Credit
The Independent 100 list was curated by dan (@irl_danB) and published on 2026-09-24 at independent.prose.md (announcement). The ranks and section names are his, and number 100 is his open wild-card slot. This study is ours and is not endorsed by him or by anyone on the list.
Method
- Scope. The 99 people named on the list, taken with the curator's rank and section as given. We did not add, drop or reorder anyone.
- First-pass survey, 2026-09-24. Ten AI agent sessions each profiled about ten people from public sources: personal sites and blogs, GitHub repositories and the GitHub API (licenses, stars, last activity, contributor counts), papers, and a small number of public posts. Each record says what the person publicly does, lists their public work with links and licenses, scores its relevance to our own work, and names what we would study next.
- Deep dives, 2026-09-24 and 2026-09-25. For a few people whose work overlaps ours most, a separate agent session read the source code and writing in depth and compared it with how we work. Each deep dive has its own page.
- Public edition, 2026-09-25. The records and deep dives were edited for publication under the rules below, every link was checked, and the result was schema-checked before any page was rendered. A reviewer from another model family then reviewed it, and its corrections were applied the same day.
- Nothing was installed, run or benchmarked for the survey. Where a deep dive ran something, it says so.
Relevance rubric
Each record carries a score from 0 to 5 for relevance to our work: how much the person's public work bears on what we build and study (agent harnesses and orchestration, evals and verification, context and memory, agent security, research and evidence practice, local models, and writing and publishing). It is not a measure of the person, their influence or the quality of their work, and it is not a ranking of people.
- 5: public, openly licensed work squarely in our lanes that we could study or adopt now.
- 4: strong overlap, with a caveat such as a missing license, early maturity or a partial fit.
- 3: a useful method or useful context, less directly adoptable.
- 2: background for our work, with little to adopt.
- 1: little connection to our work.
- 0: no connection (no record uses it).
Each score is one reviewer's judgement: one agent session scored each person, and the ten sessions were not cross-calibrated. Read a 5 as "has something open and in our lanes", nothing more. The lenses on each record (for example agents, evals, security) name the lanes the score refers to.
Deep dives
- dex, Kyle Mistele and HumanLayer (12-Factor Agents, ACE-FCA, skills, shannon, fold) written 2026-09-25; covers dex, Kyle Mistele
How HumanLayer's 12-Factor Agents, ACE-FCA context compaction, control-loop skills, shannon and fold handle context, compaction and agent loops, and what they suggest for agent harnesses. - Geoffrey Huntley: the Ralph loop, the porting recipe, preflight and underclass written 2026-09-24; covers geoff (Geoffrey Huntley)
What Geoffrey Huntley built (the Ralph loop, a four-step porting recipe, the preflight and underclass proxies), how it compares with Jeffrey's port method, and which of its practices carry general lessons. - Hermes Agent, with Teknium, Shannon Sands and mephisto written 2026-09-25; covers Teknium, Shannon Sands, mephisto
How Nous Research's Hermes Agent handles skills, memory, MCP and reliability, and the general lessons it offers anyone building an agent harness. - Omar Khattab and alex zhang, GEPA for skills and Recursive Language Models for long context written 2026-09-25; covers Omar Khattab, alex zhang
What GEPA's outcome-scored optimizer, gskill, Recursive Language Models and spec-ptc do, how they are licensed and documented, and which parts carry risks to know before reuse. - Mario Zechner and Pi, the upstream of OMP written 2026-09-24; covers Mario Zechner
How Pi's small core, project-trust gate and enforced supply-chain rules compare with OMP, the fork built on Pi, and which of Pi's practices carry general lessons.
The 99, in the curator’s order
Grouped by the curator’s sections. The last column is relevance to our work (0 to 5) and its lenses; it is one reviewer’s judgement per person, not a ranking of people. Scores: 24 at 5, 14 at 4, 24 at 3, 26 at 2, 11 at 1.
