| Round | Thesis | Key Change |
|---|---|---|
| v1 | “AI agent that screens incoming members for WhatsApp group admins” | WA-only framing constrained TAM, distribution, and verdict |
| v2 | “Agentic vetting service that uses AI agents to verify people have certain attributes, as infrastructure for community gating — on any platform” | Platform-agnostic. WA is one channel, not the product. Buyer changes from casual admin → community operator / platform. |
Eric runs multiple WhatsApp groups (GenieFriends, angel-era groups, Meta Threads pipeline — ~5–20 groups). He is a community admin who manually vets join requests. This is the strongest PMF signal.P0
v1 sized the market as “WhatsApp admin tooling ≈ $0.” That was accurate for WA-only but missed the point. The real market is community gating infrastructure across all platforms.
“AI community vetting” is not a tracked market segment — this hasn’t changed from v1. But the adjacent segments are significantly larger than previously cited, and the overlap is the opportunity.
| Layer | Market | Size | Avet’s Slice |
|---|---|---|---|
| Formal KYC/IDV | Identity verification (Jumio, Onfido, Veriff, Persona) | US$16B1 | avet is “soft KYC” — qualitative, not document-based. Different buyer. |
| Content moderation | Hive, Spectrum Labs, Meta internal | US$14B2 | Overlap on spam/bad-actor filtering. But avet gates at ENTRY, not post-entry. |
| Community tools | Khoros, Sprinklr, Higher Logic, community bots | ~US$800M–1.2B3 | Closest adjacency. Avet plugs in as a gating layer for existing platforms. |
| Token gating (Web3) | Guild.xyz, Collab.land, Unlock Protocol | ~US$20–50M4 | Rule-based gating (hold X tokens). Avet does qualitative gating. Complementary, not competitive. |
| Platform | Users | Bot API | Gated Communities | Avet Viability |
|---|---|---|---|---|
| Discord | 200M+ MAU5 | Rich, public, free | Millions of servers with role-gating | HIGH |
| Telegram | 900M+ MAU6 | Open, free, well-documented | Millions of private groups | HIGH |
| Slack | 32M+ DAU7 | Rich API, app directory | Enterprise workspaces | MEDIUM (enterprise sales cycle) |
| 2.78B MAU8 | Restricted, paid, no directory | Communities (2023), 1K cap | LOW (closed ecosystem) | |
| Circle / Skool | ~30K+ communities9 | API varies | Paid communities native | MEDIUM (may build natively) |
Every existing community gating tool is rule-based: hold X tokens, solve a CAPTCHA, answer a form, pay a fee. None of them do what avet proposes — a conversational AI agent that assesses whether a person has certain attributes.
| Approach | Examples | What It Checks | Limitation |
|---|---|---|---|
| CAPTCHA / verification | Captcha.bot, Double Counter, Wick | “Are you human?” | Only filters bots, not bad-faith humans |
| Token gating | Guild.xyz, Collab.land4 | “Do you hold asset X?” | Binary. Wallet balance ≠ community fit. Web3-only. |
| Form / questionnaire | Google Forms, Tally, Discord app forms | “Answer these 5 questions” | Static. Easy to game. No follow-up probing. |
| Manual review | Human admins | “Does this person seem legit?” | Doesn’t scale. Subjective. Slow. The pain avet solves. |
| Agentic vetting (avet) | No direct competitor | “Conversation to verify attributes, follow up on inconsistencies, assess fit” | Unproven. Agent reliability on nuanced judgment is the risk. |
| Company | Model | Scale | What They Got Right | Why Avet Is Different |
|---|---|---|---|---|
| Guild.xyz | Token/credential gating for Discord + Telegram | 60K+ communities4 | Multi-platform. Composable rules. Web3-native. | Rule-based only. “Hold 10 ETH” is a rule, not a judgment. Can’t assess “is this person an experienced developer?” |
| Collab.land | Token-gating bot for Discord/Telegram | 50K+ communities, raised US$25M+10 | First mover in Web3 gating. Simple UX. | Same limitation: binary rule check. No qualitative assessment. |
| Commsor | Community analytics + member CRM | Raised US$16M11 | Tracks member engagement post-entry. B2B SaaS. | Post-entry analytics, not pre-entry vetting. Complementary, not competitive. |
| Persona | Identity verification platform (KYC) | Raised US$200M+ at US$1.5B valuation12 | Programmable identity flows. Used by major fintechs. | Document-based KYC, not conversational vetting. Different buyer (compliance teams, not community admins). |
| Company | Model | Revenue / Scale | Playbook | Transferability to Avet |
|---|---|---|---|---|
| MEE6 | Freemium Discord bot | 19M+ servers, ~US$10M+ ARR13 | First-mover on Discord bot marketplace. Moderation → premium features. | HIGH — if avet starts on Discord, same playbook applies. Open API, bot directory, viral server-to-server spread. |
| Combot | Freemium Telegram analytics + moderation | Millions of groups, ~US$1–3M ARR14 | Telegram Bot API. Group analytics wedge → upsell moderation. | HIGH — same open-API playbook. Avet could wedge via free vetting → paid custom criteria. |
