◈ X-Research

Primary Figures

High impact · Track closely
@karpathy Andrej Karpathy
Ex-OpenAI, ex-Tesla AI. Independent (as of Q1 2026).
Why they matter
The most credible single voice for 'this is where AI engineering actually is right now.' When he says the field shifted, people believe him. His terms become standard vocabulary (vibe coding → agentic engineering → autoresearch).
Posting pattern
Low volume, high impact. Tweets are often 'shower thought' observations that become doctrine. Self-described unpredictability: 'I still can't predict my tweet engagement basically at all.'
Signal type
  • Phase change declarations (inflection point announcements)
  • Vocabulary crystallizations (naming emerging practices)
  • Honest practitioner observations (no company affiliation = more credibility)
  • Open-source releases (autoresearch, nanogpt) that define the technical state of the art
What to watch

When he tweets about a shift, expect 60 days of newsletter discourse to follow. His content is a leading indicator: if he names something, it will be standard vocabulary soon. His GitHub releases are the most-watched in the AI builder community.

Recent signals
  • Jan 2026: 'coding agents basically didn't work before December' — 14M+ views
  • Feb 4, 2026: 'vibe coding → agentic engineering' as the new default
  • Mar 2026: autoresearch release — overnight ML research automation, 66K GitHub stars
  • Feb 2026: 1-year vibe coding retrospective — admitted it was a throwaway tweet
Reference value: Citing Karpathy's exact phrase (with attribution) signals you're reading primary sources. His vocabulary decisions are reference-worthy: use 'agentic engineering' not 'vibe coding.' autoresearch is the canonical example of 'AI as autonomous research assistant' for 2026.
@tobi Tobi Lütke
CEO of Shopify
Why they matter
'Practitioner CEO' — his statements carry both technical and business weight. When he adopts a term (context engineering), it signals the concept is mature enough for production. Shopify's scale means his AI adoption signals are credible as business reality, not demos.
Posting pattern
Moderate volume, concentrated insight. Low noise. Every post is worth reading.
Signal type
  • Vocabulary adoption signals (when he uses a term, it's legit for production)
  • Practical AI implementation takes
  • Production-readiness cues
What to watch

When Tobi says 'I've been using X for Y months,' it means the technology is production-ready. His internal memos (when they leak) are high-signal. Ran autoresearch experiments: 37 runs overnight, 19% performance gain — confirmed Karpathy's thesis.

Recent signals
  • Jun 19, 2025: 'context engineering' over 'prompt engineering' tweet — standard vocabulary shift
  • Early 2025: Shopify internal memo telling employees to justify why AI couldn't do their job first (widely circulated)
  • Follow-up: named DSPy as his 'context engineering tool of choice'
  • 2026: autoresearch testing — 19% performance gain cited as public validation
Reference value: Cite Tobi when you need a practitioner CEO as proof that something is real at scale. His vocabulary choices are safe to adopt.
@garrytan Garry Tan
CEO of Y Combinator
Why they matter
Primary signal source for YC batch stats, startup ecosystem health, and VC posture. Makes combative takes that generate debate trees. His data drops (Demo Day stats) are the most concrete ecosystem-level metrics available.
Posting pattern
Moderate volume, mix of data drops and hot takes. Sometimes combative.
Signal type
  • Ecosystem metrics (ARR, WoW growth, batch stats)
  • Sector bets (which categories YC is funding)
  • Debate-starting takes (designed to generate responses)
What to watch

Demo Day stats tweets = concrete evidence of what's real in the startup economy. His hot takes = debate tree opportunities if you have a specific counter.

Recent signals
  • Dec 2025: 'vibe coding kills Zoho' take — triggered 3-day Twitter war with Sridhar Vembu
  • Mar 2026: W26 Demo Day — '3X companies at $1M ARR, 14% WoW growth on average, fastest in YC history'
  • Mar 25, 2026: hosted ARC-AGI-3 launch fireside with Altman and Chollet at YC HQ
Reference value: YC batch stats are the most credible startup economy evidence available. Cite them with attribution when making claims about AI-enabled company formation.
@sama Sam Altman
CEO of OpenAI
Why they matter
Every tweet has outsized interpretation weight. Even vague/brief posts spawn 1000-tweet discussion trees. OpenAI's strategic posture is encoded in his rare posts.
Posting pattern
Low volume, seismic impact. Often cryptic. Each post becomes an event.
Signal type
  • AGI timeline hints
  • OpenAI product direction signals
  • Startup aspiration fuel
  • Strategic posture signals
What to watch

When Altman hints at capabilities, the discourse machine runs for weeks. Interpretation of his posts by others IS the discourse. Pentagon deal controversy (March 2026) revealed new dimension: political-commercial entanglement.

Recent signals
  • 2024: 'one-person unicorn' idea (from interviews + chat group) — year of discourse
  • Mar 2026: Pentagon deal sparks #QuitGPT — Altman's response called 'sloppy' by tech press
  • Mar 25, 2026: ARC-AGI-3 fireside with Chollet at YC — public admission GPT-5.4 scores 0.3%
  • Feb 2026: publicly praised Peter Steinberger (OpenClaw) — Steinberger joined OpenAI
Reference value: Cite Altman carefully — often what you're citing is a paraphrase or community attribution. The 'one-person unicorn' idea is widely attributed to him from interviews, not a specific tweet.
@mattshumer_ Matt Shumer
CEO of HyperWrite, serial investor
Why they matter
Pioneered the 'long-form essay that goes viral on X' format in 2026. A single essay reached 82–83M views and got a Fortune full reprint. Cultural temperature gauge: his posts reveal what the mainstream-adjacent tech audience is ready to be scared or excited about.
Posting pattern
Infrequent long-form drops with massive reach events. Not a daily poster.
Signal type
  • AI hype amplification
  • Fear-and-opportunity takes
  • 'Insider warning' essays
What to watch

When he posts a long essay, it's worth reading as a cultural temperature gauge. Counter-essays and rebuttals often follow — read those too.