| Rank | Name | Handle | Section | Relevance to our work |
|---|---|---|---|---|
| The Canon (ranks 1 to 10) | ||||
| #1 | gwern | @gwern | The Canon | 2 of 5 research-method, writing |
| #2 | Jürgen Schmidhuber | @SchmidhuberAI | The Canon | 3 of 5 research, research-method |
| #3 | Robin Hanson | @robinhanson | The Canon | 2 of 5 research-method, business |
| #4 | janus | @repligate | The Canon | 3 of 5 memory, agents |
| #5 | Eliezer Yudkowsky | @allTheYud | The Canon | 2 of 5 research-method, writing |
| #6 | Beff | @beffjezos | The Canon | 1 of 5 research |
| #7 | Pliny the Liberator | @elder_plinius | The Canon | 3 of 5 security, agents, mcp, evals |
| #8 | Scott Alexander | @slatestarcodex | The Canon | 3 of 5 writing, research-method |
| #9 | deepfates | @deepfates | The Canon | 5 of 5 agents, orchestration, evals, memory, search, mcp |
| #10 | Perry E. Metzger | @perrymetzger | The Canon | 2 of 5 security, verification |
| The Frontier Outside (ranks 11 to 31) | ||||
| #11 | Omar Khattab | @lateinteraction | The Frontier Outside | 5 of 5 agents, evals, search, research-method, orchestration |
| #12 | davidad | @davidad | The Frontier Outside | 3 of 5 verification, research-method, security |
| #13 | Jeremy Howard | @jeremyphoward | The Frontier Outside | 5 of 5 writing, agents, content, research-method, business |
| #14 | will brown | @willcb | The Frontier Outside | 4 of 5 evals, training, agents, verification |
| #15 | Connor Leahy | @NPCollapse | The Frontier Outside | 2 of 5 writing, business |
| #16 | Rob Miles | @robertskmiles | The Frontier Outside | 2 of 5 content, writing, security |
| #17 | George Hotz | @realGeorgeHotz | The Frontier Outside | 3 of 5 local-models, research-method, writing |
| #18 | DHH | @dhh | The Frontier Outside | 3 of 5 business, writing, content |
| #19 | Varun | @varun_mathur | The Frontier Outside | 3 of 5 agents, security, memory, orchestration, local-models |
| #20 | alex zhang | @a1zhang | The Frontier Outside | 5 of 5 agents, memory, orchestration, evals, research-method |
| #21 | watermark (anthrupad) | @anthrupad | The Frontier Outside | 1 of 5 content, research-method |
| #22 | Stella Biderman | @BlancheMinerva | The Frontier Outside | 5 of 5 evals, verification, research-method, training |
| #23 | xlr8harder | @xlr8harder | The Frontier Outside | 5 of 5 evals, local-models, verification, agents |
| #24 | doomslide | @doomslide | The Frontier Outside | 1 of 5 writing, research-method |
| #25 | Charles Goddard | @chargoddard | The Frontier Outside | 4 of 5 training, local-models |
| #26 | Teknium | @Teknium | The Frontier Outside | 5 of 5 agents, orchestration, mcp, memory, security |
| #27 | Larissa Schiavo | @lfschiavo | The Frontier Outside | 3 of 5 orchestration, agents, research-method |
| #28 | samsja | @samsja19 | The Frontier Outside | 4 of 5 training, evals, orchestration |
| #29 | Erik Bernhardsson | @bernhardsson | The Frontier Outside | 4 of 5 search, verification, research-method, business |
| #30 | Jeff Huber | @jeffreyhuber | The Frontier Outside | 5 of 5 search, memory, evals, mcp, rust |
| #31 | vas | @vasuman | The Frontier Outside | 2 of 5 business, agents |
| The Sensemakers (ranks 32 to 52) | ||||
| #32 | Simon Willison | @simonw | The Sensemakers | 5 of 5 agents, security, verification, writing, evals |
| #33 | Ben Thompson | @benthompson | The Sensemakers | 2 of 5 business |
| #34 | Will Manidis | @WillManidis | The Sensemakers | 2 of 5 writing, business |
| #35 | Ethan Mollick | @emollick | The Sensemakers | 4 of 5 evals, verification, research-method, writing |
| #36 | swyx | @swyx | The Sensemakers | 3 of 5 agents, content, business |
| #37 | Joe Weisenthal | @TheStalwart | The Sensemakers | 2 of 5 business |
| #38 | Dan Shipper | @danshipper | The Sensemakers | 5 of 5 agents, orchestration, evals, writing, business |
