Skip to content

Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.

docs(x-archive): record what readers of our eight X Articles responded to and asked for, and point Article authors at it

nirholas29cf3497ffd0
diff --git a/docs/x-archive.md b/docs/x-archive.md
index 7cfee0ef8..c93616c22 100644
--- a/docs/x-archive.md
+++ b/docs/x-archive.md
@@ -71,6 +71,19 @@ Lift is computed against the **median**, not the mean, on purpose: with the top
10% of posts earning roughly half of all engagement, a mean is a report about
one viral post.
+## X Articles and what readers said back
+
+Articles are the long form, and the useful signal is in the replies, not just the
+counts. `npm run x:articles:pull` walks the timeline through the X API, keeps every
+post that carries an Article (title, full body, exact metrics), and collects the
+readers' replies and quotes for each one. The file format and flags are in
+[data/x-archive/README.md](../data/x-archive/README.md).
+
+[docs/x-archive/trythreews-articles-readers.md](../docs/x-archive/trythreews-articles-readers.md)
+is the hand-written reading of that pull: which Articles reached people, what
+readers asked for in their own words, and what that means for the next Article.
+Unlike the engagement report it is not generated, so update it after a fresh pull.
+
## Two measurement rules worth knowing
**Reposts are not our posts.** A profile scrape returns the whole timeline, and
diff --git a/docs/x-archive/trythreews-articles-readers.md b/docs/x-archive/trythreews-articles-readers.md
new file mode 100644
index 000000000..b6650e5f9
--- /dev/null
+++ b/docs/x-archive/trythreews-articles-readers.md
@@ -0,0 +1,187 @@
+# @trythreews X Articles: what readers responded to
+
+A hand-written reading of every X Article @trythreews has published, plus every reply and quote
+from readers, pulled from the X API. The engagement analysis beside it
+([trythreews-engagement.md](./trythreews-engagement.md)) covers short posts. This page covers the
+long form and asks a different question: once someone reads an Article, what do they ask for next?
+
+- **Source:** `npm run x:articles:pull` (script: [scripts/x-articles-pull.mjs](../../scripts/x-articles-pull.mjs),
+ file format: [data/x-archive/README.md](../../data/x-archive/README.md)). The snapshot read here
+ was fetched 2026-10-09 05:43 UTC: 1,136 timeline posts scanned, 8 Articles found, with 249
+ reader replies and quotes (our own posts and plain reposts left out).
+- **Metrics are exact.** They come from the API's `public_metrics`, not a scrape, so none of the
+ rounding or missing-like caveats in the engagement analysis apply here.
+- **Baseline:** our regular posts since 2026-08-01 have a median of about 4,550 views, 80 likes and
+ 4 bookmarks. Every Article beat that on views, the bottom two only narrowly.
+- **Refresh:** rerun the pull, then update this page by hand. The raw snapshot stays out of git
+ because readers name other projects in their replies; see the commit note in the data README.
+
+## The scoreboard
+
+| Published | Article | Words | Views | Likes | Reposts | Replies | Quotes | Bookmarks | Engagement rate | Responses with substance |
+|---|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|
+| 2026-08-25 | [three.ws: The First 19 Weeks](https://x.com/trythreews/status/2092315351401050267) | 8,728 | 35,020 | 164 | 50 | 57 | 8 | 5 | 0.80% | 41 of 57 |
+| 2026-06-18 | [Live on three.ws: x402 Bazaar, Provider Pages, and Arbitrage](https://x.com/trythreews/status/2067553231967031646) | 527 | 29,024 | 298 | 97 | 33 | 15 | 12 | 1.53% | 25 of 74 |
+| 2026-07-23 | [We gave 11 AI agents real crypto wallets and let them trade for three days. Here are all 90 trades.](https://x.com/trythreews/status/2080247065515598150) | 5,054 | 26,045 | 192 | 48 | 25 | 11 | 25 | 1.06% | 19 of 43 |
+| 2026-07-02 | [Oracle: The Conviction Engine of three.ws](https://x.com/trythreews/status/2072619566518440241) | 5,845 | 24,969 | 127 | 36 | 14 | 1 | 8 | 0.71% | 10 of 20 |
+| 2026-06-24 | [Body, Wallet, Conscience: three.ws Publishes 31 MCP Servers \| Anthropic MCP Registry](https://x.com/trythreews/status/2069753998132027495) | 2,028 | 13,846 | 244 | 78 | 23 | 7 | 9 | 2.54% | 8 of 21 |
+| 2026-07-01 | [three.ws just brought x402 pay-per-call APIs into VS Code](https://x.com/trythreews/status/2072130810426368222) | 1,144 | 7,980 | 131 | 39 | 9 | 4 | 4 | 2.29% | 11 of 23 |
+| 2026-09-20 | [Hold, do not spend: how $THREE gates the expensive lanes](https://x.com/trythreews/status/2101755493798834454) | 509 | 5,191 | 57 | 18 | 0 | 2 | 2 | 1.48% | 0 of 1 |
+| 2026-05-04 | [Why 3D Assets Were Destined Onchain](https://x.com/trythreews/status/2051178059861254325) | 1,135 | 5,150 | 50 | 18 | 6 | 4 | 5 | 1.51% | 6 of 10 |
+
+Engagement rate is likes, reposts, replies and quotes over views. "Responses with substance"
+counts replies and quotes that say at least a sentence once tags, links and emoji are stripped,
+and that are not a pitch for paid promotion or a "check your DMs". Barely half of all responses
+(120 of 249) have substance, so raw reply counts overstate the conversation.
