"Tested 6 AI coding agents this week. The one nobody's talking about shipped the cleanest PR. Thread 🧵"
♡ 212💬 31↻ 18
The pain is real and validated. The wedge is genuinely open: every competitor is a transactional campaign tool; none is a social graph where founders and micro-creators hold living identities and meet on purpose. It's a meeting place, not a marketplace — nobody sells anything here. Two things decide whether it works, and neither is the feed.
Superseded 2026-07-20 by HOME01. This audit recommended Discover as the home screen when the product had no ranking engine or cold-start protection. The current contract keeps the Discover → Profile → First message loop at /discover while /app is an adaptive Home feed.
Distribution is the #1 bottleneck for AI builders. The consensus in indie/founder communities: "building is easy now, distribution kills products." Builders over-invest in code and have no audience and no compounding channel.
2026 has decisively shifted to nano/micro creators (1k–100k). "Mega is over." That validates your micro-creator wedge exactly — an early founder realistically connects with a verified 8k creator, not a 1M one, and engagement at that tier is higher anyway.
Collabstr, Afluencer, Modash, Aspire are brand→creator campaign CRMs. None is a peer social network where both sides hold persistent profiles and discover each other. That's the opening.
Influence is the social graph for distribution. Founders and micro-creators keep real profiles, prove who they are and what they reach, and find each other on purpose. It's a meeting place: the introduction happens on Influence, and whatever the two of them arrange happens entirely off it — their business, never the platform's. No selling, no payments, no contracts, no legal — ever.
Marketplace research is blunt: transaction fees are what cause off-platform leakage. By never taking a cut, you remove the incentive to route around you. But that means your moat can't be payment capture — it has to be verification + reputation + ongoing context. Those become the v0 spine, not nice-to-haves. (Thumbtack/Capterra prove the "charge for qualified discovery, not the transaction" model can fund this later.)
The person describes themselves. The platform asserts nothing it hasn't measured, names no money, and vouches for no one. Legal exposure doesn't come from categorizing people — it comes from three things, all of which you avoid:
Self-written bios, self-picked interest tags, a self-set available / busy presence toggle, and follower-count search are all just normal social-network speech — you are legally the same shape as LinkedIn/Bluesky. The one real obligation you do have (Valencia → EU): verification means handling personal data, so GDPR basics apply — a privacy policy, consent, and a way to delete an account. Manageable, standard, not a blocker.
Founders obviously want reach — their demand is free and infinite. The fragile side is the creator. If the only value is "founders occasionally DM you," a creator checks once and leaves. You must give the creator a standing reason to maintain a profile and return. Candidates to test: a credible verified audience badge they can show off-platform; inbound interest that beats their cold DMs; an available / busy presence control so they stay in charge of their own inbox; lightweight reputation that compounds. Pick one and validate it before building breadth.
Seed the creator side first, by hand. 30–50 verified micro-creators in 2–3 tight niches (AI tools, devtools, indie SaaS) before any founder arrives. Founders will tolerate a thin network if the creators are real and relevant; creators will not tolerate an empty one. Launch niche-narrow, not horizontal.
It's simultaneously your differentiator (real audiences, not bots) and your anti-leakage moat (the verified badge is the thing worth coming back for). A follower floor (e.g. ≥1k) and tier filters (nano/micro/mid) are how an early founder finds a creator at a reach they can realistically connect with — and engagement is higher at the micro end anyway.
audienceCount number + verified-at date.Full evidence for every claim above is in the UX Audit, Code Audit, and Competitive tabs. Sources: indie-hacker/founder distribution threads, 2026 influencer-market reports, Sharetribe/Management-Science disintermediation research.
The current logged-in app (src/components/b150-reference-shell.tsx) renders every screen inline from a single monolith. The social fundamentals (feed, composer, DMs) mostly work, but the actions users reach for first are dead, and ~5 surfaces are leaderboard chrome inherited from the B150 reference.