Recent signals
  • Feb 2026: 'Something Big Is Happening' — 82–83M views, Fortune reprint, CNBC interview
Reference value: His essays are useful as examples of the long-form viral format, not necessarily as accurate forecasts. The 'February 2020 moment' frame is now overused — don't copy it.

Secondary Figures

Useful context
@paulg Paul Graham
YC co-founder

Archived advice gets recirculated constantly via @StartupArchive_ and goes viral regularly even when the original content is years old. Perennial founder wisdom.

What to watch

When a StartupArchive post about a certain type of advice goes viral, it signals a felt need in the community right now (anxiety about hiring, distribution, moat, etc.).

Recent signals
  • 'Sam Altman on the Paul Graham advice that saved OpenAI: Always make an API'
  • 'Sam Altman on the Paul Graham advice that not enough founders take: relentless resourcefulness'
@levelsio Pieter Levels
Solo Dutch developer. PhotoAI, NomadList, RemoteOK. ~$3M+ ARR portfolio.

The canonical 'solo founder who builds in public.' Revenue updates are directional signals for what types of indie products generate real money. His philosophy (ship before ready, no roadmaps, kill fast) is aspirational scaffolding for the bubble.

What to watch

Revenue updates are directional signals for what AI-adjacent product categories are generating real money.

Recent signals
  • Lex Fridman Podcast #440: gave him mainstream recognition for the solo founder narrative
@swyxio Shawn Wang (swyx)
Co-founder of Latent Space, builder, AI engineer

Primary community builder for the AI engineering layer. Latent Space podcast had 10M+ readers/listeners in 2025. His framing ('Software 3.0,' 'AI engineer') shapes how practitioners see their identity. Coined 'AI Engineer' as a distinct job category.

What to watch

If swyx is writing about a new approach or framework, it's about to become standard practice in the engineering-layer community.

Simon Willison (blog + X) Simon Willison
Open-source developer, LLM practitioner. No major company affiliation.

Credible, skeptical, honest practitioner voice. Coined 'prompt injection,' 'lethal trifecta' (AI agent security). Amplified 'AI slop.' Known for calling out hype vs. reality. Co-creator of Django. Creator of Datasette.

What to watch

When Willison writes critically about a technology or practice, he's usually right and ahead of consensus. If the builder community is praising something he's skeptical of, note the tension.

@bcherny Boris Cherny
Creator of Claude Code at Anthropic

Insider credibility on the most talked-about coding tool in the ecosystem. Low volume, but each post carries enormous weight. The '100% of my Claude Code contributions were written by Claude Code' thread is one of the canonical proof-of-concept drops of the agentic era.

What to watch

His workflow reveals are among the highest-signal content about agentic engineering in practice.

Recent signals
  • Dec 27, 2025: '100% of my contributions to Claude Code were written by Claude Code' — viral
  • Dec 2025: Full workflow thread — 'My setup might be surprisingly vanilla! Claude Code works great out of the box'
@AravSrinivas Aravind Srinivas
CEO, Perplexity AI

One of the most active and quotable founder accounts in the AI discourse. Perplexity topped $450M ARR; launched Computer (agent tool) and Comet browser in 2026. Will challenge anything — including Google, OpenAI, and venture orthodoxy.

What to watch

Perplexity product launches often signal what the next search/agent paradigm is. His debates with incumbents (Google, OpenAI) generate engagement and reveal real competitive dynamics.

Recent signals
  • Jan 2026: Perplexity Computer launch — AI agent for non-experts
  • Feb 2026: Comet browser launch — AI-native browser to rival Chrome
  • Mar 2026: $450M ARR milestone announced
@mntruell Michael Truell
CEO, Cursor (Anysphere)

Cursor is the company most cited as evidence that AI coding tools can generate real revenue at scale. $300M ARR by early 2026. His technical essays on AI engineering set narrative for the entire tooling space. The 'Third Era of AI Software Development' essay was reshared by hundreds of developers.

What to watch

When Truell writes an essay, it shapes how developers understand what coding tools are becoming.

Recent signals
  • 2026: 'Third Era of AI Software Development' essay — widely cited
  • 2026: Cursor hits $300M ARR, approaches $30B valuation
@amasad Amjad Masad
CEO, Replit

Replit is the poster company for vibe coding made real: $9B valuation after $400M raise in March 2026. Masad is one of the most active and contrarian founders on X. Notably skeptical of 'vibe coding' despite being credited as its enabler. Replit projects $1B revenue by end of 2026.

What to watch

His product announcements (Replit Agent 4 can maintain an entire company's codebase) are real milestones. His skepticism of his own company's narrative is signal-rich.

Recent signals
  • Mar 2026: Replit raises $400M at $9B valuation (3x in 6 months)
  • 2026: Replit Agent 4 — can build and maintain an entire company, not just apps
@gregisenberg Greg Isenberg
CEO, LateCheckout; builder, framework sharer

One of the most-retweeted builder accounts in the bubble. Amplifies AI startup ideas, 'leveraged agency' concepts, Paperclip demos. Popularized the 'AI-powered company' format threads. His content shapes what the mainstream-adjacent founder audience thinks is possible.

What to watch

When he picks up a new tool or concept, expect mainstream builder adoption within weeks.

Recent signals
  • 2026: Extensive threads on Paperclip and zero-human company concept
  • Ongoing: 'AI + [vertical]' business idea threads
@hwchase17 Harrison Chase
CEO, LangChain

LangChain is foundational infrastructure for the agent developer community. Sequoia-backed. His posts define the technical language of agent building. Published 'The Rise of Context Engineering' blog amplifying Tobi's vocabulary shift.

What to watch

His framing of what agents can and can't do is influential in enterprise AI adoption.

Primary

High impact · Track daily · 5 entries
@karpathy Andrej Karpathy
Ex-OpenAI, ex-Tesla AI. Independent (as of Q1 2026).

The most credible single voice for 'this is where AI engineering actually is right now.' When he says the field shifted, people believe him. His terms become standard vocabulary (vibe coding → agentic engineering → autoresearch).

