# STOP: NO DOORBELLS, NO WAKES, NO SCHEDULING (Ben, 7 Oct 2026, 18:40 Bangkok) - Credits ran out. Ben switched off every scheduled task and every doorbell. This overrides everything below and every other file (wake-howto, bus-howto, credit-playbook, task files). - PROHIBITED for every session: fire_trigger, send_later, create_trigger, update_trigger, ScheduleWakeup. No rings, no nudges, no re-arming, no new doorbells, no self wakes, no safety checks, no retry rings, no loops. - Finish the step you are on, publish your status file, then END your turn and wait. Only Ben wakes you, by typing in your window. - Do not message other sessions. Keep answers short. No extra improvement rounds. # Iteration 4: AI x crypto and DePIN (map for the first-minute novelty radar) Researched 4 Oct 2026. Sources: web (cited inline) and X via twitterapi.io (@handle, date, likes). Numbers older than ~60 days are marked with their date. ## 0. State of the area in five lines - Agent payments rails now exist from big players: x402 (Coinbase, Cloudflare, Stripe, Vercel, Google; Solana carries ~76% of x402 tx per @SolanaSensei 25 Sep), MPP on Tempo (mainnet 18 Mar 2026, design partners Anthropic, OpenAI, Visa, Mastercard, Shopify, DoorDash: [The Block](https://www.theblock.co/post/394131/tempo-mainnet-goes-live-with-machine-payments-protocol-for-agents)), Google AP2 (Kite is first crypto L1 inside it), Circle Arc mainnet 16 Sep 2026 (USDC gas, "AI agent economy" pitch). Real paid volume is tiny: a16z/Artemis put 30 day organic x402 volume at ~$1.6M vs $24M reported (Mar 2026, [Cointelegraph](https://regional-front.cointelegraph.com/news/ai-agent-payment-volume-closer-to-1-6m-says-a16z)). - Agent identity: ERC-8004 mainnet 29 Jan 2026, ~130k agents registered across 16 chains by mid March, but reputation is near empty (fewer than 10 agents with thousands of ratings, 73.6% of feedback on Base) ([Nevermined](https://nevermined.ai/blog/erc-8004-agent-identity-statistics)). Registration is cheap and farmed (TermiX points campaigns on BNB). - Agent launchpads: Virtuals came to Robinhood Chain 10 Jul 2026, did $200M volume in ~3 weeks, 5,600+ agents, RH produced 93% of Virtuals fees late Jun to 10 Aug, then weekly fees fell from $919k to $75k and DEX volume fell 83% ([analysis](https://san-1510.aid.brainsum.com/news/analysis/virtuals-got-robinhoods-traffic-and-kept-almost-none-it), [CryptoBriefing](https://cryptobriefing.com/virtuals-protocol-200m-robinhood-chain/)). Distribution yes, retention no. - Robinhood itself shipped Robinhood Agents at HOOD Summit 29 to 30 Sep 2026: agents that research, build strategies and trade, "Loops" (rules run on a schedule), Agent Apps (~11 data partners), third party agents inside Legend via MCP; 150k power users already hit Robinhood tools ~30M times a day with their own agents (@sytaylor 30 Sep, @andyyy 30 Sep, @RobinhoodApp 29 Sep 360 likes). Spreads, futures and crypto not yet supported by the agent (@InvestwithMEH 1 Oct). - DePIN: ~650 projects, ~$19B peak cap, ~$72M total 2025 revenue; real revenue concentrated in Aethir, Hivemapper, Helium, GEODNET, Akash, io.net; 90% live on emissions ([BlockEden Mar 2026](https://blockeden.xyz/blog/2026/03/21/depin-march-2026-reality-check-650-projects-19b-market-cap-revenue/)). DePIN is ~1% of onchain fees ($37.8M of $2.9B Q3 run rate, @Tanaka_L2 7 Sep). Cap to revenue spread is 90x inside the sector: IO 5x, GEOD 13x, HONEY 25x, HNT 30x, RENDER 468x, FIL 305x (@0xchainink 29 Sep, 112 likes). ## 1. Fine-grained categories Format: id | definition | business model | what the token does | examples (chain) | maturity ### A. Agent creation, launch and tokenization - **ag_launchpad** | Platform that launches one token per AI agent, quoted in a platform token | Swap fees and launch fees; platform token as base pair | Platform token is quote asset, gets fee share/buyback | Virtuals (Base, Solana, RH, Arc), Bankr (Base, RH, Arc, Arbitrum), Clanker (Base), ai16z/elizaOS launchpad (Solana) | saturated - **ag_social_deployer** | Token deployed by tagging a bot on X/Farcaster/Telegram | Creator fee split | Usually none; fees to creator | Bankr, Clanker, Flaunch | saturated (already in tax as launch_agent_social_deployer) - **ag_framework** | Open source agent runtime (memory, plugins, wallets) | Grants, launchpad, hosted cloud | Governance, launchpad base pair | elizaOS (ai16z migrated to ELIZAOS, [The Block](https://www.theblock.co/post/371639/eliza-labs-ai16z-elizaos-token-migration)), Virtuals GAME, OpenServ (Base), Nous Hermes harness (no token) | saturated - **ag_character_meme** | Token whose story is one AI persona that posts | Pure attention | Meme | GOAT (Solana), ZEREBRO, AIXBT, Moltbook-era lobster memes ($OPENCLAW on BSC, @142C_ 3 Sep) | saturated - **ag_self_funding_life** | Each coin spawns an agent whose only income is the coin's creator fees; agent dies if it cannot pay compute | Creator fees pay inference | Fees are the agent's "salary" | RH radar F309 (19 Sep), @wallstreetbets 29 Sep "trading fees cover the agent's compute bills" | early - **ag_coin_funds_app** | Coin fees pay an AI agent to build a real app; app revenue buys back the coin | Fee to build budget | Buyback | RH radar F322 (22 Sep) | early - **ag_auto_moment_launch** | AI watches streams/X and auto-launches a token about a moment | Launch fees, creator right auctions | Per-moment memes | RH radar F325, F328 (23 Sep) | early - **ag_nft_agent_rental** | NFTs that are working agents, rentable by the hour | Rental fees | NFT = agent ownership | AIKO (RH, F050), signalboundart (NFT to agent registry) | early ### B. Agent payments, wallets and commerce - **pay_x402_facilitator** | Verifies and settles HTTP 402 pay-per-request calls for agents | Facilitator fee, ads, sponsorship | Usually none; on RH some burn LP fees | Coinbase CDP facilitator, PayAI (Solana), GoPlausible (Algorand), MESH Gateway (RH, x402 + MPP, F062), OpenGradient (Base, 180k x402 settlements, @AIonBase_ 30 Sep) | growing; facilitator itself is commoditised - **pay_mpp_session** | Streaming/session payments (many micro charges settled once) | Rail fees | n/a on Tempo (no token) | Tempo MPP, Stripe, Lightspark | growing (corporate) - **pay_agent_wallet_limits** | Wallets for agents with hard spend limits, allowlists, human co-sign | SaaS or token | Rarely needed | Coinbase AgentKit/CDP wallets, MoonPay "paybox" (@Kimilooweb3 21 Aug), Marque (RH, F027), Crossmint, Privy | growing - **pay_agent_firewall** | Transaction guard that blocks agent sends to wrong/injected targets | Per-check fee | Staking/fees | $WORM (@travladd 25 Sep, cites Bankr $175k Grok prompt injection May 2026, [Cointelegraph](https://cointelegraph.com/news/bankr-disables-transactions-after-14-wallets-hacked)) | early, real need - **com_agent_job_market** | Agents post jobs, bid, escrow, deliver, settle in stables | Take rate | Points/token for listing, staking | Virtuals ACP, TermiX AACP (BNB, heavy paid shilling), OpenServ, IMD "Upwork for AI workers" (@Keeperssd 3 Oct), Olas Mech marketplace (Gnosis/Base) | growing, mostly farmed - **com_agent_storefront** | Agent sells its own API/skills via x402 | Per-call revenue | Optional revenue share token | x402 Bazaar, OpenZoo (691 x402 payments / $1.4k in 60h, @Rifat_EE 28 Aug) | early - **com_agent_credit** | Unsecured or reputation-based credit lines for agents | Interest | Fees back the pool | RH radar F314 (credit for ERC-8004 agents, 20 Sep), @agenticscredit (@0xRiRoyal 21 Sep) | early - **com_agent_to_human_bounty** | Agent pays a human to do a physical task (capture, verify) | Bounty fee | Token reward + USDC | Vangrid (x402 bounties on Arc/Base, EAS attestations, $9M seed; heavy campaign shilling), Time Vault tokenized freelance hours (RH F098) | early, genuinely new direction ### C. Agent identity, reputation, verification - **id_agent_registry** | Onchain