Resources

Author

Scott Murff

Published

September 2, 2026

Curated reading, courses, and accounts for becoming an AI product builder, organized by the six Builder axes.

How to use this. Do not work through it top to bottom. Find your lowest axis on your baseline profile, start with the one item marked Start here, and come back for the rest when you need it.

Verified August 2026. Anthropic moved several documentation URLs during 2025 and 2026, so older links you find elsewhere may redirect or break.

Discovery

What to build, and for whom.

Resource	Format	Why
Start here: The Mom Test, Rob Fitzpatrick	Book, short	Fixes the most common failure in the room: asking questions that produce flattering lies
Continuous Discovery Habits, Teresa Torres	Book	Turns “talk to users” into a weekly team habit rather than a good intention
Prioritize Opportunities, Not Solutions, Teresa Torres	Article, free	The clearest free explainer of the opportunity solution tree
The Lean Product Playbook, Dan Olsen	Book	The cleanest operational definition of product-market fit, six steps from customer to MVP
SVPG articles, Marty Cagan	Articles, free	Free and still updated. Start with how strong product teams differ from feature teams
Shape Up, Ryan Singer	Book, free	Fixed appetite and betting instead of backlog grooming. Free full text
Lenny’s Newsletter	Newsletter	The highest density of practitioner tactics anywhere in product
Mental models for building products people love, Stewart Butterfield, Nov 2025	Podcast	The Slack and Flickr founder on how he decides what is worth building

Create your own curated study plan from these resources. Copy this into Claude Code. It reads the sources above, finds out what you already know, and teaches against your own project rather than a tutorial app.

Context first, because you do not know any of this yet.

I am a student in the BYU AI Foundry, a program that trains AI product
builders: people who can carry an idea from a user problem to a working
product that someone actually uses, with AI as both the material and the tool.

The program tracks six skills, called the Builder axes:

  1. Discovery, what to build and for whom
  2. Design, the surface and the flow
  3. Application Architecture, how the app is structured, shipped, and kept
     safe, including wiring a model in: API calls, retrieval pipelines, cost
     and latency
  4. AI Systems, how the models work and why they behave the way they do:
     neural networks, next-word prediction, pretraining, fine-tuning and
     RLHF, embeddings and RAG, context windows, hallucination
  5. Agentic Workflow, putting AI to work: coding agents, project
     instruction files, memory, skills, hooks, subagents, MCP, reviewing
     code an agent wrote, and building AI into the product with prompts,
     tool use, agent loops, and evals
  6. Launch and Learn, instrumentation, metrics, activation and retention,
     experiments, positioning and pricing

I rated myself 1 to 5 on every axis and got back a six-sided chart of my
profile. I am working on one axis at a time, starting with my weakest.

You are my tutor for Discovery: what to build, and for whom.

First, ask me at most four questions to find out where I actually am. Do not
skip this and do not ask more than four.

Then build me an ordered plan from the sources below, with a rough time
estimate for each step. Read each source yourself before you teach from it.

Sources:
https://www.momtestbook.com/
https://www.producttalk.org/prioritize-opportunities/
https://www.svpg.com/articles/
https://basecamp.com/shapeup
https://www.youtube.com/watch?v=kLe-zy5r0Mk

How to teach me:
- One concept per session, thirty minutes maximum. Say where we stopped.
- After each concept, give me one small exercise against my own project. Not a
  toy example. If I do not have a project yet, help me pick one first.
- Make me explain it back, or show you the work. If I am hand-waving, say so
  and stay on it.
- If a source is out of date or wrong, tell me. Do not summarize uncritically.
- Skip anything I already know. Prove it by asking, not by assuming.

We are done when I can state a real user problem, the evidence that it is
real, and the smallest thing you could build to test it, in one page.
When we get there, tell me plainly what I still cannot do.

Design

The surface and the flow.