| #39 | John Scott-Railton | @jsrailton | The Sensemakers | 2 of 5 security, research-method |
| #40 | jessicat | @jessi_cata | The Sensemakers | 1 of 5 research-method |
| #41 | Alex Volkov | @altryne | The Sensemakers | 4 of 5 mcp, agents, evals, content |
| #42 | Zvi Mowshowitz | @TheZvi | The Sensemakers | 3 of 5 research-method, writing, evals |
| #43 | Teortaxes | @teortaxesTex | The Sensemakers | 2 of 5 local-models, research-method |
| #44 | Matt Busigin | @mbusigin | The Sensemakers | 2 of 5 agents, orchestration |
| #45 | Michael Thiessen | @MichaelThiessen | The Sensemakers | 3 of 5 evals, verification, agents |
| #46 | Mike Taylor | @hammer_mt | The Sensemakers | 5 of 5 evals, verification, agents, writing, content |
| #47 | Julia Galef | @juliagalef | The Sensemakers | 2 of 5 research-method, writing |
| #48 | Roko | @RokoMijic | The Sensemakers | 1 of 5 research-method |
| #49 | Kylie Robison | @kyliebytes | The Sensemakers | 2 of 5 business, content |
| #50 | Andrew Curran | @AndrewCurran_ | The Sensemakers | 2 of 5 research-method, business |
| #51 | Tenobrus | @tenobrus | The Sensemakers | 4 of 5 agents, memory, orchestration, security |
| #52 | Eli Lifland | @eli_lifland | The Sensemakers | 3 of 5 research-method, evals |
| The Basement Labs (ranks 53 to 71) | ||||
| #53 | kache | @yacineMTB | The Basement Labs | 2 of 5 agents, content, writing |
| #54 | vik | @vikhyatk | The Basement Labs | 5 of 5 local-models, content, search |
| #55 | Jeffrey Emanuel | @doodlestein | The Basement Labs | 5 of 5 agents, orchestration, rust, memory, verification, research-method |
| #56 | xjdr | @_xjdr | The Basement Labs | 4 of 5 agents, orchestration, local-models, evals, training |
| #57 | Michael P. Frank | @MikePFrank | The Basement Labs | 2 of 5 research-method, agents |
| #58 | fabian | @fabianstelzer | The Basement Labs | 3 of 5 content, mcp, business |
| #59 | Minh Nhat Nguyen | @menhguin | The Basement Labs | 3 of 5 evals, local-models, research-method, agents |
| #60 | Jack Morris | @jxmnop | The Basement Labs | 4 of 5 memory, search, security, research-method |
| #61 | Cody Blakeney | @code_star | The Basement Labs | 3 of 5 training, local-models, security |
| #62 | Danielle Fong | @DanielleFong | The Basement Labs | 3 of 5 evals, agents, memory, research-method |
| #63 | Shannon Sands | @max_paperclips | The Basement Labs | 2 of 5 training, agents |
| #64 | kalomaze | @kalomaze | The Basement Labs | 4 of 5 evals, training, local-models, memory, research-method |
| #65 | Simo Ryu | @cloneofsimo | The Basement Labs | 3 of 5 training, research-method, writing |
| #66 | Ellie Huxtable | @ellie_huxtable | The Basement Labs | 4 of 5 memory, orchestration, rust |
| #67 | Ivan Burazin | @ivanburazin | The Basement Labs | 3 of 5 security, agents, orchestration, business |
| #68 | orph | @orphcorp | The Basement Labs | 1 of 5 writing |
| #69 | Petr Baudis | @xpasky | The Basement Labs | 5 of 5 agents, orchestration, security, mcp |
| #70 | Florian Brand | @xeophon | The Basement Labs | 5 of 5 evals, agents, mcp, local-models, research-method |
| #71 | John David Pressman | @jd_pressman | The Basement Labs | 3 of 5 agents, training, evals, research-method |
| The Backrooms (ranks 72 to 83) | ||||
| #72 | Andy Ayrey | @AndyAyrey | The Backrooms | 2 of 5 agents, security, content, research-method |
| #73 | liminal_bardo | @liminal_bardo | The Backrooms | 2 of 5 orchestration, evals, security, agents |
| #74 | lumpenspace | @lumpenspace | The Backrooms | 3 of 5 security, training, agents, writing |
| #75 | Andrew (dremnik) | @dremnik | The Backrooms | 2 of 5 writing, business |
| #76 | thebes | @voooooogel | The Backrooms | 1 of 5 research-method, local-models, evals, security |
| #77 | Rob Haisfield | @RobertHaisfield | The Backrooms | 3 of 5 research-method, business, content |