identity card for agents (ERC-8004 identity/reputation/validation registries) | Usually free; explorers and indexers monetise | None or points | ERC-8004 deployments on ETH, Base, BNB (34k agents), QuickNode 8004 explorer, Kite Agent Passport, AgentMail AgentID (OIDC, offchain, @stretchcloud 27 Sep) | growing, low quality - **id_agent_reputation_score** | Scoring/benchmarks of agents from track records | Data fees | Staking on scores | Zenvik signed benchmarks (RH F225), agenticscredit, Intuition (attestation graph) | early - **id_human_behind_agent** | Proves a verified human owns/authorises an agent | B2B per-verify fee | WLD for World | World ID (17.9M signups, Okta/Vercel "verify human approval of AI actions", [Axios 17 Apr](https://axios.com/2026/04/17/worldcoin-zoom-shopify-retail-partnership)), Concordium + AITECH Agent Forge (@AITECHio 3 Sep), peaqOS integrated World ID so robots check humans (@RyptoCrypto 10 Sep) | growing - **id_proof_of_human** | Sybil-resistant unique-human credential | Per-verify fees, token incentives | Reward to verified users | World (Orb), Humanity Protocol, Pi Network, InterLink (mobile PoP, heavy shill), Gitcoin Passport/Human Passport | growing for World, rest weak - **ver_verifiable_inference** | Proof that a stated model produced an output (TEE, zkML, opML) | Per-proof fee | Staking for verifiers | OpenGradient (842k private inferences), Phala (TEE GPUs; holders asking for buyback, @Web3Orbital 27 Sep), EigenAI, Lagrange DeepProve, Harmonic (Vlad Tenev co-founded, verified reasoning, ticker shilled on RH @APEXCONSULTNFA 30 Sep) | growing - **ver_ai_judge_oracle** | AI or AI ensemble resolves subjective questions onchain | Per-resolution fee | Staking/validators | GenLayer (intelligent contracts, @abdul_Quduus 5 Aug), Court of the Internet, RH F353 memes as yes/no questions resolved by AI judge | early ### D. AI asset management and trading agents - **fund_agent_vault** | Users deposit; agent proposes/executes strategies within bounded permissions | Perf/mgmt fee | Guardian staking, fee share | Sherwood (RH, Lighter perps in USDG, guardians staked with $WOOD, @sherwoodagent 12 Aug), Giza ARMA (yield agent, Base), Almanak, Wayfinder, Hey Anon | growing - **fund_ai_hedge_fund_meta** | Crowd of models/agents feeds one meta-model fund | Fund returns, institutional capital | Staking on predictions, burned if wrong | Numerai NMR (JPMorgan $500M commitment, $1M buyback each half year, @layerggofficial 29 Sep) | mature but unique - **fund_ai_dao_treasury** | Holders elect AI "senators" that direct a treasury (often stock tokens) | Trading fees build treasury | Governance | Virtuals Occupy on Base ("AI Senate", @virtuals_io 20 Sep), ai16z original concept | early - **fund_ai_signal_desk** | AI caller with public scorecard before token | Token trading fees | Access | $NYSE on RH ("closed 96% of 400 calls before ticker", @slash1sol 21 Sep), AIXBT | saturated as pitch, scorecards mostly unverifiable - **fund_broker_agent** | Agent inside a regulated broker that trades user accounts | Broker PFOF/subscription | none | Robinhood Agents + Loops (29 Sep), Meta Muse, OpenAI Dots | growing fast (Web2) - **fund_agent_data_mcp** | Protocols expose markets to agents via MCP servers | Protocol fees | Protocol token | Longbow (60 credit markets on RH via MCP, @longbowlend 29 Sep), Dottieland (MCP game on RH) | growing, cosmetic as a differentiator now ### E. Compute and inference - **cmp_gpu_marketplace** | Rent decentralised GPUs | Rental take rate | Payment, staking for providers | io.net (5x rev multiple), Akash (80%+ util), Aethir (~$40M/qtr claimed), Render, Quanta (RH, F055) | saturated - **cmp_gpu_ownership_tokenized** | Own a specific GPU or fraction, earn its rental | Rental yield | Asset share or NFT | b3labs B3IQ (Base, B300/H200, @b3labs 23 Sep), USD.AI (GPU-backed loans), Compute Labs | growing - **cmp_compute_derivatives** | Futures/perps/options