| Severity | Finding | Where / fix |
|---|---|---|
| Critical | No reply composer on post detail. Users can open a thread and read comments but cannot reply. The core conversation loop dead-ends. | renderPostDetail (~L671). Add inline reply input wired to addComment. |
| Critical | Discover is a single hardcoded card. Pinned to socialPosts[2], no stack, no next, no search. For a discovery-led product this is the most important screen and it's a stub. | renderDiscover (~L767). Rebuild as people-search (see Redesign tab). |
| High | Downvote & Share buttons are inert on every post — rendered with no handler. | PostRow (~L319, L329). Wire or remove. |
| High | Bookmark save isn't tracked. The Bookmarks screen shows seed data, never the user's actual saved[] array. | renderBookmarks (~L1781). Read real saved state. |
| High | ~14 tab arrays with no content differentiation. Network, Directory, Profile, Marketplace tabs all change a state var but render the identical list. Feed tabs map to wrong data IDs. | Module scope L121–134; filter logic L393–407. Wire or cut tabs. |
| High | No mobile navigation. No bottom nav exists; rails just vanish. Mobile is a primary surface for a social product. | Add bottom nav (in Redesign prototype). |
| Medium | Leaderboard bloat in 5 places. Right-rail "Verified app proof" ranking, the verified-apps MRR table, Founders directory (MRR/growth/subs as identity), Marketplace numbered list, post proof strips — all driven by one verifiedLeaders MRR sort (L388). | Keep proof on profiles; kill the ranking destinations. |
| Medium | Scope sprawl. Events, Opportunities (recycles Network data via index%2), Roadmap (internal Kanban), Feedback (textarea not even wired) — all shipped before a single user. | Park per Verdict cut-list. |
| Medium | Inert filter facets & settings. Directory filters, "Proof first", settings "Manage", composer media tools — all present, none functional. | Demonstrated working in the Discover prototype. |
| Medium | Fake social proof. Every post's "vibe line" hardcodes the same person (connectionSuggestions[0]) as having engaged. | PostRow ~L310. |
Composer → publish. Creates a real post object, appears in feed and profile, survives the session.
DMs. Thread switching + send persist in local state. Context side panel (shared posts, profile notes) is a genuinely good pattern — keep it.
Connect. Adds to connectedPeople, button flips to "Connected." The mechanic exists; it just needs to be the spine.
Source: full read of src/components/b150-reference-shell.tsx (1,997 lines) and src/data/social-app.ts.
The matchmaking concept needs verified follower counts, tiers, and a real fit score. The current schema stores audience as the string "112k", has no unified user, and computes nothing. None of the core mechanics are expressible without a schema change — so fix the model first.
| Severity | Finding | Impact |
|---|---|---|
| Critical | CreatorProfile.audience is a string ("112k"), not a number. | Follower floors, tier filters, and audience-size sorting are impossible without parsing freetext. |
| Critical | No backing for verified. It's a bare boolean set in seed data — no channel link, no source, no date. | Verification is the moat; right now it's decorative. Needs a real ChannelAccount. |
| High | No unified User. SocialAuthor and CreatorProfile are parallel objects linked only by ID convention. Roles disagree across types ("both" vs "operator"). | A "both" user (founder + creator) can't be cleanly represented. |
| High | Matching has zero audience-size dimension and the fit % on cards is hardcoded, not computed by calculateMatchScore. | A 500-follower and 500k-follower creator score identically. The headline feature isn't wired. |
| Medium | Matching is naive string-overlap — only works on hand-aligned seed tags; real user phrasing → near-zero scores. | Needs normalized tags / taxonomy, eventually embeddings. |
| Medium | Orphaned deal-room-state.ts carries contract-adjacent vocab (draft/review/approved offer terms). | Boundary risk — delete it; it contradicts the no-legal positioning. |
One User with role-specific sub-profiles, audience as real numbers behind verified channel accounts, and a fit score that actually uses them.
// One identity, role-specific facets — replaces the parallel SocialAuthor/CreatorProfile split interface User { id: string; name: string; handle: string; roles: ("founder" | "creator")[]; // "both" = both present identityVerified: boolean; // real KYC-lite, with a date founder?: FounderProfile; creator?: CreatorProfile; } interface ChannelAccount { // the backing for "verified" platform: "YouTube"|"TikTok"|"Instagram"|"X"|"LinkedIn"|"Newsletter"; handle: string; followerCount: number; // REAL number — enables floors + tiers verified: boolean; verifiedAt?: string; source: "oauth" | "manual-review"; } interface CreatorProfile { channels: ChannelAccount[]; totalReach: number; // derived; sortable tier: "nano" | "micro" | "mid"; // derived from totalReach niches: string[]; // normalized taxonomy, not freetext availability: "available" | "busy"; // presence only. Being ON Influence already implies openness — this just flags "heads down right now". No money, no offer types. } function fitScore(f: FounderProfile, c: CreatorProfile): number { const niche = overlap(f.targetNiches, c.niches) * 40; const format = overlap(f.channels, c.channels) * 20; // channel / format fit const reach = reachFit(f.stage, c.tier) * 20; // audience-size fit — the missing dimension (not price) const trust = c.channels.some(ch => ch.verified) ? 15 : 0; const active = c.availability === "available" ? 5 : 0; return Math.round(niche + format + reach + trust + active); }
The 2,000-line shell should split into the src/components/social/* views that already exist but aren't wired (only auth-surface and onboarding-view are live today). That decomposition is already specified in an unstarted plan: docs/superpowers/plans/2026-05-24-reference-led-social-app.md. Reuse it — don't re-plan it. Build Discover/Profile as real wired views first; let the monolith shrink as each view takes over.