| #78 | mephisto | @karan4d | The Backrooms | 1 of 5 agents, orchestration, training, mcp |
| #79 | Sauers | @Sauers_ | The Backrooms | 1 of 5 research |
| #80 | Mira | @_Mira___Mira_ | The Backrooms | 3 of 5 verification, evals, agents, research-method |
| #81 | nightwing | @yaboilyrical | The Backrooms | 1 of 5 search, memory, local-models |
| #82 | veryvanya | @veryvanya | The Backrooms | 1 of 5 content |
| #83 | 0.005 Seconds | @seconds_0 | The Backrooms | 3 of 5 research-method, search, content |
| The Builders on Main (ranks 84 to 99) | ||||
| #84 | Taelin | @VictorTaelin | The Builders on Main | 5 of 5 verification, agents, memory, research-method |
| #85 | Mario Zechner | @badlogicgames | The Builders on Main | 5 of 5 agents, orchestration, mcp, evals, security, writing |
| #86 | Lewis | @ctjlewis | The Builders on Main | 2 of 5 agents, business |
| #87 | dex | @dexhorthy | The Builders on Main | 5 of 5 agents, orchestration, evals, research-method, writing |
| #88 | bone | @bonegpt | The Builders on Main | 2 of 5 content, business |
| #89 | dax | @thdxr | The Builders on Main | 5 of 5 agents, orchestration, local-models, business |
| #90 | geoff (Geoffrey Huntley) | @GeoffreyHuntley | The Builders on Main | 5 of 5 agents, orchestration, verification, security, research-method, writing |
| #91 | Aaron Francis | @aarondfrancis | The Builders on Main | 4 of 5 orchestration, agents, business, content |
| #92 | Rob Pruzan | @RobKnight__ | The Builders on Main | 5 of 5 agents, orchestration, verification |
| #93 | Jesse Genet | @jessegenet | The Builders on Main | 2 of 5 agents, business, content |
| #94 | Raymond Weitekamp | @raw_works | The Builders on Main | 5 of 5 agents, orchestration, evals, rust, security, research-method |
| #95 | Nathan Baschez | @nbaschez | The Builders on Main | 4 of 5 agents, writing, verification, business |
| #96 | Nick Dobos | @NickADobos | The Builders on Main | 2 of 5 agents, evals, writing |
| #97 | Kyle Mistele | @0xblacklight | The Builders on Main | 4 of 5 agents, orchestration, mcp, security, local-models |
| #98 | sunil pai | @threepointone | The Builders on Main | 5 of 5 agents, orchestration, memory, business, writing |
| #99 | Adam | @adamdotdev | The Builders on Main | 5 of 5 agents, orchestration, mcp |
Number 100 is the curator’s open wild-card slot. Nominations go to @irl_danB, not to us.
Evidence labels
Claims carry the labels used across Franken Research. [Verified] means we read the page, file or API response ourselves. [Reported] means a source says so and we did not check the underlying thing; a role that comes only from a person's X profile is labelled [Reported: X profile] and paraphrased. [Inference] is our own reasoning. Each record also has an overall confidence (High, Medium or Low) and a list of what could not be verified. A [Verified] label that rests on an X API read or a GitHub license read names the file in evidence/ that holds the row, and every X post the study cites has a row there.
Dates
- List published by the curator: 2026-09-24.
- First-pass survey: 2026-09-24. Stars and activity dates are as returned by the GitHub API that day.
- Licenses of GitHub repositories: the authenticated GitHub license API, 2026-09-25, one row per repository in
evidence/licenses.jsonl. "No license file" means the API answered 404; "license file present, not recognised by GitHub" means GitHub could not classify the file. - Deep dives: 2026-09-24 and 2026-09-25, as dated on each page.
- Public edition, link check and review corrections: 2026-09-25.
What the public edition leaves out
- Public work first. Each page is about what a person has published: code, writing, papers, talks and products. Biography is limited to a sourced professional role where it explains the work, and nothing here judges anyone's character or states their motives.