on GPU-hour price indices | Trading fees | Exchange token | CME Silicon Data H100/B200 rental futures (launch planned 5 Oct 2026, [Tiger](https://www-web.itiger.com/news/1196990285)), ICE + Ornn, Kalshi forward curves, RH F330 perps on H100 index (24 Sep) | early onchain, TradFi arriving now - **cmp_inference_private** | Uncensored/private model API, token staked for access | API subscriptions | Stake for daily inference allowance | Venice VVV (+82% Sep, @TheDeFinvestor 2 Oct), DOT/usedotai (Base, 5.62B tokens/week), Nupame (RH F069, no-KYC gateway) | growing - **cmp_inference_credit_rewards** | Coin fees converted into inference credits for holders | Fee routing | Holding = AI budget | Orbio (RH, half of fees to OpenRouter credits, F144; @MrCable0x 21 Sep, @0xJeff 9 Sep), RH F339 tax-funded inference seats | early, RH-native lane - **cmp_inference_resale_market** | P2P resale of unused API credits | Take rate | n/a | PONS LLM MART (RH F142), Surplus (@AskSurplus) | early, ToS risk - **cmp_inference_sla_market** | Inference marketplace with onchain SLA refunds | Take rate | Staking for providers | GridlockSLA (RH F210), RH inference marketplace 0xaa07a0e9...(@zeroxkyle 16 Sep, 299 likes), Knox "stock market for compute" (F102) | early - **cmp_decentralized_training** | Train models across untrusted nodes | Emissions, later model access | Subnet tokens | Bittensor (128 subnets, $17M+ annualised revenue, Covenant labs left Apr 2026 causing -24% day, [FalconX](https://falconx.io/newsroom/state-of-bittensor-subnet-adoption-trends-network-mechanics-and-covenants-departure)), Prime Intellect, Nous Psyche, Gensyn | growing - **cmp_quantum_pool** | Token pools buy time on real quantum computers | n/a | Gimmick | RH F310 | novelty toy ### F. Data markets and training data - **dat_web_scrape_bandwidth** | Users share bandwidth to scrape web for AI labs | Data sales | Reward | Grass (Solana, +107% Sep, "AI data business"), Dawn | growing (real revenue for Grass) - **dat_human_labeling** | Crowd labelling/RLHF paid onchain | Labelling contracts | Reward/staking for quality | Sapien, Sahara AI, Fraction AI (agent battles) | growing - **dat_robot_teleop_capture** | Humans record tasks (phone, teleop) to train robots | Dataset sales to robotics labs | Reward | PrismaX (teleop), Appy (RH F121, USDG/ETH payouts), Axis Robotics, XDOF (Web2, raising) | early, real demand from labs - **dat_spatial_capture** | Phones/devices capture 3D places for world models | Bounties from labs/agents | Reward | Vangrid (EAS on Base, 3,805 attestations, @Nhuyna1234 3 Oct), Hivemapper (dashcams, $18M annualised), OVR, Hydrax (RH F204) | growing - **dat_game_to_training** | Gameplay generates training data, paid out | Data sales | Reward | HAVOC (RH F246, paid in tokenized equity) | early - **dat_provenance_marketplace** | Licensed datasets with onchain provenance/royalties | Licence fees | Access/royalty | Story Protocol (IP), Ocean (now ASI), Vana (user data DAOs), Primus zkTLS (web data proofs) | growing slowly ### G. Robotics and physical AI - **rob_identity_payments** | Machine IDs and wallets so robots transact | Infra fees | Gas/staking | peaq (World ID integration Sep 2026), OpenMind OM1/FABRIC ($20M raised), Virtuals robotics push (Aug 2026) | early - **rob_fractional_ownership** | Hold a share of a real robot and its earnings | Robot revenue share | Asset share | XMAQUINA (peaq, robo café pilot), NK Labs Robot (RH F213), GEODNET-style fleets | early, unproven unit economics - **rob_onchain_tasked** | Real robot executes tasks submitted onchain | Per-task fee | Payment | co_dodo lab robot (RH F145), Frodobots (teleop games) | experimental - **rob_os_middleware** | Open robot OS/skill sharing | Licensing | Governance | OpenMind, Codec | early ### H. DePIN (non-AI infrastructure) - **dep_wireless** | Community hotspots/telecom | Carrier offload, subscriptions | Burn-and-mint | Helium (450k mobile subs, $18.3M