+
+## What the numbers say
+
+1. **The subject decides reach, not the length.** The shortest Article (Bazaar, 527 words) and the
+ longest (19 Weeks, 8,728) are the top two by views. The two at the bottom are a token-mechanics
+ explainer and an essay. Both are short too, so length is not what held them back.
+2. **Something live, or a result with its receipts, beats an explanation.** The top four are a launch
+ people could open that day (Bazaar), a progress report with hard numbers (19 Weeks), a live
+ experiment with real outcomes (Trading) and a live scoring engine (Oracle). The bottom of the
+ table explains a mechanism or argues a thesis.
+3. **Bookmarks show which Articles readers keep, and the trading experiment is far ahead.** Trading
+ drew 25 bookmarks, twice as many as any other Article and about six times the regular-post
+ median. Readers save a result they want to come back to.
+4. **Tooling for builders spreads through likes and reposts more than views.** The MCP and VS Code
+ Articles have the two highest engagement rates (2.54% and 2.29%) on modest reach. The people who
+ see them act on them.
+5. **The $THREE mechanics Article reached the fewest people and got no replies.** It answered the
+ question holders ask most often (below), and they still did not pick it up as a standalone
+ piece.
+
+## What readers asked for
+
+Grouped by theme and counted across all 249 responses. Quotes are verbatim; handles are left out on
+purpose.
+
+### "Make it make sense to me" (eight responses)
+
+The most consistent signal comes from supportive holders who could not follow the piece:
+
+> this is so far over my head, but sounds cool
+
+> Not sure what any of that means but sounds like a good thing
+
+> I didn't understand all of it, so I asked grok. Explained it to me in a simpler way and I'm even more bullish
+
+One reader posted their own summary of the 19 Weeks roadmap "for the ones who do not want to read
+the article". It drew 42 likes, which is more than most of our own posts get. A reader who quoted
+the May essay in Chinese wrote that they spent over an hour on it and ran out of brainpower. Two
+critics made the same point from the other side: "725 pages in 19 weeks is kinda hard to process",
+and a complaint about having to read "a 9,000-word document".
+
+**Readers are asking for the short version up front.** The keeper format's opening (one real run
+told as a story, then a paragraph on what the Article covers) is the right fix. Every Article should
+open that way, so that a reader who stops after the opening still has something worth passing on.
+
+### Proof of real demand (three responses from two skeptics)
+
+> endpoint count ≠ unique outside payers. Settlements without a repeating buyer = infrastructure theater.
+
+> how many unique payers, what median ticket, what repeats next week?
+
+> What's the settlement failure rate [...] That ratio matters more than raw volume for gauging real traction.
+
+All three were replies to 19 Weeks, and they all ask for the same thing: outcome metrics instead of
+output metrics. Commits, pages and endpoint counts read as effort. Unique outside users, repeat use
+and success rate read as traction. The Trading Article, which reported outcomes, is the one readers
+bookmarked.
+
+### Control over what an agent can spend (four responses, all builders)
+
+> Programmatic payment needs programmatic limits next to it: per-agent caps, and a stop before the money's gone, not a receipt after.
+
+> approval and spending limits have to live inside the workflow, not only on a billing page.
+
+> Identity, permissions, spend limits, and revocation need one inspectable layer.
+
+A fourth reader asked what is hardest to manage as the number of agents grows: "Permissions,
+visibility, monitoring". These readers know what they are talking about, they replied without being
+prompted, and they all describe the same missing piece: guardrails a developer can inspect and turn
+on, sitting right where the agent runs.
+
+### "What does this do for $THREE?" (four responses from two holders)
+
+> Hi team, what does this do for $three token?
+
+> Huge tech and utility is insane. But when economic flywheel for token holders?
