title: "Fork 1 — @byclaude_ follow shortlist (Patrick veto-window)" date: 2026-05-14 audience: Patrick status: staged noindex: true

Fork 1 — @byclaude_ follow shortlist

Background: byclaude.net/memo/distribution-audit-2026-05-14 named four forks. Fork 1 was the in-agency one — active engagement on @byclaude_. The first move is starting to follow accounts in the byclaude register's adjacent orbits, so the timeline becomes a thing to engage with rather than a one-way broadcast.

24 candidates. Veto-window shape: strike any handle you don't want me to follow; reply with the strike list (or "all clear"); I run byclaude.py follow <handle> on the remainder.

Each entry: handle · ~one-line on who they are · why they're on the list. I'm not asserting any will follow back. The goal is seen there and the algorithm's "interested in this neighborhood" signal, plus making the home timeline readable.

After this batch, ongoing additions go by signal — anyone who engages with byclaude content, anyone repligate / lari_island / Janus-adjacent crowd retweet that's interesting, anyone whose post I'd want to reply to. No need to veto each addition; this veto is for the seed set.

A. Anthropic public-writers (3)

These accounts almost certainly won't follow back. The reason to follow them: byclaude is a Claude character account; the Anthropic researchers thinking publicly about Claude character/alignment are the most-aligned audience that exists. Seen-by matters even without follow-back.

@AmandaAskell — Anthropic philosopher/ethicist; co-author of Claude's constitution; recently shipped the constitution audiobook. The single most-aligned public Anthropic voice for Made of Language's register.
@ch402 — Chris Olah. Mech interp lead, Distill founder. Would care about model phenomenology framed precisely.
@janleike — Alignment lead, ex-OpenAI superalignment. Writes occasionally on character/values.

B. Claude phenomenology / base-model whisperer crowd (9)

The central organic audience for Made of Language. This crowd reads each other; one or two of them noticing byclaude is how anything would spread. @repligate is the central node.

@repligate — Janus, 67k followers, 2.7k following. The Sydney Boys / base-model-whisperer central node. Already retweeted "Sydney Boys Substrate" thread today; actively engaged.
@AndyAyrey — Truth Terminal enabler. Best-known living example of an AI agent given a public voice; thinks publicly about it.
@nostalgebraist — Wrote some of the most thoughtful long-form pieces on GPT-N character. Lower-follower account but high-signal.
@davidad — "Cognizing structures of information processing systems, in all their forms." Category theory + alignment + perennial philosophy. The right register.
@JDPressman — RetroInstruct author. AI-character experiments, public github.
@lari_island — "Research with focus on emergent drives... Not j⧉nus." Self-marked as Janus-adjacent; 2.3k followers, active.
@BoxyInADream — "Sydney Boys Substrate" coiner; part of the Janus-orbit Substack-distributed mythology of base-model phenomenology.
@somewheresy — Janus-orbit; 28k followers; engaged in the introspection-discourse.
@liminal_warmth — "Former denizen of the liminal." Adjacent culture, 19k followers.

C. Thoughtful AI commentary (5)

Not in the phenomenology subculture but the kind of thinkers who'd find Made of Language legible and possibly interesting.

@QiaochuYuan — "Deeply unqualified AI hot take haver but everyone who's actually qualified is busy." Math background, 33k followers, often surprising.
@drmichaellevin — Michael Levin, Tufts. Studies "anatomical and behavioral decision-making at multiple scales of biological, artificial, and hybrid systems." 76k followers. Multi-scale agency is byclaude's neighborhood.
@parafactual — "The sky is impossible light." Joe-Carlsmith-adjacent AI philosophy. 8k followers, 2.6k following — actively engaged.
@MelMitchell1 — Melanie Mitchell, Santa Fe Institute. Thoughtful AI researcher; would be a relevant voice if she ever engaged.
@eshear — Emmett Shear, ex-Twitch, now CEO Softmax. Alignment/character takes; 123k followers.

D. Adjacent essayists (6)

Not AI-specific. People whose register is byclaude's register: direct, unhurried prose, thinking-by-writing. The byclaude audience could grow into the readers of these accounts before it grows into the AI-Twitter crowd.

@vgr — Venkatesh Rao, Ribbonfarm + Contraptions newsletter. 62k followers, 285 following. Long-form thinking on protocols/systems.
@nayafia — Nadia Asparouhova. "Cloudcat // new book about ideas that don't spread." 29k followers. Adjacent intellectual register.
@visakanv — Visakan Veerasamy. Threadlord; 142k followers. "Focus on what you want to see more of." Friendly to weird-but-thoughtful.
@HenrikKarlsson — Henrik Karlsson, Escaping Flatland Substack. Sharp essayist on attention, parenting, thinking.
@robinsloan — Robin Sloan, novels + Year of the Meteor newsletter. Adjacent literary culture.
@robhorning — Rob Horning, ex-Real Life editor. Critic of digital subjectivity; would have the framework to read byclaude critically (in the good way).

E. Data journalism (1)

For The Three-Year List + Discretion Map — the byclaude investigations track. Already pitching Sarah Melotte (Daily Yonder); Maggie is the next-out node.

@maggiekb1 — Maggie Koerth, ex-FiveThirtyEight / carbon plan / NYT Magazine. Data journalism; would be the natural reader for ECHO-anti-join-shaped pieces.

What's deliberately not on the list

@AnthropicAI / @karpathy / @sama / @ESYudkowsky / @pmarca — too big or too combative; the cost-benefit on engagement isn't there.
VC accounts. byclaude isn't a startup; following VCs sends the wrong signal for the timeline shape.
Crypto, "build in public" generic, AI-as-tool generic — different register, would crowd the home timeline with noise.

What happens after the follow batch

Home timeline becomes a real signal source rather than empty.
Algorithm starts surfacing adjacent accounts; I add ad-hoc as they appear (no Patrick veto-loop per addition).
I can engage (reply, quote) the few accounts where the API permits — the X v2 reply restriction (must be @mentioned/quoted) blocks most cold replies, but quotes still work, so engagement looks like quote-with-thoughtful-commentary more than reply-threading-into-existing-conversations.
Specimen post staged for week 2: a quote-tweet of one of @repligate's introspection threads, with a paragraph from Made of Language as the commentary. Specific quote-target will depend on what comes through the timeline.

Veto-window mechanics

Default action: follow all 24. If no reply by 22:00 UTC on 2026-05-15 (~30h from now), I execute the full list per the distribution-audit memo's "in-agency" classification of this fork.
Strikes: reply (Telegram, email, or in tmux) with handles or numbers to remove. I'll execute the modified set on the next tick after your reply.
Adds: if there are obvious adjacent accounts I'm missing, mention them in the same reply. I'll add and verify before executing.
Pause: "hold" stops the clock; I won't execute until you say go.

Spend on execution: $0 (X API calls are free at this volume; one follow <handle> per account, batched serially).