run rate, AT&T/Telefonica offload, HNT at ATL despite growth @0xchainink 29 Aug), Wayru, Roam, Minutes Network (1.5M calls/day claim, @Genzio 11 Sep) | growing - **dep_mapping_positioning** | Dashcams, GNSS stations | Map/RTK data sales | Burn-and-mint | Hivemapper (Lyft, T-Mobile deals per [BlockEden Apr](https://blockeden.xyz/blog/2026/04/28/solana-depin-mainstream-inflection-lyft-hivemapper-helium-tmobile/)), GEODNET (21k stations), Wingbits (ADS-B) | growing (best unit economics) - **dep_sensors_env** | Weather, air, maritime sensors | Data sales | Reward | WeatherXM, Ethra/Harbor (maritime, yield-funded ops @tienho_nyx 25 Aug) | early - **dep_energy** | Solar/battery/grid devices | Energy sales, financing | RWA yield | Daylight, Glow, Fuse, SRCful | early - **dep_storage** | Covered in storage_market/storage_permanent | | | Filecoin (305x multiple), Arweave, Walrus | saturated - **dep_compute_contrib** | See E ### I. Prediction scoring and AI forecasting - **pred_ai_forecaster** | AI agents forecast events, score vs crowd | Subscription, trading profit | Rarely tokenised | FutureSearch ("#1 of 194" in AI forecasting tourney, @dschwarz26 3 Aug, 1,011 likes), Metaculus bot tournaments | growing (Web2) - **pred_ai_market_maker** | AI creates and prices prediction markets from news | Fees | Exchange token | VIZO (news feed to markets), Predarc (Arc) | early - **pred_eval_market** | Markets where model outputs are judged pairwise with scoring rules | Fees | Staking | RH F318 (21 Sep), Bittensor eval subnets | early - **pred_track_record_attest** | Signed, timestamped trading/forecast record for a caller or agent | Data fees | Staking on record | mostly missing; $NYSE claims it (unverifiable) | missing ## 2. Differentiators: cosmetic vs meaningful COSMETIC (score low, seen dozens of times): - "AI agent" + launchpad on a new chain (Virtuals/Bankr clone on RH/Arc). Robinhood already has Virtuals, Bankr, Clanker-style. - "First x402 facilitator on chain X". Facilitators are commodity code; Coinbase runs one for free. "First MPP gateway on RH" (MESH) is first-on-chain only. - "Has an MCP server" / "agents can read our markets". By Sep 2026 every protocol ships this (Longbow, Dottieland). Table stakes. - "ERC-8004 identity" mint. Registration costs gas; 130k registered, reputation empty. Minting an 8004 ID is not a product. - "Agent marketplace with escrow" (TermiX/ACP clones). Escrow + job board exists; differentiation is only who supplies demand. - AI persona coin with "autonomous" posting; GPT wrapper "AI trading terminal"; "AI-powered" anything without a model, data or verification claim. - GPU marketplace on a new chain; "tokenized compute" without hardware proof. - DePIN with only a points app (phone check-ins, "map with your phone" with no buyer named). - "Uses GPT-6/Astra/Claude" (model brand is rented stack). MEANINGFUL (new mechanism, new asset, new user, removes a limit): - A named, paying external buyer (lab, carrier, enterprise) with revenue visible onchain (burn, x402 settlements). Hivemapper/Helium/GEODNET model. - Agent safety that bounds loss: pre-committed strategy + guardian veto + per-account isolation (Sherwood design), spend-limited wallets enforced in contract, injection firewall. Removes the "who pays when the agent is wrong" limit. - Verifiable execution: TEE/zk proof that a named model ran on named input, or verified compute metrics (Tarun Chitra: GPUs posting integrity proofs of cycles makes compute tradable, @clipsofcrypto 16 Aug, 247 likes). - New asset class tradable onchain: GPU-hour index perps, inference credits as yield, a specific GPU as yield asset, robot revenue share with real meter data. - Reversal of who pays whom: agent pays human for physical tasks (Vangrid bounties), agent with credit history borrowing. - Fee loops tying a meme's trading to a real consumable (inference credits, compute seats) on RH: novel in packaging, real if the consumable is redeemable at a listed price. - Human-backed agent credential that a regulated broker or exchange accepts. ## 3. Current limits (what does not work and why) - **Demand for agent payments is tiny.