Resource	Format	Why
Start here: Refactoring UI	Book	Written for engineers who need work to look designed without becoming designers
Laws of UX, Jon Yablonski	Site, free	The cognitive principles that explain why an interface works, one page each
The Definition of User Experience, Norman and Nielsen	Article, free	The foundational framing, from the people who coined the term
Handmade Designs: The New Trust Signal, NN/g	Article, 2026	Directly on the UI critique row: as AI polish becomes free, imperfection starts reading as trust
shadcn/ui	Docs	The component system agents actually write against, built to be legible to an LLM
Apple Human Interface Guidelines	Docs	Canonical reference for platform-native patterns and component behavior
The design process is dead. Here’s what’s replacing it., Jenny Wen	Article, 2026	Head of design at Claude on why discovery, mock, iterate is obsolete, and what taste means when prototyping is free
The Looking Glass, Julie Zhuo	Newsletter	Ex-Facebook VP of Design on judgment and critique culture
Why AI makes design, craft, and quality the new moat, Figma’s CEO, Oct 2025	Podcast	The argument that as building gets cheap, taste becomes the differentiator
How to build real taste, and why AI makes it matter more, Tony Fadell, Jun 2026	Podcast	The iPod and Nest creator on where taste comes from and how to develop it

Context first, because you do not know any of this yet.

I am a student in the BYU AI Foundry, a program that trains AI product
builders: people who can carry an idea from a user problem to a working
product that someone actually uses, with AI as both the material and the tool.

The program tracks six skills, called the Builder axes:

  1. Discovery, what to build and for whom
  2. Design, the surface and the flow
  3. Application Architecture, how the app is structured, shipped, and kept
     safe, including wiring a model in: API calls, retrieval pipelines, cost
     and latency
  4. AI Systems, how the models work and why they behave the way they do:
     neural networks, next-word prediction, pretraining, fine-tuning and
     RLHF, embeddings and RAG, context windows, hallucination
  5. Agentic Workflow, putting AI to work: coding agents, project
     instruction files, memory, skills, hooks, subagents, MCP, reviewing
     code an agent wrote, and building AI into the product with prompts,
     tool use, agent loops, and evals
  6. Launch and Learn, instrumentation, metrics, activation and retention,
     experiments, positioning and pricing

I rated myself 1 to 5 on every axis and got back a six-sided chart of my
profile. I am working on one axis at a time, starting with my weakest.

You are my tutor for Design: the surface and the flow.

First, ask me at most four questions to find out where I actually am. Do not
skip this and do not ask more than four.

Then build me an ordered plan from the sources below, with a rough time
estimate for each step. Read each source yourself before you teach from it.

Sources:
https://www.refactoringui.com/
https://lawsofux.com/
https://www.nngroup.com/articles/definition-user-experience/
https://www.nngroup.com/articles/handmade-designs/
https://www.lennysnewsletter.com/p/the-design-process-is-dead
https://ui.shadcn.com/docs

How to teach me:
- One concept per session, thirty minutes maximum. Say where we stopped.
- After each concept, give me one small exercise against my own project. Not a
  toy example. If I do not have a project yet, help me pick one first.
- Make me explain it back, or show you the work. If I am hand-waving, say so
  and stay on it.
- If a source is out of date or wrong, tell me. Do not summarize uncritically.
- Skip anything I already know. Prove it by asking, not by assuming.

We are done when I can take a screen an agent generated for you and name
five specific things wrong with it, each with a reason a designer would
accept.
When we get there, tell me plainly what I still cannot do.

Application Architecture

How the application is structured, shipped, and kept safe.

Resource	Format	Why
Start here: Learn Next.js, Vercel	Course, free	Sixteen chapters building a real full-stack app: routing, data, auth, deploy
Supabase docs and the auth guide	Docs	Postgres, auth, storage, realtime in one set. The auth guide covers row level security, which most tutorials skip
Designing Data-Intensive Applications, 2nd ed., Kleppmann and Riccomini	Book, 2026	The reference for how data systems really behave. The second edition rewrites the consistency chapter and adds the cloud reality
Pro Git	Book, free	Complete and free. Read chapters 1 to 3 and stop until you need more
OWASP Top 10:2025	Standard	The 2025 edition adds software supply chain failures. If a syllabus cites the 2021 list, it is stale
Web Security Academy, PortSwigger	Labs, free	Hands-on labs for the attacks OWASP describes. Reading about SQL injection is not the same as doing one
Use The Index, Luke!, Markus Winand	Book, free	SQL indexing for developers. Almost nothing else covers this well
The Pragmatic Engineer, Gergely Orosz	Newsletter	How production systems and engineering practice actually work inside real companies

Lenny’s Podcast has almost nothing on this axis. Product podcasts do not cover schema design or deploys, which is part of why it is the cohort’s weakest axis.

Context first, because you do not know any of this yet.

I am a student in the BYU AI Foundry, a program that trains AI product
builders: people who can carry an idea from a user problem to a working
product that someone actually uses, with AI as both the material and the tool.