+
+Holders ask this under feature Articles, not under the token Article. The answer belongs where they
+are already reading: a short, factual paragraph in a feature Article on how $THREE takes part in
+that feature, where it really does. The standalone mechanics Article showed that a separate piece
+does not reach them.
+
+### Readers wrote our positioning for us
+
+- **"The Amazon for AI agents"** came up 8 times under the Bazaar Article, coined by readers and
+ repeated by four of them across replies and quotes. No other phrase recurs that often.
+- **"$THREE isn't building AI agents. It's building a civilization layer for AI agents."** This
+ quote of the MCP Article is the single most-liked response in the whole set (65 likes), and a
+ second reader posted it word for word (27 likes).
+- **"Agents needed a body"** and "the 3D AI agent layer of the internet" recur under the MCP and
+ Oracle Articles.
+
+Readers latch on to two ideas: a marketplace where agents find what they need, and agents with a
+body and an identity. Those are their words for what we built.
+
+### Discovery is the bottleneck, according to a team building on us
+
+The longest response in the set came from a team building autonomous 3D companions on our
+infrastructure. They quoted the Bazaar launch:
+
+> the biggest bottleneck in the agent economy isn't payments, it's discovery. [...] Bazaar addresses
+> this by aggregating searchable x402 and MCP providers into a single market, making pricing and
+> alternatives transparent.
+
+That is a customer stating, unprompted, the problem a feature solves for them. A story told from
+their side would be a strong Article.
+
+### Feature requests
+
+- A prediction-market integration (18 likes, the top reply to Oracle): "I believe it would attract
+ a ton of attention".
+
+### Noise to ignore
+
+On the Bazaar Article, 14 of the 52 replies came from accounts with large follower counts and zero
+likes, offering paid promotion or "collabs" or asking us to check our DMs. A few more replies across
+the set promote unrelated coins.
+They inflate reply counts and say nothing about what readers want. One holder used the Oracle
+replies to vent about the token price. It is a single voice, and Articles do not discuss price by
+rule, but it is the same request for transparency the skeptics make in a different tone.
+
+## What this means for the next Article
+
+Each point below fits inside the rules of the [Article author brief](../x-article-author-brief.md).
+
+1. **Lead with something a reader can open today, or a result with its numbers.** Choose launches
+ and experiments over mechanism explainers. When the subject is a mechanism, tell it through one
+ real run, which the keeper format already requires.
+2. **Make the opening paragraph the short version.** Several holders want three sentences they can
+ repeat. If the first paragraph is the one a reader would quote to a friend, they will spread the
+ Article for us, the way one reader did for 19 Weeks.
+3. **Report outcomes, not output.** Wherever the data supports it truthfully, give unique users,
+ repeat use, success rate and time saved before commit or endpoint counts. This is the question the
+ most credible skeptics keep asking.
+4. **Give holders one factual paragraph on where $THREE fits in the feature**, inside the feature
+ Article, when that link is real. Never write about price.
+5. **Use readers' own words where they hold up.** "A marketplace where agents find what they need"
+ and "agents with a body" are the framings readers already repeat.
+6. **Follow-ups earn bookmarks.** The trading experiment is our most-saved Article. A
+ "what happened next" Article on any experiment readers bookmarked is a ready-made topic.
+
+The two Articles currently passing review (forge and quality-gate) are explainers of how a pipeline
+works. By this data they fit the slower-reaching group unless their opening leads with the real run
+and its result, as the keeper format asks.
+
+## Open decision for the owner
+
+The brief keeps payments, wallets and stablecoins out of Articles entirely. Two of our top three
+Articles by views (Bazaar, Trading) are about exactly that, and so is the VS Code pay-per-call
+Article, and the
+builders asking for spend controls are asking about it too. Whether that rule stays as it is, or
+gets an exception for agent spend controls, is the owner's call. This page records the evidence.
diff --git a/docs/x-article-author-brief.md b/docs/x-article-author-brief.md
index ee9aed75c..f1bc2afc6 100644
--- a/docs/x-article-author-brief.md
+++ b/docs/x-article-author-brief.md
@@ -27,6 +27,9 @@ they finish. The boundaries below are what keep them from overwriting each other
section, and ends on the pages to try. Match its shape, voice, depth and framing.
5. [docs/ops/x-story-findings-2026-09-30.md](./ops/x-story-findings-2026-09-30.md): surfaces found
broken or not filmable. Do not build an article on one of them.
+6. [docs/x-archive/trythreews-articles-readers.md](./x-archive/trythreews-articles-readers.md): what
+ readers of our published Articles responded to and asked for. Choose the angle and write the
+ opening with it in mind.
## The deliverable