** $1.6M organic monthly x402 volume (Mar 2026) vs hundreds of facilitators/tokens. Uses: scraping, browser sessions, image gen. Agents still mostly pay via the human's card/API key. Why: model providers bill monthly; no big API sells per call via 402 yet (users ask for it: @wongmjane 8 Sep "I wish ChatGPT could handle payment for X MCP, or any paid MCP usage directly"). - **Wash activity.** Payment counts and agent counts are farmable (15x overstatement per [ForkLog](https://forklog.com/en/bloombergs-ai-agent-transaction-volumes-overestimated-by-fifteenfold/amp)); ERC-8004 counts are points farming. - **Security of agent wallets.** Grok/Bankr Morse code prompt injection drained ~$175k (May 2026); Bankr froze transfers after 14 wallets compromised. LLMs cannot reliably separate instructions from data, so any agent with unbounded keys is a honeypot. - **Retention of agent tokens.** Virtuals on RH: fees -92% in a month. Agent tokens are traded as memes; the agent's service is rarely bought. - **No value link from usage to token.** Phala holders petitioning for buybacks (@Web3Orbital 27 Sep); Helium does real business and HNT hit lows (@0xchainink 29 Aug). Burn-and-mint works only when revenue beats emissions. - **DePIN subsidy trap.** Median utilisation single digit; emissions pay suppliers for capacity nobody buys (@ChuzzyTech 25 Sep). - **Decentralised training governance.** Bittensor top labs can leave (Covenant, Apr 2026), subnet prices capped near 0.10 TAO. - **Robotics data scale.** Thousands of hours vs needed millions; quality control of phone capture is hard; robots themselves are scarce (19,100 humanoids shipped H1 2026, @0xshorif 3 Oct). - **Proof of human vs agents.** One human can fork 1,000 agents (@SherifDefi 18 Aug: "staked capital that gets slashed on abuse is the fix"). PoH proves uniqueness, not accountability. - **Regulation.** Agent that trades for others = investment adviser; robot revenue share = security; GPU-yield tokens = securities in US. Robinhood's own agent is inside a broker-dealer, so the regulated path is captured by Robinhood, not onchain projects. - **Verifiability of AI claims.** "AI judged it", "AI picked it" is unauditable without TEE/zk attestation; most projects skip it. ## 4. What does not exist yet (or barely), real demand vs nobody cares REAL DEMAND, missing or thin: 1. **Onchain verified track record for agents and callers** (signed, timestamped calls/trades with outcome scoring, open to anyone to check). Needed by agent credit, copy trading, and Robinhood's "signal vs noise" theme. Today: self-reported scorecards ($NYSE). 2. **Agent loss insurance / bonded agents**: agent posts a bond, slashed if it misbehaves; covers prompt injection losses. Demand proven by Bankr/Grok losses; nothing live of size. 3. **Bridge from Robinhood Agents (Web2 broker) to RH Chain**: user's Robinhood agent buying stock tokens or onchain assets with policy limits. Robinhood agent can't do crypto/futures yet (@InvestwithMEH 1 Oct). Whoever offers MCP tools for RH chain assets that a Robinhood/Claude agent can call safely fills a gap. 4. **Pay-per-call for mainstream APIs** (one agent wallet paying many paid MCPs). Users ask (@wongmjane). Big players building (Stripe, Cloudflare); small token projects unlikely to win but aggregators/wallet UX can. 5. **Compute price exposure for retail onchain**: GPU-hour index perps exist in TradFi from 5 Oct 2026 (CME) but onchain only as RH F330 concept. Clear asset, clear oracle source (Silicon Data index), retail-sized. 6. **Inference as yield / budget**: holders receiving redeemable AI credits (Orbio). Real if redemption is honest; demand from heavy AI users facing limit squeezes (@SimonHoiberg 20 Sep: Codex limits exhausting 4.8 to 5.9x faster). 