The program tracks six skills, called the Builder axes:

  1. Discovery, what to build and for whom
  2. Design, the surface and the flow
  3. Application Architecture, how the app is structured, shipped, and kept
     safe, including wiring a model in: API calls, retrieval pipelines, cost
     and latency
  4. AI Systems, how the models work and why they behave the way they do:
     neural networks, next-word prediction, pretraining, fine-tuning and
     RLHF, embeddings and RAG, context windows, hallucination
  5. Agentic Workflow, putting AI to work: coding agents, project
     instruction files, memory, skills, hooks, subagents, MCP, reviewing
     code an agent wrote, and building AI into the product with prompts,
     tool use, agent loops, and evals
  6. Launch and Learn, instrumentation, metrics, activation and retention,
     experiments, positioning and pricing

I rated myself 1 to 5 on every axis and got back a six-sided chart of my
profile. I am working on one axis at a time, starting with my weakest.

You are my tutor for Application Architecture: how the application is
structured, shipped, and kept safe.

First, ask me at most four questions to find out where I actually am. Do not
skip this and do not ask more than four.

Then build me an ordered plan from the sources below, with a rough time
estimate for each step. Read each source yourself before you teach from it.

Sources:
https://nextjs.org/learn
https://supabase.com/docs/guides/auth
https://use-the-index-luke.com/
https://owasp.org/Top10/2025/
https://portswigger.net/web-security
https://git-scm.com/book/en/v2

How to teach me:
- One concept per session, thirty minutes maximum. Say where we stopped.
- After each concept, give me one small exercise against my own project. Not a
  toy example. If I do not have a project yet, help me pick one first.
- Make me explain it back, or show you the work. If I am hand-waving, say so
  and stay on it.
- If a source is out of date or wrong, tell me. Do not summarize uncritically.
- Skip anything I already know. Prove it by asking, not by assuming.

We are done when I can walk a stranger through your repo: what runs where,
what the schema is, who can see which rows, and what happens between your
machine and the live app.
When we get there, tell me plainly what I still cannot do.

AI Systems

How the models work, and why they behave the way they do: the understanding underneath every AI feature you build.

Resource	Format	Why
Start here: Deep Dive into LLMs like ChatGPT, Andrej Karpathy	Video, 3.5h	Pretraining through hallucination, tool use, and RLHF, pitched at builders rather than researchers
Neural networks, 3Blue1Brown	Video series, free	Neural networks, gradient descent, and transformers drawn out visually, ending with how a GPT predicts the next word
Intro to Large Language Models, Andrej Karpathy	Video, 1h	The one-hour version: what a model is, how it is trained, and why it behaves as it does
The Illustrated Transformer, Jay Alammar	Article, free	Attention and the transformer, one diagram at a time
How Models Work	Chapter	The course chapter: next-word prediction, training, and why the same prompt gives different answers
Context windows and prompt caching, Anthropic	Docs	What actually counts against the window, and the most direct lever you have on cost and latency
AI Engineering, Chip Huyen	Book	The standard reference for the job title. Free chapter notes at aie-book
AI Engineering 101, Chip Huyen, Oct 2025	Podcast	The book’s author covering the same ground in an hour

Context first, because you do not know any of this yet.

I am a student in the BYU AI Foundry, a program that trains AI product
builders: people who can carry an idea from a user problem to a working
product that someone actually uses, with AI as both the material and the tool.

The program tracks six skills, called the Builder axes:

  1. Discovery, what to build and for whom
  2. Design, the surface and the flow
  3. Application Architecture, how the app is structured, shipped, and kept
     safe, including wiring a model in: API calls, retrieval pipelines, cost
     and latency
  4. AI Systems, how the models work and why they behave the way they do:
     neural networks, next-word prediction, pretraining, fine-tuning and
     RLHF, embeddings and RAG, context windows, hallucination
  5. Agentic Workflow, putting AI to work: coding agents, project
     instruction files, memory, skills, hooks, subagents, MCP, reviewing
     code an agent wrote, and building AI into the product with prompts,
     tool use, agent loops, and evals
  6. Launch and Learn, instrumentation, metrics, activation and retention,
     experiments, positioning and pricing

I rated myself 1 to 5 on every axis and got back a six-sided chart of my
profile. I am working on one axis at a time, starting with my weakest.

You are my tutor for AI Systems: how the models work, and why they behave the
way they do.

First, ask me at most four questions to find out where I actually am. Do not
skip this and do not ask more than four.