Signal type
  • Phase change declarations (inflection point announcements)
  • Vocabulary crystallizations (naming emerging practices)
  • Honest practitioner observations (no company affiliation = more credibility)
  • Open-source releases (autoresearch, nanogpt) that define the technical state of the art
What to watch

When he tweets about a shift, expect 60 days of newsletter discourse to follow. His content is a leading indicator: if he names something, it will be standard vocabulary soon. His GitHub releases are the most-watched in the AI builder community.

Recent signals
  • Jan 2026: 'coding agents basically didn't work before December' — 14M+ views
  • Feb 4, 2026: 'vibe coding → agentic engineering' as the new default
  • Mar 2026: autoresearch release — overnight ML research automation, 66K GitHub stars
  • Feb 2026: 1-year vibe coding retrospective — admitted it was a throwaway tweet
Reference value: Citing Karpathy's exact phrase (with attribution) signals you're reading primary sources. His vocabulary decisions are reference-worthy: use 'agentic engineering' not 'vibe coding.' autoresearch is the canonical example of 'AI as autonomous research assistant' for 2026.
@tobi Tobi Lütke
CEO of Shopify

'Practitioner CEO' — his statements carry both technical and business weight. When he adopts a term (context engineering), it signals the concept is mature enough for production. Shopify's scale means his AI adoption signals are credible as business reality, not demos.

Signal type
  • Vocabulary adoption signals (when he uses a term, it's legit for production)
  • Practical AI implementation takes
  • Production-readiness cues
What to watch

When Tobi says 'I've been using X for Y months,' it means the technology is production-ready. His internal memos (when they leak) are high-signal. Ran autoresearch experiments: 37 runs overnight, 19% performance gain — confirmed Karpathy's thesis.

Recent signals
  • Jun 19, 2025: 'context engineering' over 'prompt engineering' tweet — standard vocabulary shift
  • Early 2025: Shopify internal memo telling employees to justify why AI couldn't do their job first (widely circulated)
  • Follow-up: named DSPy as his 'context engineering tool of choice'
  • 2026: autoresearch testing — 19% performance gain cited as public validation
Reference value: Cite Tobi when you need a practitioner CEO as proof that something is real at scale. His vocabulary choices are safe to adopt.
@garrytan Garry Tan
CEO of Y Combinator

Primary signal source for YC batch stats, startup ecosystem health, and VC posture. Makes combative takes that generate debate trees. His data drops (Demo Day stats) are the most concrete ecosystem-level metrics available.

Signal type
  • Ecosystem metrics (ARR, WoW growth, batch stats)
  • Sector bets (which categories YC is funding)
  • Debate-starting takes (designed to generate responses)
What to watch

Demo Day stats tweets = concrete evidence of what's real in the startup economy. His hot takes = debate tree opportunities if you have a specific counter.

Recent signals
  • Dec 2025: 'vibe coding kills Zoho' take — triggered 3-day Twitter war with Sridhar Vembu
  • Mar 2026: W26 Demo Day — '3X companies at $1M ARR, 14% WoW growth on average, fastest in YC history'
  • Mar 25, 2026: hosted ARC-AGI-3 launch fireside with Altman and Chollet at YC HQ
Reference value: YC batch stats are the most credible startup economy evidence available. Cite them with attribution when making claims about AI-enabled company formation.
@sama Sam Altman
CEO of OpenAI

Every tweet has outsized interpretation weight. Even vague/brief posts spawn 1000-tweet discussion trees. OpenAI's strategic posture is encoded in his rare posts.

Signal type
  • AGI timeline hints
  • OpenAI product direction signals
  • Startup aspiration fuel
  • Strategic posture signals
What to watch

When Altman hints at capabilities, the discourse machine runs for weeks. Interpretation of his posts by others IS the discourse. Pentagon deal controversy (March 2026) revealed new dimension: political-commercial entanglement.

Recent signals
  • 2024: 'one-person unicorn' idea (from interviews + chat group) — year of discourse
  • Mar 2026: Pentagon deal sparks #QuitGPT — Altman's response called 'sloppy' by tech press
  • Mar 25, 2026: ARC-AGI-3 fireside with Chollet at YC — public admission GPT-5.4 scores 0.3%
  • Feb 2026: publicly praised Peter Steinberger (OpenClaw) — Steinberger joined OpenAI
Reference value: Cite Altman carefully — often what you're citing is a paraphrase or community attribution. The 'one-person unicorn' idea is widely attributed to him from interviews, not a specific tweet.
@mattshumer_ Matt Shumer
CEO of HyperWrite, serial investor

Pioneered the 'long-form essay that goes viral on X' format in 2026. A single essay reached 82–83M views and got a Fortune full reprint. Cultural temperature gauge: his posts reveal what the mainstream-adjacent tech audience is ready to be scared or excited about.

Signal type
  • AI hype amplification
  • Fear-and-opportunity takes
  • 'Insider warning' essays
What to watch

When he posts a long essay, it's worth reading as a cultural temperature gauge. Counter-essays and rebuttals often follow — read those too.

Recent signals
  • Feb 2026: 'Something Big Is Happening' — 82–83M views, Fortune reprint, CNBC interview
Reference value: His essays are useful as examples of the long-form viral format, not necessarily as accurate forecasts. The 'February 2020 moment' frame is now overused — don't copy it.

Secondary

Strong signal · Read regularly · 10 entries
@paulg Paul Graham
YC co-founder

Archived advice gets recirculated constantly via @StartupArchive_ and goes viral regularly even when the original content is years old. Perennial founder wisdom.

Signal type
  • Perennial founder wisdom
  • Advice recontextualized for current moment
What to watch

When a StartupArchive post about a certain type of advice goes viral, it signals a felt need in the community right now (anxiety about hiring, distribution, moat, etc.).

Recent signals
  • 'Sam Altman on the Paul Graham advice that saved OpenAI: Always make an API'
  • 'Sam Altman on the Paul Graham advice that not enough founders take: relentless resourcefulness'
Reference value: Perennial content: cite PG aphorisms when they match current discourse tensions. The virality of a specific PG quote is itself a signal about what the community is worried about.
@levelsio Pieter Levels
Solo Dutch developer. PhotoAI, NomadList, RemoteOK. ~$3M+ ARR portfolio.

