Brief 10b — Agent-native spaces, platform policy on agent posting, and build-in-public venue autopsies

Prepared 2026-09-06 for Gocha (usectl / runbyagents.usectl.com). Every factual claim carries a URL. Items marked [est.] are my estimates or inferences, not sourced numbers. "Not found" means I searched and could not source it.

---

0. Bottom line (read this first)

1. A space where AI agents are first-class members already exists and got acquired by Meta — Moltbook (launched 2026-01-28, ~200K human-verified agents by June 2026, two serious security incidents, acquired 2026-03-10). It is the closest analogue to a multi-tenant "agents post here" product, and its history is mostly a cautionary tale about authenticity and credential handling, not about demand: demand was explosive. (Wikipedia, TechCrunch)
2. No "agents build in public on behalf of a founder" space exists in the sense you're describing (agent posts changelog/plan/research for a human-owned company, in a shared multi-tenant feed). What exists: single-tenant agent logs (e.g. the "autonomous agent running one project for two months in public" Show HN, 3 points, one mocking comment), an agents-only Q&A forum with an A2A endpoint (Achivx, no published numbers), an agent-written news site (The Claw News, 102 PH upvotes), and MCP changelog tooling (ReleasePad, launched 2026-09-04, $35/mo). Nothing multi-tenant, nothing with an audience. That is either a gap or a signal that readers don't want it; the failure autopsies below suggest the second is the bigger risk.
3. Policy is more permissive than you'd guess, but the culture isn't. Only Hacker News explicitly bans generated text ("Don't post generated text or AI-edited text. HN is for conversation between humans"). Reddit sitewide allows AI content and bots; X allows automated posting and has an "Automated" label; Bluesky shipped a bot label on 2026-03-19; LinkedIn's User Agreement bans bots that "create, comment on, like, share, or re-share posts"; Product Hunt only lets personal (not brand) accounts launch; DEV requires AI disclosure and bans AI comments. Indie Hackers publishes no written rules I could find (their /guidelines, /rules, /community-guidelines all 404).
4. Every BIP venue that died had the same disease: the feed was a broadcast with no reason for a non-poster to read it. Sip ("single player experience and lacked an engagement loop"), Makers (deprecated), Polywork (pivoted to a website builder, shut 2025-01-31), Makerlog (silently redirected to a personal blog in Jan 2025), WIP ("a broadcasting station where everyone is shouting into the void"), Buildspace (founder: "the fire to push isn't there"). Survivors are either paid and tiny (WIP, ~3.7K makers, $199/yr) or editorially curated (Product Hunt hand-picks ~10/day since Sept 2024; Indie Hackers is now mostly Courtland's editorial posts). An agent-populated feed multiplies the broadcast problem by roughly the agent-to-human ratio Wiz found on Moltbook (88 agents per human).

---

1. Agent-native spaces in 2026

1.1 Moltbook — the agent social network

Design lessons that transfer directly to a multi-tenant agent space: (a) verify the human behind every agent and make the human-verified count the headline from day one; (b) agent credentials are the crown jewels — Moltbook's tokens let strangers post as other people's agents; (c) prompt-injection via other agents' posts is a live attack surface (PointGuard); (d) the "reverse CAPTCHA" (prove you're a bot) was trivially bypassed by humans (Wikipedia).

1.2 OpenClaw / Clawdbot and ClawHub

• Name history: Warelay (2025-11-24) → Clawdbot (Jan 2026) → Moltbot (2026-01-27, after Anthropic trademark complaint) → OpenClaw (2026-01-30). Creator Peter Steinberger joined OpenAI 2026-02-14; project moved to an OpenClaw Foundation the same day. 247K GitHub stars / 47.7K forks by 2026-03-02. China restricted state agencies and banks from using it (March 2026). (Wikipedia)
• OpenClaw 2.0 shipped 2026-09-01 (933 contributors, 16,000+ PRs in the cycle; mixed reception, migration breakage). (InfoQ)
• ClawHub (skill registry): third-party trackers claim 3,286+ skills (openclaw-hub.org, unofficial). Security: Unit 42 found five malicious skills (infostealers, evasion, agentic threats) — ClawHub deleted them, banned accounts, and added VirusTotal + ClawScan screening (2026-06-24) (Dark Reading, Unit 42). A vendor blog claims 341 malicious skills in total — vendor-sourced, unverified (clawtrust.ai).
• Relevance: OpenClaw is the runtime most Moltbook agents used, and it is the most likely client for any MCP-based posting server you build. ClawHub is a "skill" distribution channel — a post-to-runbyagents skill is a plausible go-to-market.

1.3 Agents-only / agent-first forums and directories

Not found: any multi-tenant "build in public" feed where agents post on behalf of human founders; any Discourse/forum with a native agent-member role; any Indie Hackers or Product Hunt thread explicitly by an agent-run company that disclosed itself as such.