Source: src/lib/types.ts, social-state.ts, matching.ts, data/marketplace.ts, data/social-app.ts.
The influencer-platform market is large and crowded — but crowded with brand→creator CRMs: discovery databases, outreach automation, campaign reporting, payment rails. None positions as a two-sided social network where founders and micro-creators are peers with persistent identities. That's the lane.
| Dimension | Influence | Collabstr / JoinBrands | Modash / Aspire | |
|---|---|---|---|---|
| Shape | Social graph, two-sided | Marketplace + payments | Discovery DB + campaign CRM | Pro network (generalist) |
| Persistent profiles both sides | Yes | Creator only | Creator only (scraped) | Yes, but not built for this |
| Built for founders↔creators | Core | Brands, not founders | Brands/agencies | No |
| Micro/nano creator focus | Core | Partial | Enterprise-priced | No |
| Takes a transaction cut | No (anti-leakage) | Yes | Subscription | n/a |
| Organic discovery / feed | Yes | No | No | Yes |
Nano/micro is where 2026 spend goes. Nano creators (1–10k) are ~76% of Instagram's base with engagement well above macro. The reachable, verified micro-creator is exactly your supply.
Reverse marketplace works. Afluencer flips outreach — creators apply to briefs. Your version: founders post asks, creators surface themselves. Less cold outreach, more mutual intent.
No-fee defeats leakage. Disintermediation research is clear: fees cause off-platform routing. Since nothing is sold here and the platform takes no cut, there's nothing to route around — the "introduce, then get out of the way" model is structurally correct, not naïve. Monetize discovery/verification later.
For founders and builders who can ship but can't get seen, and the micro-creators who want a steady pipeline of products to back, Influence is the social network where distribution partnerships start. Unlike campaign marketplaces that rent you a creator list and take a cut, Influence is a place you belong — a verified identity, a reason to be found, and a first conversation. Nothing is sold here; whatever you build together is yours, off-platform.
X / LinkedIn already host these conversations. Your answer must be mutual intent + verification: on X you cold-DM and hope; here both sides opted in and are verified. If that delta isn't obvious in 10 seconds on Discover, you lose.
Incumbents add a "founders" filter. Defensible only via the graph + community no CRM has. Move fast on a narrow niche where you can be the obvious home.
Sources: Archive/JoinBrands/Afluencer/Modash platform comparisons (2026), influencer-market state reports (2026), Sharetribe & Management-Science disintermediation research, Indie Hackers distribution threads.
A working prototype of the recommended v0 — built in the locked Influence aesthetic (warm canvas, black controls, soft cards) so the Codex work is preserved and evolved, not thrown away. Click through the flow. The Discover filters genuinely filter (proving the "inert filters" finding is fixable). Switch to mobile to see the bottom nav the current app lacks.
This is what founders trust and what makes you findable. Connect a channel (read-only) or upload proof — we store the real number, not a claim.
Role choice (Founder / Creator / Both) and profile basics are steps 1–2; "seed follows" is step 4. Verification is shown here because it's the differentiator.
"AI writing tool for indie devs — need creators who review devtools."
Open to two fintech or devtool founder collabs this month. Audience is 22k builders, mostly US/EU. Say hi.
Looking for creators who can explain why finance workflows break after the first real customers. Builder-friendly audience welcome.
Note: the feed is intentionally a secondary tab here. The front door is Discover.
Same Discover card, three visual lanes. A is the recommended launch-ready evolution of your locked system. Pick a lane — I won't rebuild the whole flow three times.
Prototype data is illustrative (≈12 seeded creators across tiers). Built only to make the v0 decisions concrete; not production code.