AI people on X: ideas, agents and working software
A model demo, a working app and a research paper answer different questions. Meet the people behind them, inspect what they made and choose a useful starting material.
Start with a question
A useful first work is a better starting point than a long list of names.
7 reading tasks
Build software with AI
Follow product prototypes, coding-agent workflows and the practices used to inspect and test generated code.
Builders & productsUnderstand language models
Explore explanations, training systems and reasoning studies, then choose an original paper or implementation.
Foundations & explanationsDelegate a multi-step task
Connect models to tools, preserve execution state and inspect how an agent checks the result.
AI agents & automationMake an AI application dependable
Work with context and documents, build useful evaluation sets and trace quality, latency and cost.
Applied AI engineeringRun a model on your own hardware
Inspect local runtimes, model adaptation and the memory and performance choices behind inference.
Local models & open projectsUnderstand attacks and data boundaries
Read incident reports, permission models and experiments that examine model behavior.
AI security & interpretabilityCreate images and video
Explore original media work, generation code and ways to direct a result across successive revisions.
AI for creative workExplore a reading list
All AI topicsMeet different approaches to the same subject.
- Vibe coding & coding agentsMeet people building with LLMs: product demos, coding-agent workflows and practical ways to understand and check generated code.10selected people
- LLM research & explanationsFind LLM researchers through their original work: Transformer explanations, training recipes, model systems, reasoning and evaluation. Choose a useful first read.12selected people
- AI agents & automationFind people building and studying AI agents: tool use, workflow architecture, persistent state and controls over actions. Open a specific original material first.7selected people
- Applied AI engineeringRead people building AI applications around useful data, product-specific evaluations and observable systems. Each selection starts with a concrete original work.7selected people
- Local models & open projectsFind people who explain running AI on your own device, choosing and adapting downloaded models, and understanding memory and speed.6selected people
- AI security & interpretabilityRead people investigating prompt injection, data leaks, tool permissions and model internals, with an original report or explanation to open first.7selected people
- AI for creative workFind authors who explain creating and editing images and video: visual instructions, reference images, consistent changes and assembling generated clips.6selected people
-
Builds open-source learning materials for Solidity and ethers.js, and works on xAPI, an API gateway for AI agents.
Explore their work -
Practical ways to plan, inspect and test software made with coding agents, from a small change to a team's development process.
Explore their work -
The product and infrastructure thinking behind Replit's move from a coding editor to building software through natural language.
Explore their work -
Explains how language models change software development, from his original description of vibe coding to experiments where an agent changes training code and measures the result.
Explore their work -
Builds bullish token theses around AI products and app incentives. His KLED article connects data collection, user rewards and token demand for an asset he says he holds.
Explore their work -
Explains the design of a coding agent and shares how he uses it in everyday work, including project instructions, reusable commands and ways to verify a change.
Explore their work
Guides
All AI guidesContext for reading the people and conversations on X.
Twitter shadowban test: check search, replies and labels on X
Test an X (Twitter) shadowban yourself: search suggestions, from: searches and replies from a second account, X's own labels, and innocent causes to rule out.
From a vibe-coded prototype to a checked change
Use a small search-filter bug to connect a requirement, code walkthrough, failing case and focused fix, with work from Simon Willison and Addy Osmani.
How to read an LLM research result: a worked example
Read a Tülu 3 experiment from table to conclusion: compare training settings, interpret the score and keep the limits, with useful authors to read next.