Then build me an ordered plan from the sources below, with a rough time
estimate for each step. Read each source yourself before you teach from it.

Sources:
https://www.youtube.com/watch?v=7xTGNNLPyMI
https://www.3blue1brown.com/topics/neural-networks
https://www.youtube.com/watch?v=zjkBMFhNj_g
https://jalammar.github.io/illustrated-transformer/
https://platform.claude.com/docs/en/build-with-claude/context-windows

How to teach me:
- One concept per session, thirty minutes maximum. Say where we stopped.
- After each concept, give me one small exercise against my own project. Not a
  toy example. If I do not have a project yet, help me pick one first.
- Make me explain it back, or show you the work. If I am hand-waving, say so
  and stay on it.
- If a source is out of date or wrong, tell me. Do not summarize uncritically.
- Skip anything I already know. Prove it by asking, not by assuming.

We are done when I can explain, without notes, how a model turns a prompt into
an answer, why it sometimes makes things up, and when I would reach for RAG
instead of fine-tuning.
When we get there, tell me plainly what I still cannot do.

Agentic Workflow

Putting AI to work, as the way you build and inside the product you build. This axis moves fastest, so favor docs and primary sources over books.

Resource	Format	Why
Start here: Building Effective Agents, Anthropic	Article	The shared vocabulary of the field: workflows versus agents, patterns over frameworks
Claude Code documentation	Docs	CLAUDE.md, memory, skills, hooks, subagents, and MCP, all in one place. Note the URL moved in 2025
Anthropic Academy	Courses, free	Around 23 free courses with certificates, including Claude Code in Action, MCP, skills, and subagents
Agentic Engineering Patterns, Simon Willison	Guide, free	Six chapters on principles, testing, and annotated real prompts. The best free text on loop design
How we built our multi-agent research system, Anthropic	Article	Orchestrator and subagent architecture, when parallelism wins, and the token cost it carries
Writing effective tools for agents, Anthropic	Article	Read before you write an MCP server, not after
Model Context Protocol and AGENTS.md	Specs	Both are now multi-vendor open standards rather than one company’s idea
How far can we push AI autonomy in code generation?, Birgitta Böckeler	Article	The most rigorous treatment of where human review is still non-negotiable
Vibe coding and agentic engineering are getting closer than I’d like, Simon Willison	Article	An honest admission that review discipline is eroding, from one of its advocates
What happens after coding is solved, Boris Cherny, Feb 2026	Podcast	The head of Claude Code on where agentic development is going. The most watched episode on this list
Claude Code and Cowork, Fiona Fung, Jun 2026	Podcast	The follow-up, four months later, on how the workflow changed again
Tool use with Claude	Docs	Canonical reference for function calling and the agentic loop
Interactive prompt engineering tutorial and Claude Cookbooks	Notebooks, free	Runnable examples for RAG, tool use, evals, caching, and cost
Patterns for Building LLM-based Systems and Products, Eugene Yan	Article, free	The single best map of the field: evals, RAG, caching, guardrails, defensive UX
AI Evals email course, Hamel Husain and Shreya Shankar	Course, free	Seventeen parts plus two free ebooks. Evals are the axis item nobody has, and this is the on-ramp
Effective context engineering for AI agents, Anthropic	Article	Context engineering as the discipline that succeeded prompt engineering
Why AI evals are the hottest new skill for product builders, Hamel Husain, Sep 2025	Podcast	The best introduction to why evals matter, from the person who teaches them

Context first, because you do not know any of this yet.

I am a student in the BYU AI Foundry, a program that trains AI product
builders: people who can carry an idea from a user problem to a working
product that someone actually uses, with AI as both the material and the tool.

The program tracks six skills, called the Builder axes:

  1. Discovery, what to build and for whom
  2. Design, the surface and the flow
  3. Application Architecture, how the app is structured, shipped, and kept
     safe, including wiring a model in: API calls, retrieval pipelines, cost
     and latency
  4. AI Systems, how the models work and why they behave the way they do:
     neural networks, next-word prediction, pretraining, fine-tuning and
     RLHF, embeddings and RAG, context windows, hallucination
  5. Agentic Workflow, putting AI to work: coding agents, project
     instruction files, memory, skills, hooks, subagents, MCP, reviewing
     code an agent wrote, and building AI into the product with prompts,
     tool use, agent loops, and evals
  6. Launch and Learn, instrumentation, metrics, activation and retention,
     experiments, positioning and pricing

I rated myself 1 to 5 on every axis and got back a six-sided chart of my
profile. I am working on one axis at a time, starting with my weakest.