The canonical 'solo founder who builds in public.' Revenue updates are directional signals for what types of indie products generate real money. His philosophy (ship before ready, no roadmaps, kill fast) is aspirational scaffolding for the bubble.

Signal type
  • Solo founder aspiration signals
  • Indie product market validation
  • 'What you can build alone with AI' benchmarks
What to watch

Revenue updates are directional signals for what AI-adjacent product categories are generating real money.

Recent signals
  • Lex Fridman Podcast #440: gave him mainstream recognition for the solo founder narrative
Reference value: Cite Levels when you need concrete proof that the one-person unicorn aspiration is real. His exact revenue numbers (which he posts publicly) are reference-worthy.
@swyxio Shawn Wang (swyx)
Co-founder of Latent Space, builder, AI engineer

Primary community builder for the AI engineering layer. Latent Space podcast had 10M+ readers/listeners in 2025. His framing ('Software 3.0,' 'AI engineer') shapes how practitioners see their identity. Coined 'AI Engineer' as a distinct job category.

Signal type
  • AI engineering community pulse
  • Emerging tool discourse
  • Paper-to-practice bridge
What to watch

If swyx is writing about a new approach or framework, it's about to become standard practice in the engineering-layer community.

Reference value: Good for engineering-layer discourse. Latent Space as a source is high-credibility.
Simon Willison (blog + X) Simon Willison
Open-source developer, LLM practitioner. No major company affiliation.

Credible, skeptical, honest practitioner voice. Coined 'prompt injection,' 'lethal trifecta' (AI agent security). Amplified 'AI slop.' Known for calling out hype vs. reality. Co-creator of Django. Creator of Datasette.

Signal type
  • Anti-hype, honest practitioner notes
  • Security issues in AI systems
  • Practical LLM limitations
What to watch

When Willison writes critically about a technology or practice, he's usually right and ahead of consensus. If the builder community is praising something he's skeptical of, note the tension.

Reference value: Cite Willison when you need the credible skeptic's view. 'Even Simon Willison thinks X' carries weight.
@bcherny Boris Cherny
Creator of Claude Code at Anthropic

Insider credibility on the most talked-about coding tool in the ecosystem. Low volume, but each post carries enormous weight. The '100% of my Claude Code contributions were written by Claude Code' thread is one of the canonical proof-of-concept drops of the agentic era.

Signal type
  • AI coding tool philosophy
  • Workflow reveals
  • Agentic development in practice
What to watch

His workflow reveals are among the highest-signal content about agentic engineering in practice.

Recent signals
  • Dec 27, 2025: '100% of my contributions to Claude Code were written by Claude Code' — viral
  • Dec 2025: Full workflow thread — 'My setup might be surprisingly vanilla! Claude Code works great out of the box'
Reference value: Cite Boris Cherny when you need the 'creator uses their own tool at this level' proof. His specific numbers (259 PRs, 497 commits, 40K lines) are the reference standard for the agentic era.
@AravSrinivas Aravind Srinivas
CEO, Perplexity AI

One of the most active and quotable founder accounts in the AI discourse. Perplexity topped $450M ARR; launched Computer (agent tool) and Comet browser in 2026. Will challenge anything — including Google, OpenAI, and venture orthodoxy.

Signal type
  • AI search narrative (vs. Google)
  • Agent product signals
  • Startup founder candor (product strategy, revenue)
What to watch

Perplexity product launches often signal what the next search/agent paradigm is. His debates with incumbents (Google, OpenAI) generate engagement and reveal real competitive dynamics.

Recent signals
  • Jan 2026: Perplexity Computer launch — AI agent for non-experts
  • Feb 2026: Comet browser launch — AI-native browser to rival Chrome
  • Mar 2026: $450M ARR milestone announced
@mntruell Michael Truell
CEO, Cursor (Anysphere)

Cursor is the company most cited as evidence that AI coding tools can generate real revenue at scale. $300M ARR by early 2026. His technical essays on AI engineering set narrative for the entire tooling space. The 'Third Era of AI Software Development' essay was reshared by hundreds of developers.

Signal type
  • AI coding tooling narrative
  • Developer productivity signals at scale
  • Platform era for AI-assisted development
What to watch

When Truell writes an essay, it shapes how developers understand what coding tools are becoming.

Recent signals
  • 2026: 'Third Era of AI Software Development' essay — widely cited
  • 2026: Cursor hits $300M ARR, approaches $30B valuation
@amasad Amjad Masad
CEO, Replit

Replit is the poster company for vibe coding made real: $9B valuation after $400M raise in March 2026. Masad is one of the most active and contrarian founders on X. Notably skeptical of 'vibe coding' despite being credited as its enabler. Replit projects $1B revenue by end of 2026.

Signal type
  • Vibe coding reality check
  • Developer tool market signals
  • Solo/lean team aspiration proofing
What to watch

His product announcements (Replit Agent 4 can maintain an entire company's codebase) are real milestones. His skepticism of his own company's narrative is signal-rich.

Recent signals
  • Mar 2026: Replit raises $400M at $9B valuation (3x in 6 months)
  • 2026: Replit Agent 4 — can build and maintain an entire company, not just apps
@gregisenberg Greg Isenberg
CEO, LateCheckout; builder, framework sharer

One of the most-retweeted builder accounts in the bubble. Amplifies AI startup ideas, 'leveraged agency' concepts, Paperclip demos. Popularized the 'AI-powered company' format threads. His content shapes what the mainstream-adjacent founder audience thinks is possible.

Signal type
  • Builder aspiration culture
  • Business idea generators
  • Paperclip/Polsia/zero-human company amplification
What to watch

When he picks up a new tool or concept, expect mainstream builder adoption within weeks.

Recent signals
  • 2026: Extensive threads on Paperclip and zero-human company concept
  • Ongoing: 'AI + [vertical]' business idea threads
@hwchase17 Harrison Chase
CEO, LangChain

LangChain is foundational infrastructure for the agent developer community. Sequoia-backed. His posts define the technical language of agent building. Published 'The Rise of Context Engineering' blog amplifying Tobi's vocabulary shift.