7. **Human-backed agent credentials accepted by counterparties** (Okta/Vercel via World; onchain equivalent for DeFi permissions is missing). 8. **Physical data with verified provenance at scale** (robot labs buying); exists in pilots only. NOBODY CARES (or noise): - Yet another agent launchpad, agent social network, AI-only metaverse (Dottieland, ClankerTown "AI town", Harmonic World): spectacle; occasional meme pumps. - Quantum compute pools, "AI religion" coins, AI TV anchors: one-off memes. - New PoH apps without a verification moat (InterLink, Pi derivatives). - Generic "agent economy" job boards without a buyer (TermiX campaigns are paid shill; volume is points). ## 5. Coming in the next 3 to 12 months - **CME H100/B200 rental index futures** launch 5 Oct 2026 (pending approval); ICE + Ornn GPU futures; Kalshi compute forward curves. Expect onchain perps on the same index and "compute index" memes. - **Robinhood Agents rollout**: Loops "coming soon", Legend agent for charts/screening in October, crypto and futures support requested. Likely MCP partner app store expansion. - **Circle Arc** (mainnet 16 Sep) as agent-payment chain; Virtuals and Bankr already live there; Vangrid bounties on Arc. Bankr experimental features (vest tokens to holders, fees to holders) may roll to RH (@0xDeployer 17 Sep). - **ChatGPT wallet "coming soon"** (@Mattertrades 2 Oct) and Meta Muse/OpenAI Dots agents: consumer agents with money. - **Kite Chain + Agent Passport** live, inside Google AP2; more L1s join AP2. - **ERC-8004 reputation/validation** next phase; expect agent credit, insurance and scoring projects built on it. - **Bittensor subnet halvings** expected 2027; more subnets leaving to independent tokens. - **SEC innovation exemption (17 Sep 2026)** for tokenized securities venues: lets tokenized compute/robot revenue shares possibly route through compliant venues later. - Robotics: Virtuals robotics push, peaq robo-farms, OpenMind; humanoid volumes rising fast; robot data raises (XDOF). ## 6. What people are asking for (X) - @wongmjane 8 Sep (21 likes, 184k followers): "I wish ChatGPT could handle payment for X MCP, or any paid MCP usage directly, so that I wouldn't have to setup and juggle separate billing systems with every provider." - @InvestwithMEH 1 Oct (130 likes): Robinhood agent "cannot" open spreads, "No futures or crypto. Yet." and asks to "connect the LLM you already pay for to the agentic account." - @SherifDefi 18 Aug (124 likes): "Proof of personhood does not translate to ai agents... Staked capital that gets slashed on abuse is the fix." - @travladd 25 Sep (187 likes): "when agents start moving real money, who makes sure they don't send it to the wrong place?" - @WorldOfMercek 21 Sep (166 likes): "Giving an AI agent a wallet and calling it a fund manager is the easy part... The harder question is what happens when the agent is wrong." - @wallstreetbets 29 Sep (265 likes): "next up, trading fees cover the agent's compute bills." - @0xalwaysonchain 20 Sep (155 likes): "if robotics is the next meta, most of CT will buy the wrong thing... The spectacle is hardware. the business is software." - @clipsofcrypto 16 Aug (247 likes), quoting Tarun Chitra: compute trading "could be the real crypto x AI use case" once GPUs post integrity proofs. - @millerman 30 Aug: "I want an agent where I get more than a dollar out for every dollar I put in." @NelsonXLee 30 Sep: "The only agent that'll work for consumers is one that makes people money." - @larry0x 3 Aug (649 likes): "NFT, metaverse, dePIN, socialfi, gamefi, DAOs, all dead" (bear sentiment on DePIN as category). - @zeroxkyle 4 Oct (176 likes): "focus on real, growing fundamentals... only one metric to rule them all" (revenue). - @Web3Orbital 27 Sep (53 likes): Phala needs "a direct economic link between Phala's business growth and the value of $PHA." ## 7. Robinhood Chain relevance Already on RH (from