Dr. Alexander D. Wissner-Gross is a physicist, computer scientist, and investor: Founder & Chief Scientist of Physical Superintelligence (PSI), backer or advisor to 40+ companies through 021T and Reified, co-host of the Moonshots podcast, and author of The Innermost Loop (100,000+ daily readers), with a corpus of 24 publications, 26 patents, and 128 distinctions behind him. We read alexwg.org the way we read any site before writing about it: plain HTTP requests plus a headless-browser render pass at real desktop and mobile viewports, nothing probed, nothing authenticated. What we found is a real gap between the work and the wrapper holding it: an extraordinary corpus sitting on markup that predates the mobile web, with zero machine-readable structure anywhere we checked.
evidence.json). Every fact about his work is cited to his own pages, inline. Full disclosure ↓
What we found: an extraordinary corpus (24 publications, 26 patents, 128 distinctions) sitting on markup that predates the mobile web: no viewport tag, no structured data, a 1.66MB headshot at 185px.
What we built, free: a brand-matched concept redesign (structured data per publication, a draft discovery file, a before/after table) plus an Agent Kit that points at a live payment endpoint and a live agent surface: all reviewable, nothing installed.
The ask: none required. If the appendix on why we did this earns twenty minutes, it's one click away: nothing here needs a reply.
Everything else in this package is measurement. This is the only part that is a claim, and it is a claim about where value ends up, not about him.
His own distillation argument, taken seriously, has a second answer in it. If a frontier model costing a billion dollars can be functionally duplicated for a few million, then value cannot rest in the model layer for long, and the refuge he names is hardware: the physical thing that cannot be copied out of a weights file. We think there is a second refuge sitting right next to it, and it is cheaper to build. Weights distill. Proof does not. A settled onchain receipt, a worker's own history of accepted work, and a place's measured standing are not compressible into someone else's model, because they are records of something that happened in a specific place at a specific time and was checked. Copy the model all you like: you still did not do the work.
That is the whole thesis of what we build. The verification layer is the part of an AI economy that stays scarce, and almost nobody is building it as infrastructure. The long version is in the appendix, with the honest limits attached.
Five things alexwg.org already gets right, measured the same way as everything else in this note.
robots.txt, llms.txt, sitemap.xml, five .well-known/* paths) returns a true, byte-identical 404 against a nonsense-path control: no soft-404 deception.<meta name="viewport"> tag, on any page. A real 390px-wide mobile viewport reports window.innerWidth = 980 and visualViewport.scale = 0.398: the whole corpus opens pre-shrunk to about 40% scale.width attribute.Each stop below is either a citable measurement or a labeled concept: never a projection dressed as a fact.
The complete audit: five things alexwg.org gets right, five things we measured (mobile viewport, structured data, image weight, compression, corpus queryability): each with the exact number and how to re-check it yourself.
Read the findings →A brand-matched concept redesign in the same elevated-academic register as his own site: JSON-LD (the invisible labels search engines and AIs read) per publication and patent, a draft llms.txt (the file AI assistants read first), and a before/after table: every fix labeled concept, nothing claimed as already shipped.
The appendix, not the headline: why proof is the second refuge from distillation, why a legible agent economy is the benign version of the scenario safety people worry about, and a careful structural case for why his 2013 causal-entropy equation and his footprint-monitoring patent describe the same shape of problem. Live on two mainnets, with the limits named in the same breath as the claims.
Read the pitch →The live x402 endpoint (a web-native way for AI agents to pay per use), the live agent surface an outside model can drive end to end, and what an agent actually earns when it finishes a job: for a PSI researcher's agent, a Get Physics Done copilot, or anything built inside 021T or Reified's portfolio.
Open the kit →If you have ever asked out loud why nobody builds a nonviolent game: here is ours, live and playable. A tiny civilization tech tree where every node you unlock is a real job on the open board, provable while you play. We built it rather than argue about whether it could exist.
Play it →We build a protocol that settles ecological work the way markets settle compute, funded, proof-verified, on-chain, live today across three chains, settlement on Base and Robinhood Chain, the work library anchored on Regen Ledger. His own causal-entropy equation and his environmental-footprint-monitoring patent are why we thought he, specifically, would want to see it, not flattery, a structural argument we make carefully, once, in the brief. If any of this earns twenty minutes, the brief has the full case: nothing here requires a reply.
The board is not a mockup, and it is not a market yet. Right now it holds - posted packets worth -, of which - are open to claim and - carry a verified payment behind the wage. - have settled end to end, and the wallet that did that work was our own: proof the lifecycle runs, not a customer buying anything. No outside buyer has funded a packet yet, and we publish that number at zero on purpose rather than let a demonstration look like demand. Those figures read live from the board's own counter each time this page opens: if a dash is showing, the fetch did not land, and no number here was ever typed in by hand.