- No raw X data. We do not republish profile text, pinned posts, locations or follower counts, and the raw X API responses are not in this repository.
evidence/x-reads.jsonlkeeps only what the study relies on: for a cited post, its text (with any email address or phone number it quotes replaced, and any withheld site replaced) and the URLs it links; for a pseudonymous account's profile, the URLs its record uses and nothing else. - Nothing personal. No home or work location, contact details, family or relationship details, health, financial or legal information about a person, private communications, images of people, or inferences about sensitive traits.
- No identity resolution. Pseudonymous accounts appear under their handle and the name shown on the list, with no gendered pronouns. Routes and handles are those on the curator's public list. A URL may appear in a pseudonymous record only under one of four rules, and gate N checks every URL against them:
- a. It is on x.com (or twitter.com) under the listed handle.
- b. It is a GitHub or Hugging Face account or repository whose owner is exactly the handle, ignoring case.
- c. The listed X account links it itself, one hop: its profile link, a link in its profile description, or a link in one of its own posts, with the read in
evidence/x-reads.jsonl. - d. A project page or README names the handle, with the quoted line in
evidence/credits.jsonl. - Never: personal sites or blogs (even ones named after the handle), biography, profession, employer or location, anything reached through a chain of more than one link, or third-party mirrors of X such as Thread Reader or Nitter. Where that leaves no work to list, the record lists none.
- Withheld everywhere. A site that a pseudonymous account links from its X profile and that its record does not use, and any site named after a pseudonymous handle, appears nowhere in the study: not in another person's record, a deep dive, an evidence file or a page. A claim about someone else that rested on such a site was given another primary source or dropped.
- Roles and shared affiliations are sourced. A person's role, or a statement that two people share a lab or employer, appears only with a primary source (their own site, repository or paper) or their own profile, labelled as such.
- No private material. Local file paths, machine names, process ids, internal ticket ids and the names of private repositories were removed or replaced with a short public description. Each record's list of where the person's work appears in our own files was replaced by one public sentence (
in_our_work) or omitted. - Nothing about our own deployment. Tools we use are named where they are public and studied on this site, but the study does not say how we configure or run them, what our own systems lack or have not checked, their internal sizes or thresholds, or our internal plans. Where a record once compared a person's work with our setup, it now states the general lesson from the person's work.
- Neutral on sensitive topics. Crypto tokens, politics and persona accounts are mentioned only where they are part of the person's public work, without judgement and without allegations.
- Pliny the Liberator (#7) is covered from public metadata and press coverage only, for defensive study. No repository contents were run or summarised; the only line quoted from a repository is its contact line, which names the handle (
evidence/credits.jsonl). - Security behaviour of third-party tools is stated only as documented behaviour, with a link to the project's own documentation or source, never as a way to reproduce it.
- Dead links. A link that returned 404 or 410 on 2026-09-25 was removed together with its claim, or kept and marked "link dead as of 2026-09-25" where the item itself still stands.
Limits
- This is a survey of public material, read in one or two days. Nothing was installed, run or benchmarked for the survey.
- Every relevance score is one reviewer's judgement, made by one of ten agent sessions that were not cross-calibrated. The scores say how much a person's public work bears on our own work; they do not rank people.
- Coverage of each person is uneven. Some people publish mostly code, some mostly writing, and some mostly on X, which we read only lightly.
- Stars and activity dates are as of 2026-09-24 and will drift. Licenses of GitHub repositories are as the GitHub license API reported them on 2026-09-25; other licenses are as the source states. Check the license file before reusing anything.
- Roles and affiliations can change quickly. A role labelled [Reported: X profile] was taken from the person's own profile and not checked elsewhere.
- The agents that wrote the records and deep dives are AI coding agents. Their work was checked by other agent sessions and by the build gates, not by the people described.
Corrections
If anything about you or your work is wrong or out of date, or you would rather not be covered here, open a correction with the "Independent 100 study" option: github.com/JYeswak/franken-research/issues/new?template=correction.yml. Every page also has a correction link that fills in its address. Accepted corrections are recorded in the repository changelog with the date. A request to remove a record is honoured without argument.
Request a correction
Is something about this study wrong or out of date, or would you rather not be covered? Open a correction (a GitHub issue form with this page’s address filled in). A request to remove a record is honoured without argument; accepted corrections are recorded in the changelog with the date.