.

Lenny Rachitsky

The best on how products and careers actually get built - and behind the one paid AI course I would point a non-technical builder to.

Thariq Shihipar

Anthropic engineer on Claude Code. The way he plans and runs build loops with agents - write the plan, let it run for hours, verify each step - is the backbone of how I work now.

Start with: How I plan, build and run loops with Claude Code ↗

Bill Gates

Fifty years watching where computing goes next. His essays on AI agents as the next computing platform are a builder's long view, not a pundit's - worth your time.

Arnaud Doko

Applied AI at Anthropic. His 'How we Claude Code' talk is the three-phase build loop this whole site runs on - ask, plan in HTML, build, and a verify step that checks the work before it counts as done.

Start with: How we Claude Code (Code with Claude London) ↗