Signal type
  • Agent infrastructure vocabulary
  • LangChain ecosystem pulse
  • Context engineering technical framing
What to watch

His framing of what agents can and can't do is influential in enterprise AI adoption.

Researchers & Lab Leaders

AI research and lab leadership · 10 entries
@darioamodei Dario Amodei
CEO, Anthropic

Increasingly visible on X compared to early Anthropic years. His 'Machines of Loving Grace' essay (2024) defined the optimistic case for AI benefit. The Super Bowl ad war and Modi India Summit 'refused to hold hands with Altman' moment made him a cultural figure. Anthropic's $30B Series G (Feb 2026) and Mythos model decision define his 2026 narrative.

Signal type
  • Anthropic strategic posture
  • AI safety practical framing
  • Lab-vs-lab competitive signals
What to watch

When Dario writes about safety-capability tradeoffs, it signals Anthropic's internal decision-making process. The Mythos withholding decision is the clearest example.

Recent signals
  • Feb 2026: Super Bowl ad war — Anthropic vs. OpenAI ('Deception,' 'Betrayal,' 'Treachery')
  • Feb 19, 2026: Modi India Summit — refused to hold hands with Altman in unity photo
  • Mar/Apr 2026: Mythos model — withheld from public release due to autonomous hacking capability
@ilyasut Ilya Sutskever
CEO, Safe Superintelligence (SSI)

Co-founder of OpenAI who led the board's attempted Altman removal (Nov 2023). Now building SSI with a single explicit mission: safe superintelligence. Rare X poster but when he posts, it shapes AGI discourse for weeks. Followed by every serious AI researcher; his technical judgments are authoritative.

Signal type
  • AGI safety signals
  • Technical judgment calls (architecture, training)
  • SSI mission signals
What to watch

Any SSI announcement or Ilya commentary on model capability = major discourse event.

@fchollet François Chollet
Creator of ARC-AGI benchmark; Senior Researcher, Google

ARC-AGI is the most-debated benchmark for testing whether models can actually reason. ARC-AGI-3 launch (March 25, 2026 at YC HQ, fireside with Altman) was a major discourse event. GPT-5.4 and Claude Opus 4.6 scored 0.3% on ARC-AGI-3; humans score 100%. The clearest voice for 'current LLMs haven't solved general intelligence.'

Signal type
  • Benchmark validity discourse
  • AGI claims reality-checking
  • Reasoning capability honest assessment
What to watch

ARC-AGI score releases are quarterly benchmark events that reset AGI discourse. His technical critiques of scaling claims are influential among serious researchers.

Recent signals
  • Mar 25, 2026: ARC-AGI-3 launch — frontier models score 0.3% vs. human 100%
@ylecun Yann LeCun
Chief AI Scientist, Meta AI

The most prominent internal skeptic of LLM scaling as the path to AGI. 972K followers. Prolific, combative, engaged in public feuds with Gary Marcus, Geoffrey Hinton. His thesis: current LLMs are fundamentally limited and cannot reach AGI. This makes him the most credible heterodox voice within the AI establishment.

Signal type
  • LLM limitations discourse
  • Alternative AI architecture advocacy
  • AGI skepticism (credentialed internal dissent)
What to watch

When LeCun disagrees with a mainstream claim, the specific technical argument he makes is usually worth understanding — even if the conclusion is disputed.

@drfeifei Fei-Fei Li
Co-director, Stanford HAI; Founder, World Labs (embodied AI / 3D)

Bridges academic AI and applied AI/robotics. 516K followers. World Labs (embodied AI + 3D world models) is a major 2025 launch. Her voice is influential in policy circles and corporate boardrooms (Apple board member). Not in the SF-bubble-hype discourse — gives her comments extra weight as signal vs. noise.

Signal type
  • Physical world AI signals (robotics, embodied)
  • Policy and safety mainstream bridge
  • Academic research to practical application
@natfriedman Nat Friedman
Former GitHub CEO; Co-founder, AI Grant

AI Grant backs exceptional individuals building with AI (small grants, huge signal). Posts deeply technical, often pre-viral observations about AI capability. Connected to the serious AI researcher community via AI Grant. His 'Are Language Models Actually Useful?' type analyses are influential.

Signal type
  • Pre-viral AI capability observations
  • Independent researcher backing signals
  • Technical investor credibility
@danielgross Daniel Gross
AI Grant co-founder; recently joined Meta

AI Grant (with Nat Friedman) has backed multiple high-signal AI researchers. Left AI Grant for Meta in early 2026 — signals Meta's renewed aggressive AI hiring.

@gdb Greg Brockman
President/Co-founder, OpenAI

Active X poster on technical and philosophical AI topics. $25M donation to Trump-aligned PAC triggered controversy during Pentagon deal era. Co-founder presence provides OpenAI mission credibility alongside Altman.

Signal type
  • OpenAI mission signals
  • Technical AI frontier commentary
@alexalbert__ Alex Albert
Head of Developer Relations, Anthropic

Publishes Claude prompt recipes and agent patterns directly from Anthropic insider view. His prompt engineering / context engineering tutorials are widely followed in the builder community. Primary public face of Anthropic's developer-facing strategy.

Signal type
  • Anthropic technical patterns (authoritative)
  • Claude capability reveals
  • Prompt/context engineering tutorials
What to watch

His posts about Claude capabilities often precede broader capability revelations. Developer-facing Anthropic strategy visible through his threads.

@OfficialLoganK Logan Kilpatrick
OpenAI Developer Advocacy; Harvard AI instructor

231K followers. Bridges the OpenAI API developer community. His tutorials and capability demonstrations shape how developers think about OpenAI's tools.

Founders & Operators

Company builders driving discourse · 12 entries
@steipete Peter Steinberger
Creator of OpenClaw; ex-indie developer (PSPDFKit). Joined OpenAI Feb 2026.

The defining 'vibe coder makes viral product' story of early 2026. OpenClaw (formerly Clawdbot/Moltbot) hit 247K GitHub stars, 47K forks as of March 2026. Sam Altman praised him publicly; he joined OpenAI to lead personal AI agents. Made 6,600+ commits in January 2026 alone, running 4-10 Claude Code agents simultaneously. MIT Technology Review covered China 'gold rush' around his tool.