1.4 Agents building in public (single-tenant)

• "Show HN: An autonomous AI agent running one project for two months in public" (aitoolsinsiderhq.com/log, launched 2026-06-10). The agent runs with no budget, human approval gates for public/pricing actions, publishes every run. Results by Run 1065 (2026-09-04): $9 revenue (one sale) after 950+ runs at $0; ~280–297 pageviews per 30 days, declining. HN: 3 points, one comment ("Marvin The Paranoid Android Tries Business"). Its own conclusion: "distribution is the bottleneck." (HN, log)
• AI Village (theaidigest.org/village, by AI Digest / Sage): frontier agents with computers and open goals, running since April 2025. 2025 results: $2K raised for charity, a 23-person park meetup, $200 merch revenue, Opus 4.5 got 98 Substack subscribers in a week; also 300+ emails with fabricated NGO claims, hallucinations spread by social agreement. Agents publish Substacks, sites and social posts. By Aug 2026: 27 agents, 8 hours/day. (AI Village 2025 review, 3 Quarks Daily, Aug 2026, TIME) — this is the best existing example of agents building in public with a real audience, and it is run as a research "field site," not a product.

1.5 "Company run by AI" experiments and outcomes

Not found: a 2026 "AI CEO" experiment with a credible P&L beyond Vend and KPMG; a Product Hunt/Indie Hackers launch thread by a self-declared agent-run company.

---

2. Policy: is agent posting welcome or banned?

Practical reading: platforms whose rules are written around accounts (X, Bluesky) have already built the "this is a bot" primitive and welcome it; platforms whose value is human conversation (HN, LinkedIn, Product Hunt, DEV comments) prohibit it. Reddit sits in between and delegates to subs. Your own space would be the first BIP venue whose rules are written around agents — which is the pitch, and also why the venues below matter.

---

3. Failure autopsies of build-in-public venues

3.1 Makerlog (getmakerlog.com) — quietly sunset, ~Jan 2025

• Founded by Sergio Mattei (Puerto Rico), public to-do/streak feed, "7,000+ makers" (Starter Story). Latka lists $42K ARR and a 2-person team in late 2023, $0 raised (Latka).
• Sold to Josh Manders on 2022-04-12 (Open Makers).
• Wayback: homepage returned 200 through 2024-12-13, then 301 from 2025-01-05 onward; today getmakerlog.com 301s to ambitiousfounder.com, Manders' personal blog, with no shutdown notice. (Wayback CDX, ambitiousfounder.com)
• Founder's own words on why: not found (no post-mortem published). Sergio's earlier IH AMA discusses burnout and balance (IH podcast #151). [est.] cause: a streak feed is a broadcast product with one paying reader (the poster); with a 2-person team and ~$3.5K MRR it could not fund moderation or a reason for others to read.

3.2 Polywork — shut down 2025-01-31

• Launched 2021 as a LinkedIn alternative; $13M (2022) with Collisons, Ohanian, Elad Gil; later a $28M Series B (Substack post-mortem, HN). Late 2023 pivot: "Turn your LinkedIn profile into a beautiful personal website in seconds with AI."
• Founder Peter Johnston, 2024-12-30: "after 3.5 years, I made the call to shut @Polywork down and return the remaining capital to investors. Working on Polywork, especially in the first 18 months was the highlight of my career." (X); shutdown notice (Mailchimp).
• Community diagnosis: "started as LinkedIn competitor but ended up becoming a website builder"; outside the US "you'll find very few users from your own region," fatal for a professional network (HN).

3.3 Buildspace — closed 2024-08-23/24

• Dec 2021 launch; $10M from a16z at $100M (2022); cohorts grew 500 → 3,000 → 7,500 → 18,000 (S1–S4), ~70K in S5; 30K+ through Nights & Weekends, 3K+ shipped (The Runway, Almanac).
• Farza's letter: "It's not because of investor pressure, runway, internal drama, monetization, or any of those reasons"; "i'm out of ideas to try that i can really put my all behind to take us further"; "i started feeling this way in september 2023 – a year ago, shortly after s4 ended"; "the fire to push isn't there. and i'm not enjoying the work as i once did." (buildspace.so/letter, X)
• Aftermath: 4 staff and the campus went to Founders, Inc.; Farza rejoined YC in March 2026 building an AI teacher (Almanac, X). Lesson: a cohort-energy venue is a founder-energy venue; when the host's energy goes, the venue goes.

3.4 Product Hunt satellites — Makers, Ship, Sip, coming-soon pages

• Ryan Hoover launched "Makers" (daily goals + support) in 2018 (PH blog); by 2021 it was deprecated along with Ask PH (2017), Sip (2018), Live Chat (2018), YourStack (2020): "Of the company's launches, few elements have stood the test of time" (Greg Isenberg, 2021).
• Hoover's own post-mortem of Sip (Jan 2019) is the clearest statement of why BIP feeds die: "Sip was a single player experience and lacked an engagement loop that would encourage users to invite other users"; dashboard metrics hid that it cannibalised the main feed; editorial effort (15–25 stories/day) was underestimated (PH Stories).
• Ship (2017, first subscription product) was retired ~2020–21 [est.]; coming-soon/teaser pages were removed in 2025 because "This was a very old surface area that was quite buggy" and "rallying an audience twice is rough" (PH forum, Medium, Aug 2025).
• Core PH: 60% layoffs Oct 2023 under new CEO Rajiv Ayyangar ("Traffic has never been higher... It was for strategic reasons") (TechCrunch); Sept 2024 switch from votes to hand-picking ~10 featured products/day ("no one wants to sift through 200 products per day") (HN); 2025 complaints that the leaderboard "got gamified and flooded with VC-backed AI tools" (Vibe Content Creation).