| Skool | Community platform, US$99/mo | ~US$20M+ ARR, 30K+ communities9 | Sam Ovens’ audience. Replaced FB Groups for paid communities. | LOW — platform replacement, not a bolt-on tool. But proves willingness to pay US$99/mo for community tooling. |
| Guild.xyz | Free token-gating, API fees | 60K+ communities4 | Composable rules engine. Multi-platform. Web3 flywheel. | COMPLEMENTARY — Guild gates on tokens; avet gates on qualitative attributes. Could integrate: “hold 1 ETH AND pass avet interview.” |
| Company | What They Tried | What Happened | Lesson for Avet |
|---|---|---|---|
| Chatfuel / ManyChat (WA) | WA chatbot builders for marketing/support | Pivoted to Messenger/Instagram. WA API restrictions made it uneconomical.15 | Don’t start on WA. Start where APIs are open, expand to WA later. |
| Various DAO gating tools (2021–22) | Token-gating + role verification | Many died with the crypto winter. Collab.land and Guild survived by going multi-chain + multi-platform. | Single-chain or single-platform = fragile. Multi-platform from day one. |
| AI hiring tools (HireVue, Pymetrics) | AI-powered candidate assessment via video/games | US$100M+ raised but facing regulatory pushback. Illinois BIPA, NYC Local Law 144 require audit of AI hiring decisions.16 | Regulatory risk. If avet’s “attribute verification” looks like hiring discrimination, same laws may apply. Community gating ≠ employment, but the line blurs for professional communities. |
| Component | Per-Unit Cost | Assumption | Source |
|---|---|---|---|
| LLM inference (conversation) | US$0.002–0.01 | 5–10 turns, ~3K–8K tokens total. GPT-4o-mini at US$0.15/1M input, US$0.60/1M output17 | OpenAI pricing page |
| LLM inference (assessment) | US$0.001–0.005 | Final judgment prompt, ~1K tokens | OpenAI pricing page |
| Platform API cost (Discord/TG) | US$0 | Discord and Telegram Bot APIs are free56 | Platform docs |
| Platform API cost (WhatsApp) | US$0.005–0.08/msg | Per-conversation pricing. 10-turn vetting = US$0.05–0.8015 | Meta Cloud API docs |
| Cloud hosting | ~US$0.001 | Serverless function per vetting | Vercel/AWS Lambda pricing |
| Total COGS per vetting | US$0.005–0.02 (Discord/TG) US$0.06–0.90 (WhatsApp) | ||
| Metric | MEE6 (Discord)13 | Guild.xyz4 | Combot (TG)14 | Avet (estimate) |
|---|---|---|---|---|
| ARPU (paid) | US$11.95/mo | Free (API fees) | ~US$5–10/mo | US$10–30/mo |
| Free → paid conversion | ~2–4% | N/A | ~1–3% | ~3–8% higher if niche |
| Distribution | Bot marketplace | Web3 ecosystem | TG Bot API | Bot marketplace + “vetted by avet” badge |
| CAC (organic) | ~$0 | ~$0 | ~$0 | ~$0 if Discord/TG marketplace works |
| v1 Claim | Verdict | Corrected Assessment |
|---|---|---|
| “WA admin tooling market ≈ $0” | TRUE for WA | Correct for WhatsApp specifically. But the framing error was treating WA as the entire market. Community gating across Discord/TG/Slack is a real, active category with Guild.xyz (60K communities) and Collab.land (50K communities) as proof. |
| “Closed ecosystem = no distribution” | TRUE for WA FALSE for Discord/TG | v1 was right about WA but made a category error: concluded “no distribution for the product” when it should have said “no distribution on this specific platform.” Discord has Top.gg (400K+ bots listed). Telegram has BotFather and directories. |
| “Casual admin has $0 budget” | TRUE for casual FALSE for pro | Casual WA family-group admins: correct, $0 budget. But paid community operators (Skool creators, Discord community leaders, DAO managers) already spend $99–300/mo on community tooling. These are the real buyers. |
| “US$12K ARR ceiling” | WRONG | The $12K ceiling was a function of the WA-only constraint (100 admins × $10). Multi-platform unlocks 10–50x more addressable communities. Realistic ceiling: US$36K–120K ARR for a solo builder. Still not venture-scale but meaningful side revenue. |
| “Platform risk: Meta will build it” | PARTIALLY TRUE | True for WA (Meta will build native screening). But multi-platform avet can’t be replicated by any single platform owner. The moat becomes cross-platform consistency: “vetted by avet” works on Discord AND Telegram AND WA. No platform builds that. |
| “This is a feature, not a business” | DEBATABLE | Rule-based gating IS a feature (platforms build it natively — Discord AutoMod, TG anti-spam). But qualitative AI vetting is NOT a feature — it requires domain-specific prompting, evaluation tuning, and continuous improvement. The agent’s judgment quality is the product. Harder to commoditise than a rule engine. |
| “Viral loop has 4 weak links” | TRUE for WA NOT TESTED for Discord/TG | On Discord, the loop simplifies: admin searches “vetting bot” on Top.gg → installs avet → tries free → upgrades. Three steps, all frictionless — exactly the MEE6 playbook. v1 was testing the WA loop, which is genuinely broken. The Discord/TG loop is proven by every successful bot in those ecosystems. |
v1 focused on distribution risk. That was the wrong risk for the broadened thesis. The actual make-or-break question for platform-agnostic avet is: can an AI agent reliably assess qualitative human attributes through conversation?