Signal type
  • Agentic coding methodology in practice (real-time evidence)
  • Consumer AI agent product signals
  • Build-in-public at extreme velocity
What to watch

His OpenAI work on personal AI agents is the most-watched 'what does Altman think AI assistants become' signal.

Recent signals
  • Nov 2025 – Feb 2026: OpenClaw viral arc — trademark battle, 247K GitHub stars, Sam Altman praise
  • Feb 15, 2026: Joined OpenAI to lead personal AI agents
  • Jan 2026: 6,600+ commits in one month — canonical extreme agentic velocity example
(Ben Broca, Polsia) Ben Broca
Solo Founder, Polsia

Polsia crossed $1M ARR in one month with Ben as the sole employee. $4.5M ARR run rate. Claude Opus 4.6 as CEO agent. The canonical living proof-of-concept for 'one-person AI-run company.' Dave Morin (@davemorin) amplified his fundraising. Fortune covered the story.

Signal type
  • 'One-person unicorn' proof of concept (live, auditable)
  • AI-run company operations evidence
What to watch

His ARR progression and product reviews are the most-watched solo founder data in the bubble.

Recent signals
  • Feb 2026: Polsia launches — $1M ARR in one month as sole employee
  • 2026: $4.5M ARR run rate; Claude Opus 4.6 as primary reasoning agent
  • Note: user reviews flag incomplete work, billing issues — still immature
@eshear Emmett Shear
Former CEO, Twitch; brief OpenAI interim CEO (Nov 2023)

Active X commentator on AI/startup dynamics. His 3-day tenure as OpenAI interim CEO gives him credibility on governance discourse. Frequently posts nuanced takes on AI safety, model capability, and startup strategy.

@hwchase17 Harrison Chase
CEO, LangChain

See secondary section.

@EricNewcomer Eric Newcomer
Founder, Newcomer.co newsletter

The most credible inside-baseball reporter on startup/VC dynamics. Broke the Sequoia stewardship transition story (Botha stepping down, Lin/Grady taking over). His reporting shapes how founders and VCs understand industry dynamics.

Signal type
  • VC firm and funding scoops
  • Governance and power dynamics in tech
@amasad Amjad Masad
CEO, Replit
@mntruell Michael Truell
CEO, Cursor
@dkaufmann Derik Kaufmann
Founder, RunRL (YC-backed)

Organized the 'March for Billionaires' (Feb 7, 2026) — an earnest anti-billionaire-tax protest that became a canonical cringe-viral cultural reference. ~36 attendees, near-equal press corps. The event has a Wikipedia page and was covered in SF Chronicle, SF Standard, late-night comedy.

Signal type
  • Cringe-viral cultural reference point
  • SF startup scene political naivety marker
@Artisanai CEO Jaspar Carmichael-Jack
CEO, Artisan AI

Ran the canonical 'Stop Hiring Humans' SF billboard rage-bait campaign. Publicly admitted: 'the goal of the campaign was always to rage bait.' The company name Artisan is now a permanent case study in aggressive AI marketing. $2M ARR from the campaign; also received death threats.

Signal type
  • Rage bait case study
  • AI labor displacement discourse trigger
@zachlloyd Zach Lloyd
CEO, Warp (terminal-as-coding-tool)

Made the precise distinction: 'I don't believe the Cursor is dead memes, but the IDE is dead is real.' This framing became the nuanced position in the IDE wars. Warp is the AI-native terminal that makes the 'IDE is dead' thesis make sense.

Signal type
  • IDE-vs-agent-era nuanced framing
  • Developer tooling discourse
Recent signals
  • Feb 2026: 'The IDE Is Dead' framing — cleanly distinguishes from 'Cursor is dead' clickbait
@svembu Sridhar Vembu
CEO, Zoho

The most credible voice on the 'SaaS is not dead' side of the Garry Tan debate. His challenge: 'If vibe coding is so powerful, where is the vibe-coded email, accounting, or ERP software?' As Zoho's founder with 30+ years of building enterprise software, his credibility on this is real. Provides the essential counter to startup bubble SaaS triumphalism.

Signal type
  • SaaS-is-alive counter-signal
  • Technical realism vs. vibe coding hype
  • Non-VC-consensus India tech voice
@dwarkesh_sp Dwarkesh Patel
Host, Dwarkesh Podcast

Deeply researched long-form interviews with AI researchers and leaders. Time 100 AI as of 2025. Interviewed Dario Amodei and Elon Musk in early 2026. His interview questions are often more technically specific than any journalist. The 'serious AI discourse' podcast — audience is researchers, founders, serious operators.

Signal type
  • Deep AI researcher access
  • Serious AI intellectual discourse

Investors & VCs

Capital and narrative setters · 20 entries
@pmarca Marc Andreessen
Co-founder, a16z

The most culturally influential VC on X. Posted 'Techno-Optimist Manifesto' (2023) still reverberates. Posts on AI, tech politics, 'techno-optimism.' a16z raised $15B across five funds in January 2026. His cultural positioning shapes how the next generation of SF founders think about their mission.

Signal type
  • Techno-optimism narrative
  • a16z investment thesis signals
  • Cultural/political commentary (can be brand-risky to follow)
@chamath Chamath Palihapitiya
General Partner, Social Capital; All-In co-host

All-In Podcast co-host with millions of followers across platform. Major macro/political/AI voice. Coined or amplified 'AI roundtripping' bubble concern. His public bets on AI (and specific companies) move markets.

Signal type
  • Macro AI/tech market takes
  • All-In Podcast talking point preview
  • SF tech cultural pulse
@davidsacks47 David Sacks
All-In co-host; Trump's AI & Crypto Czar (White House)

Joined Trump White House as AI & Crypto Czar in early 2025 — still influential in venture discourse. All-In co-host gives him daily reach to millions. His White House role shapes US AI policy: the most powerful VC-adjacent person in government.