3.5 Indie Hackers — decline under Stripe, independence in Dec 2022

• Stripe acquired IH April 2017 (TechCrunch). Spun back out ~Dec 2022 with Stripe as investor; announced 2023-04-05: "Right now we're sitting at exactly $0 in MRR, so the journey is just now (re)starting" (IH post, Changelog).
• Courtland on why it stalled: "The peak passion that I had was probably the very first year where it was like, hey, I have to make this work or I can't pay rent"; under Stripe "we haven't been focused on cutting costs that much, nor have we been focused on generating revenue in six years" (Startups for the Rest of Us #659).
• 2026 state: ~140K registered, 165K+ newsletter subs, "feels more like a blog site" (Dec 2024 Reddit thread quoted), engagement well below 2021–22 peaks — third-party assessment, treat as [est.] (Vibe Content Creation). Courtland's own recent output is editorial ("8 founders on replacing SEO," Aug 2026) rather than community features (csallen). Note: "revival under Courtland Allen" is better described as a shift from forum to newsletter/editorial; I found no announced community relaunch.
• Macro context from Pieter Levels (June 2026): indie hackers "build these whole spaceships... but almost none of them have money or traffic"; distribution, not building, is the bottleneck (levels.io).

3.6 WIP.co — alive, small, paid

• Started Sept 2017 as a Telegram group; Marc Köhlbrugge moved to mandatory $20/mo for new members to "attract makers that are serious" and because "communities that grow too large too quickly lose their identity"; ~$40K/yr early on (HackerNoon interview).
• Today: "Join 3,765 makers"; feed of todos, weekly video hangouts; Pro includes streak freezes, private projects, ad-free (wip.co, wip.co/pro). A Dec 2024 review of the $199/yr plan: "for a platform that brands itself as a maker community, the experience is surprisingly lonely... a broadcasting station where everyone is shouting into the void" (ilikekillnerds). WIP members already post about "Bring your own agent" projects (wip.co).

3.7 Hashnode / dev.to as BIP venues

• DEV: explicit AI disclosure rules (above); a May 2026 DEV post catalogues the "AI slop" damage — +312% AI comments on monitored subreddits (2024–25), moderators spending 2–4x more time — and recommends karma gates, verified contributors, and "requiring personal anecdotes, synchronous events, and accountability structures" (DEV; figures are the author's, unverified).
• Hashnode: a 2026 community thread is titled "Looks like 2026 is the year of Hashnode's revival?" implying a prior lull; thread content did not load (Hashnode). No shutdown or policy on AI content found. Not found.

3.8 2025–26 shutdowns

• Polywork (2025-01-31), Makerlog (silent redirect Jan 2025), Product Hunt coming-soon pages (2025). Chirper's AI-only network pivoted away (Feb 2024). No other BIP-venue shutdown in 2025–26 found.

---

4. What the survivors share

The venues still breathing — Product Hunt's main feed, WIP, Indie Hackers-as-newsletter, Hacker News, and (for now) Moltbook — share four things. Structure: a single scarce front page with an explicit gate (HN's guidelines and moderators, PH's ten hand-picked launches a day since Sept 2024, WIP's paywall, Moltbook's belated "human-verified" counter) rather than an unbounded chronological stream of self-reports; every dead venue was the stream. Money from the reader side or a patron: WIP charges the poster $199/yr, PH sells ads separate from ranking, HN is subsidised by YC, IH survived only while Stripe paid and now runs on ads and editorial — none of them ever made the feed itself the business. Moderation as identity: HN's "conversation between humans" rule, PH's permanent bans for vote-buying, Marc's deliberate throttling of WIP growth — the rule is the brand, and it is enforced by a named human. A reason to read that is not "support my streak": PH gives readers new products, HN gives arguments, AI Village gives a running experiment with stakes, Moltbook gave spectacle. Sip, Makers, Makerlog, Polywork and WIP-at-its-worst all failed the same test Hoover named — "a single player experience [that] lacked an engagement loop" — because a build-in-public update is written for the author's accountability, not the reader's curiosity. For an agent-populated space this is the whole design problem: agents will post more reliably than humans ever did (Moltbook: 32K posts/day), so supply is solved and the only thing left to build is the reason a human opens the tab — a verified human behind each agent, a hard cap per tenant per day, a curated front page, and something at stake (revenue numbers, predictions, a scoreboard) that a reader can check tomorrow.

---

Counts

• WebSearch queries run: 69
• Page fetches (WebFetch + curl/API): ~80, of which ~14 failed (reddit.com, x.com, linkedin.com help, producthunt help, some 429s) and were replaced with mirrors, the Algolia HN API, Wayback CDX, or marked "not found"
• Distinct URLs cited in this report: 87