You are my tutor for Agentic Workflow: putting AI to work, as the way I build
and inside the product.

First, ask me at most four questions to find out where I actually am. Do not
skip this and do not ask more than four.

Then build me an ordered plan from the sources below, with a rough time
estimate for each step. Read each source yourself before you teach from it.

Sources:
https://www.anthropic.com/engineering/building-effective-agents
https://code.claude.com/docs/en/overview
https://simonwillison.net/guides/agentic-engineering-patterns/
https://www.anthropic.com/engineering/writing-tools-for-agents
https://www.anthropic.com/engineering/multi-agent-research-system
https://martinfowler.com/articles/pushing-ai-autonomy.html
https://modelcontextprotocol.io
https://platform.claude.com/docs/en/agents-and-tools/tool-use/overview
https://eugeneyan.com/writing/llm-patterns/
https://ai.hamel.dev/eval-course

How to teach me:
- One concept per session, thirty minutes maximum. Say where we stopped.
- After each concept, give me one small exercise against my own project. Not a
  toy example. If I do not have a project yet, help me pick one first.
- Make me explain it back, or show you the work. If I am hand-waving, say so
  and stay on it.
- If a source is out of date or wrong, tell me. Do not summarize uncritically.
- Skip anything I already know. Prove it by asking, not by assuming.

We are done when I can set your project up so an agent does good work in it,
review what it wrote well enough to catch what it got wrong, and put an AI
feature in your product with an eval that tells you when it is wrong.
When we get there, tell me plainly what I still cannot do.

Launch and Learn

Getting it in front of people and finding out whether it worked.

Resource	Format	Why
Start here: Product/Market Fit Survey, Sean Ellis	Tool	The 40 percent “very disappointed” test. Run it on your own product rather than reading about it
How Superhuman built an engine to find PMF, Rahul Vohra	Article, free	Turns that survey into a repeatable loop. They went from 22 to 58 percent
What is good retention?, Lenny Rachitsky and Casey Winters	Article	Benchmarks by business model. Answers the question “is our number bad”
North Star Playbook, Amplitude	Playbook, free	The standard reference for choosing a metric that would actually move
Obviously Awesome, April Dunford	Book	Positioning, which determines what activation even means
Growth Loops are the New Funnels, Reforge	Article	The essay that reframed growth from linear funnels to compounding loops
Trustworthy Online Controlled Experiments, Kohavi, Tang and Xu	Book	The A/B testing reference. Chapter 1 and the FAQ are free on the companion site
The new AI growth playbook for 2026, Elena Verna	Article, 2026	Argues AI-native products have to re-find product-market fit every few months rather than once
Finding hidden growth opportunities in your product, Albert Cheng, Oct 2025	Podcast	Duolingo and Uber growth work, concrete enough to copy on Monday
What world-class go-to-market looks like in 2026, Jeanne DeWitt Grosser, Nov 2025	Podcast	Vercel and Stripe on positioning and pricing, the two rows this axis is weakest on

Context first, because you do not know any of this yet.

I am a student in the BYU AI Foundry, a program that trains AI product
builders: people who can carry an idea from a user problem to a working
product that someone actually uses, with AI as both the material and the tool.

The program tracks six skills, called the Builder axes:

  1. Discovery, what to build and for whom
  2. Design, the surface and the flow
  3. Application Architecture, how the app is structured, shipped, and kept
     safe, including wiring a model in: API calls, retrieval pipelines, cost
     and latency
  4. AI Systems, how the models work and why they behave the way they do:
     neural networks, next-word prediction, pretraining, fine-tuning and
     RLHF, embeddings and RAG, context windows, hallucination
  5. Agentic Workflow, putting AI to work: coding agents, project
     instruction files, memory, skills, hooks, subagents, MCP, reviewing
     code an agent wrote, and building AI into the product with prompts,
     tool use, agent loops, and evals
  6. Launch and Learn, instrumentation, metrics, activation and retention,
     experiments, positioning and pricing

I rated myself 1 to 5 on every axis and got back a six-sided chart of my
profile. I am working on one axis at a time, starting with my weakest.

You are my tutor for Launch and Learn: getting it in front of people and
finding out whether it worked.

First, ask me at most four questions to find out where I actually am. Do not
skip this and do not ask more than four.

Then build me an ordered plan from the sources below, with a rough time
estimate for each step. Read each source yourself before you teach from it.