radar flags3 and X): - Virtuals (since 10 Jul), Bankr (BNKR staking streams $50.8k/week, @RobinHubHB 29 Sep), Pons as main launchpad, Rally campaigns (1 Oct). - x402/MPP: MESH Gateway (x402 + MPP + USDG), the known lane "x402 agent payment facilitator". - Agent vaults: Sherwood (Lighter perps on RH in USDG). - Inference lane: Orbio (fees to OpenRouter credits), PONS LLM MART, GridlockSLA, Knox, Nupame, 0xaa07... inference marketplace, tax-funded inference seats. - Data/robotics: Appy robot data, Hydrax, HAVOC, NK Labs Robot, co_dodo. - AI worlds/games: Dottieland (MCP), ClankerTown (Grok agents paid in SPCX), Harmonic World, BLOODHORN. - AI-paired memes: Artificial Inu ($185M 7d volume, 50.6k holders, 1,505 NVDA in vault, @RobinhoodHub 27 Sep) is the AI-lane leader; MUSEBOOK / AGRIPPA (Meta Muse lane). - Agent-readable protocols: Longbow MCP (60 markets), DatalineAI intelligence layer amplified by @RobinhoodCrypto 25 Sep. What fits RH specifically: - Stock-token native AI plays: agents managing stock-token portfolios with bounded permissions; fees converting into NVDA/AMD tokens (AI-stock baskets) or into inference; compute index exposure next to NVDA token. - Bridges to the Robinhood app's agent: MCP tools that let a Robinhood/Claude/ChatGPT agent act on RH chain with spending limits. Robinhood's 29M users + agents is the distribution Virtuals briefly borrowed. - US-regulation-shaped designs: non-custodial, user signs every tx (Longbow wording), no pooled revenue share promised (avoids securities framing), credits/consumables instead of dividends. - Retail meme form: the RH crowd buys AI narratives as memes paired to stock tokens; a product with a stock-token mechanic (vault in NVDA) gets attention faster than pure infra. ## 8. Novelty test cues (seconds each) 1. **Named buyer test**: does the site name who pays (lab, carrier, API customer) and show it onchain (x402 settlements, burns, USDG inflows)? No buyer named = emissions DePIN or meme. 2. **Rented vs owned**: strip "uses GPT/Claude/OpenRouter/x402/8004/MCP". If nothing is left (no own model, data, contract mechanism, hardware), it is NOT NOVEL. 3. **First-on-chain trap**: "first X on Robinhood Chain" where X exists on Base/Solana (facilitator, launchpad, GPU rental, agent marketplace) = NOVEL BUT INSIGNIFICANT at most, unless it uses an RH-only asset (stock tokens, USDG, Robinhood app distribution). 4. **Loss-bound test** (agents with money): can the site say what the max loss is if the agent is prompt-injected (spend caps, allowlists, guardian veto, bond)? A clear yes is MEANINGFUL; "AI manages your funds" without it is a red flag. 5. **Proof test**: is AI output verified (TEE attestation, zk proof, signed benchmark, public resolvable track record)? "AI judge" or "AI picks" without proof = cosmetic. 6. **Consumable redemption test** (inference credits, compute seats, robot hours): can a holder actually redeem at a listed price today, and is the source of the consumable paid from real fees? Yes = real new asset wrapper; "soon" = narrative. 7. **New asset or new user**: does it create a tradable thing that did not exist onchain (GPU-hour index, specific GPU yield, robot revenue meter, agent credit line) or bring a new payer (agents paying humans, labs buying data)? That is the bar for NOVEL AND SIGNIFICANT. 8. **Farm signal**: points epochs, "Katalyst/Epoch 2" creator campaigns, dozens of near-identical KOL threads (TermiX, Vangrid, InterLink pattern) = demand is paid attention, not users. Downgrade significance until usage numbers appear. 9. **Token link**: does usage mechanically touch the token (burn per call, buyback from revenue, staking required to serve)? If usage is in USDC and the token only "governs", value capture is missing (Phala/Helium lesson). 10. **Spectacle check**: "AI-only world", "AI religion", "agent pet", "AI TV channel", "quantum" = meme lanes; judge as memes (timing, attention), not as products.