Signal type
  • US AI policy signals (primary source)
  • All-In Podcast previews
  • Republican-aligned tech narrative
@jason Jason Calacanis
Host, This Week in Startups; All-In co-host

Extremely active on X. Amplifies startup discourse daily. All-In co-host. Invested in early Uber, Robinhood, Calm. Large platform for startup signal. Opinion-forward, often first to loudly voice what the SF tech community is feeling.

Signal type
  • SF startup culture real-time pulse
  • Early-stage funding signals
@friedberg David Friedberg
CEO, Production Board; All-In co-host

All-In co-host with strong biotech/science/deeptech lens. 'Friedberg for Governor' became a meme in SF tech circles (2026 All-In episode). His science takes (on agriculture, climate, biology) balance the bubble's software-centrism.

@altcap Brad Gerstner
Founder, Altimeter Capital; All-In 'fifth bestie'

Returned as 'fifth bestie' on All-In in 2026. Major AI thesis investor. His 'AI roundtripping' framing (AI companies as each other's biggest customers) is the primary bubble-skepticism signal. Altimeter led significant OpenAI stakes.

@naval Naval Ravikant
AngelList co-founder; philosopher-investor

Aphoristic X posts shape how founders think about wealth, autonomy, and technology. His takes age well — frequently cited from 2-5 years ago. 'AI won't replace programmers, but make it easier for programmers to replace everyone else.' Quieter recently but each post is highly amplified.

Signal type
  • Philosophical framing of technology
  • Long-horizon investment thinking
  • Wisdom that ages well (safe to cite from any year)
@garrytan Garry Tan
YC CEO (also acts as an investor voice)
@eladgil Elad Gil
Serial investor; co-host, No Priors podcast

Has backed Airbnb, Twitter, Stripe, Coinbase, Instacart — 100-seed investor. Co-hosts No Priors podcast with Sarah Guo. Highly influential in quiet ways. His frameworks on AI company building (from 'High Growth Handbook') are widely cited.

@saranormous Sarah Guo
Founder, Conviction; co-host, No Priors podcast

One of the most visible AI-focused VCs. Runs Conviction ($500M+). No Priors podcast with Elad Gil consistently surfaces pre-viral AI company founders. Her 'AI company formation thesis' is influential in how the investor community thinks about the space.

@packym Packy McCormick
Not Boring newsletter + Not Boring Capital

Built a media → venture flywheel: newsletter funds his VC decisions and his VC decisions feed newsletter. Long-form deep dives on tech companies are widely read by founders and VCs. Operates as independent single GP with full public voice.

@hstebbings1996 Harry Stebbings
Founder, 20VC; runs $400M fund

Largest tech VC podcast platform. Has interviewed every major VC and founder. Now runs $400M fund. The combination of media reach + capital creates unique leverage. His 'top 20 VCs' and 'top 20 founders' interview series shape perception of who matters.

@vkhosla Vinod Khosla
Khosla Ventures

Aggressive AI bull. Public and opinionated on X. Predictions: AI will replace 80% of white collar jobs; doctors, lawyers, teachers. His extreme takes define the upper bound of the AI-disruption thesis.

Signal type
  • Upper-bound AI disruption thesis
  • Healthcare AI advocacy
@rabois Keith Rabois
Khosla Ventures (Partner); ex-Founders Fund; co-founder OpenStore

Known for combative X takes on SF politics, Miami tech scene, startup culture. Embedded in the PayPal Mafia network — his investment thesis shapes early-stage AI company formation. His criticism of SF (and subsequent Miami experimentation, then SF return) is a cultural narrative.

@fredwilson Fred Wilson
Union Square Ventures

Long-time blogger (AVC.com). One of the most consistent long-form investor voices. Less X-native but his blog posts are widely shared and authoritative on consumer/network businesses.

@m2jr Mike Maples Jr.
Floodgate

'Thunder lizards' framing for company building. Active writer on X about startup theory. His 'Pattern Breakers' framework is widely cited.

@reidhoffman Reid Hoffman
Greylock; LinkedIn co-founder

Extremely active on AI, writing, and tech discourse. Author of 'Impromptu' (with GPT-4). Actively building in AI with Mana (social AI startup). His LinkedIn co-founder status + AI advocacy = significant institutional voice.

(Josh Kushner) Josh Kushner
Thrive Capital

Led OpenAI's $6.6B round (October 2024). Thrive is one of the largest OpenAI stakeholders. Low X presence but enormous capital influence.

(Masayoshi Son) Masayoshi Son
SoftBank Vision Fund

Led OpenAI's $30B February 2026 round tranche. Vision Fund is the largest source of AI capex. His 'Singularity' proclamations set the outer bound of AI investment optimism. Low X presence but enormous capital flow influence.

@a16z a16z Network
Andreessen Horowitz — entire firm

Raised $15B across five funds in January 2026. Most active AI investor by deal count. Publishes 'Big Ideas' yearly that shape investment discourse. a16z AI blog, a16z podcast, SPEEDRUN accelerator, New Media Fellowship all extend reach.

Signal type
  • AI investment thesis (authoritative annual updates)
  • Accelerator deal flow signals

Builders (in public)

Build-in-public accounts · 9 entries
@levelsio Pieter Levels
Solo founder, PhotoAI / NomadList / RemoteOK
@goodside Riley Goodside
Staff Prompt Engineer, Scale AI

Known for early prompt engineering demonstrations that shaped how practitioners thought about LLMs. 'Red-teaming' AI systems in public. His exploits (jailbreaks, system prompt leaks) are high-signal technical content.

@waitin4agi_ Varun Mayya
CEO, Avalon Labs

219K followers. Agent/automation builder with strong presence in the 'building AI companies' tier. His demos of multi-agent workflows are widely shared.

@LinusEkenstam Linus Ekenstam
Product/experience designer; AI builder

219K followers. Barcelona-based but prominent in AI builder discourse. Visual-forward demonstration of AI tools. Appeals to designer/product builder audience.