| Attribute Type | Example | Agent Reliability | Evidence |
|---|---|---|---|
| Factual / verifiable | “Do you have 5+ years of Python experience?” | HIGH — can probe specifics, cross-reference | LLMs excel at knowledge assessment (coding interviews, technical trivia)18 |
| Self-reported | “Are you an accredited investor?” | MEDIUM — can ask but can’t verify | Self-attestation is gameable. Agent adds friction but not verification. |
| Behavioural / cultural fit | “Will this person contribute positively?” | LOW — highly subjective, no ground truth | AI hiring tools (HireVue) faced accuracy challenges + regulatory pushback on this exact claim16 |
| Social graph | “Is this person connected to existing members?” | HIGH — if platform API exposes connections | Discord/TG expose mutual servers/groups. Cross-reference is deterministic. |
| Phase | Platform | Action | Timeline | Success Signal |
|---|---|---|---|---|
| Phase 0 | WhatsApp (own groups) | Dogfood. Deploy in 3–5 own groups. Test if vetting adds value. Reuse Donna infra. | 1 weekend | Do vetted groups feel different? Does anyone ask about it unprompted? |
| Phase 1 | Discord | Build Discord bot. List on Top.gg. Free tier: 1 server, 20 vettings/mo. Target: crypto/DAO communities. | 2–3 weekends | 50+ servers install. 5+ unprompted upgrade requests. |
| Phase 2 | Telegram | Port to TG Bot API. Same free/paid tiers. | 1 weekend (after Discord validates) | Cross-platform communities using avet on both Discord + TG. |
| Phase 3 | API / embeddable | Expose avet as API. Let Skool, Circle, custom platforms embed. | Only if Phase 1–2 show pull | Platform integration requests from community tools. |
| # | Project | Priority | Time |
|---|---|---|---|
| 1 | Sourcy/Karl contract | HIGH — paying | ~10hr/wk |
| 2 | Blackring (agentic hardware) | HIGH | Variable |
| 3 | Donna pilot (AI PA) | HIGH — core product | Variable |
| 4 | Wenhao Blue-Collar AI | MEDIUM | Variable |
| 5 | Talent Coop | EXPLORING | Light |
| 6 | avet (this) | EXPLORING | Weekends only |
SG IMDA pre-approved list has no “community vetting” category. PSG/EDG grants are possible under “AI-powered community management” but would require productisation (Phase 3). Not relevant for Phase 0–1. No decision blocked on this.19
Upgraded from “weekend hack” to “weekend-phased experiment worth pursuing on Discord.”
The v1 verdict was too pessimistic because it locked the product to WhatsApp’s closed ecosystem. The reframe — platform-agnostic agentic vetting for community gating — resolves the three biggest objections: distribution (Discord/TG have open bot ecosystems), buyers (paid community operators spend $99–300/mo on tooling), and platform risk (multi-platform = no single kill switch).
But the reframe also surfaces a new core risk that v1 didn’t address: agent reliability on qualitative human assessment. This is the actual make-or-break, and it can only be tested by building and dogfooding.
What to do:
1. Phase 0 (1 weekend): Build a bare-bones vetting agent. Deploy in 3–5 of your own WA/Discord groups. Define 3–5 concrete vetting criteria per group. Run for 2 weeks.
2. Measure ONE thing: Does the agent’s accept/reject decision match what you would have decided? If >80% agreement → proceed. If <60% → stop.
3. Phase 1 (if agent works): Build Discord bot. List on Top.gg. Free tier (1 server, 20 vettings/mo). Target: crypto/DAO communities that already use Guild.xyz. The “qualitative layer on top of token gating” positioning is the strongest wedge.
4. Phase 2 (if Discord shows pull): Port to Telegram. Same playbook.
What NOT to do: Skip directly to multi-platform SaaS. Don’t build billing, landing pages, or onboarding before you’ve proven the agent can actually vet. The agent’s judgment quality is the entire product — validate that first.
What changes the verdict to “go all in”:
1. Agent agreement with admin judgment exceeds 85% on defined criteria
2. 5+ unprompted requests from other admins after seeing the bot in action
3. Guild.xyz or Collab.land expresses interest in integration (turns avet into infrastructure)
Honest bottom line: The idea is significantly better than v1 gave it credit for. The market exists (60K+ gated communities on Guild.xyz alone). The distribution exists (Discord/TG bot ecosystems). The economics work (95%+ gross margin on Discord). The gap is real (nobody does qualitative vetting). But the product risk is high (agent reliability) and Eric’s time is the binding constraint. Invest weekends. Don’t invest sprints.