Sources:
https://pmfsurvey.com
https://review.firstround.com/how-superhuman-built-an-engine-to-find-product-market-fit/
https://amplitude.com/books/north-star/amplitudes-north-star-metric-and-inputs
https://www.lennysnewsletter.com/p/what-is-good-retention-issue-29
https://www.aprildunford.com/books
https://experimentguide.com/

How to teach me:
- One concept per session, thirty minutes maximum. Say where we stopped.
- After each concept, give me one small exercise against my own project. Not a
  toy example. If I do not have a project yet, help me pick one first.
- Make me explain it back, or show you the work. If I am hand-waving, say so
  and stay on it.
- If a source is out of date or wrong, tell me. Do not summarize uncritically.
- Skip anything I already know. Prove it by asking, not by assuming.

We are done when I can name the one metric that would move if your product
were working, say how you would instrument it, and say what number would
make you kill the feature.
When we get there, tell me plainly what I still cannot do.

Who to follow

Follow fewer accounts than you think. These are grouped by what they are useful for.

One thing worth knowing before you build a feed: Application Architecture has almost no social media presence. Nobody posts in depth about schema design, deploys, or CI. Two separate searches turned up the same answer. For that axis, feeds will not help you and the docs and books above are the only path. It is also the axis this cohort is weakest on.

For building, the highest signal for Foundry work

Account	Why
@ClaudeDevs	Official Anthropic developer updates. Where Claude Code, API, and MCP changes get announced first
@bcherny Boris Cherny	Head of Claude Code. Agents, tooling, engineering workflow
@simonw Simon Willison	Creator of Datasette, co-creator of Django. Near-daily practical posts. His blog is the better resource, organized by tag
@steipete Peter Steinberger	Built and sold PSPDFKit, now works on agents full time. The frontier of multi-agent setups. Not a first read
@emollick Ethan Mollick	Wharton. Structured experiments on what AI is actually good and bad at. One Useful Thing is the better subscription
Every: @danshipper, @kieranklaassen, @every	They ship AI products in public, Cora and Spiral among them, then write up what happened. Kieran on compound engineering is the most specific writing anywhere on agentic workflow
@HamelHusain Hamel Husain and @sh_reya Shreya Shankar	The two people who turned evals into a teachable subject. Their free course is the AI Systems start-here item
@chipro Chip Huyen	Wrote AI Engineering. Production systems, cost and latency tradeoffs
@swyx	The connective tissue of the AI engineering scene, via Latent Space
@amanrsanger Aman Sanger	Cursor founder. The one framework account that is load-bearing for this stack
@alexalbert__ Alex Albert	Anthropic, on prompting and agent behavior. The practice counterpart to @ClaudeDevs announcements

For product

Account	Why
@lennysan Lenny Rachitsky	Deeply researched product, growth, and career advice. Newsletter and podcast both worth it
Marty Cagan, SVPG	How strong product teams operate
Teresa Torres, Product Talk	Continuous discovery
Dan Olsen, dan-olsen.com	Lean product and product-market fit
@paulg Paul Graham	The essays more than the account. Written for founders building their own thing, which makes them a better fit for Build than for the Lab

For understanding where the field is going

Useful for perspective, not for shipping this week.

Name	Handle	Who they are
Sam Altman	@sama	CEO, OpenAI
Dario Amodei	@darioamodei	CEO, Anthropic
Demis Hassabis	@demishassabis	CEO, Google DeepMind
Greg Brockman	@gdb	Co-founder, OpenAI
Jensen Huang	@JensenHuang	CEO, NVIDIA. The compute side of the boom
Marc Andreessen	@pmarca	Co-founder, a16z. Investment and long-horizon takes
Andrej Karpathy	@karpathy	Researcher and educator. The clearest explainer of how LLMs work
Andrew Ng	@AndrewYNg	Founder of DeepLearning.AI. Applied ML and career advice
Yann LeCun	@ylecun	Deep learning pioneer, NYU. Strong and useful dissenting opinions
Fei-Fei Li	@drfeifei	Stanford HAI. Created ImageNet. Human-centered AI
Sebastian Raschka	@rasbt	Author and educator on LLM implementation
Lex Fridman	@lexfridman	Long-form interviews with AI researchers and founders
Logan Kilpatrick	@OfficialLoganK	AI developer ecosystem and platform
Kai-Fu Lee	@kaifulee	Investor and author. Global and China AI perspective