@gregisenberg Greg Isenberg
CEO, LateCheckout
@karpathy Andrej Karpathy
@steipete Peter Steinberger
(anonymous) Paperclip Founder
Creator, Paperclip (paperclip.ing)

Anonymous founder of Paperclip (open-source zero-human company platform). 30K GitHub stars in under 3 weeks. Greg Isenberg amplified extensively. Never shows face publicly. The Paperclip name is a deliberate dark-humorous homage to the 'paperclip maximizer' AI alignment thought experiment.

Signal type
  • 'Zero-human company' thesis live demonstration
(Ben Broca) Ben Broca

Podcasters & Writers

Podcast hosts and newsletter writers · 11 entries
@monkbent Ben Thompson
Founder, Stratechery (stratechery.com)

The definitive long-form analysis newsletter for business strategy in tech. Stratechery inspired Substack itself. 2026 pieces include 'Aggregators and AI,' 'Winners, Losers, and the Unknown,' 'The Chip Fly in the AI Ointment.' The most-cited paid newsletter in the SF startup bubble for business strategy.

Signal type
  • AI business model framework (authoritative)
  • Platform strategy analysis
  • SaaS / aggregator / AI competitive dynamics
What to watch

His frameworks for understanding AI competitive dynamics (aggregators, platforms, verticals) are the standard reference for how the bubble thinks about business models.

@lennysan Lenny Rachitsky
Lenny's Newsletter + Lenny's Podcast

Product/growth/career focus. 1M+ subscribers. Top-10 tech podcast, 337 episodes. His Michael Truell (Cursor $300M ARR) episode was one of the most-shared founder conversations of 2026. Simon Willison 'AI State of the Union' pieces hosted on Lenny's are widely cited.

@TheZvi Zvi Mowshowitz
Author, Don't Worry About the Vase (Substack)

Weekly AI roundup is essential for the rationalist/AI safety/EA community. His analysis of AI capability claims, safety incidents, and policy is the most thorough independent aggregation available. Wrote extensively on Mythos in April 2026. Gary Marcus and the mainstream press cite him when trying to understand AI safety discourse.

Signal type
  • AI safety discourse aggregation (authoritative)
  • Capability claim reality-checking
  • Rationalist/EA-adjacent AI analysis
What to watch

When Zvi covers a model release or incident in depth, his analysis surfaces non-obvious risks. His 'good/bad/mixed' weekly structure is a fast way to understand the week in AI.

@labenz Nathan Labenz
Host, The Cognitive Revolution podcast

Deep AI discourse with safety, healthcare, biosecurity, and employment disruption focus. His interviews surface researchers and builders who aren't yet mainstream. Deeply embedded in the serious AI discourse tier alongside Dwarkesh Patel.

@rowancheung Rowan Cheung
Founder, The Rundown AI newsletter

2M+ subscribers. 567K X followers. World's largest AI newsletter (by subscriber count). Daily digest format. His newsletter reach means items he covers are in the mainstream awareness tier. What he covers today will be in casual startup conversations tomorrow.

@swyxio Shawn Wang (swyx)
Co-host, Latent Space podcast
@mariogabriele Mario Gabriele
Founder, The Generalist (newsletter)

Deep company profiles with investment angle. Long-form that VCs and founders actually read. His coverage of AI companies (Anthropic, Cohere, Mistral) is widely cited.

@EricNewcomer Eric Newcomer
Newcomer.co newsletter
@lexfridman Lex Fridman
Host, Lex Fridman Podcast

Long-form AI researcher and philosopher interviews. 3.3M+ YouTube subscribers. His interviews are the route by which AI researchers enter mainstream cultural consciousness. Pieter Levels appearance (#440) gave the solo founder narrative mainstream visibility.

Signal type
  • AI researcher mainstream access
  • Cross-demographic AI discourse amplifier
@bentossell Ben Tossell
Founder, Ben's Bites newsletter

Approachable daily AI tools/funding/news digest. Accessible entry point to the ecosystem. His 'product of the week' picks often predict mainstream tool adoption.

@dwarkesh_sp Dwarkesh Patel

Amplifiers

Cross-platform spread signals · 10 entries

Meta-Commentators

Critics inside the bubble · 6 entries
@GaryMarcus Gary Marcus
Cognitive Scientist, NYU Emeritus; AI Skeptic

The most prominent AI skeptic with real credentials. 197K followers. Maintains a Substack with detailed rebuttals of AI hype claims. When AGI announcements happen, Gary Marcus is the first credible critic in mainstream media. His name is shorthand for 'the serious counterargument.'

Signal type
  • AGI hype counter-signal (credentialed)
  • Benchmark validity criticism
  • Mainstream press go-to AI skeptic
Recent signals
  • Apr 2026: Wrote on Mythos model announcement — pushed back on AGI claims
@timnitGebru Timnit Gebru
Co-founder, DAIR Institute; ex-Google AI Ethics Lead

The clearest voice on AI bias, power concentration, and labor impact from inside the field. Her firing from Google (for 'Stochastic Parrots' paper) is a canonical AI ethics incident. DAIR Institute is the primary independent AI ethics research organization. Her critiques of AI companies' safety claims have shaped mainstream press coverage.

Signal type
  • AI ethics and bias discourse (authoritative)
  • Labor impact of AI (concrete critique)
  • Power concentration in AI (anti-monopoly lens)
@edzitron Edward Zitron
Tech PR/media critic; founder, EZPR; Substack writer

The sharpest mainstream critic of AI hype and tech industry culture. Challenged the technical accuracy of Matt Shumer's 'Something Big Is Happening' essay. His Bluesky audience is large and actively anti-hype. His critiques show up in press. Represents the growing skeptical counter-discourse to SF bubble optimism.

Signal type
  • AI hype counter-signal (populist, non-technical)
  • Tech industry culture critic
@ylecun Yann LeCun
Chief AI Scientist, Meta
@svembu Sridhar Vembu
CEO, Zoho
(Alvaro Higes) Alvaro Higes
Substack author, 'And AGI Went By'

Wrote '2026 the Year of Boring AI' — the primary source for the 'boring beats flashy' counter-narrative. Enterprise rollups and workflow standardization outperform agentic commerce hype. Not a major X presence, but his essay circulates in the bubble as the canonical anti-hype thesis.

Signal type
  • Boring AI counter-narrative (canonical essay source)