Day 2 routes more of the board across a single stage than any other block of the conference. The move is to stop moving.
Cayman breakfast at 8:30 upstairs, then an aisle seat on Main Stage from 9:20 and do not leave it until 12:10. Intercept every target as they come off stage, never before they go on — pre-stage they are rehearsing, post-stage they are relieved and looking for a reaction.
Corrected time: 8:30, not 9:00. Half an hour earlier than the board had it, which is the difference between walking in on the fireside and walking in after it.
"Cayman and the Future of Tokenised Finance" — Kristy-Anne Leith, partner at Campbells leading their Cayman digital assets practice, in conversation with Danielle Pienaar, Web3 & Blockchain Director at Verdant Management and Chair of Women in Blockchain at BACI. Lalla Asmaa Alaoui is in the room too.
Why this is now a priority and not a breakfast. Leith is the Cayman-side legal counterpart to Bacina. You spent last night on the entity question with a web3 attorney; this is the same question asked from the jurisdiction that would have to answer it. Your thread continues here without needing anyone's approval.
Read her actual work before you speak. On 19 March 2026 Cayman published three amendments — Mutual Funds, Private Funds, and Virtual Asset (Service Providers) — creating the framework for tokenised funds, in force 24 March. Leith co-authored Campbells' analysis of it. Four months old, and she helped explain it to the market. Referencing it is not flattery, it is homework.
Two provisions matter to you:
① The Mutual Funds amendment mandates operator approval for transfers — every movement of a tokenised interest needs the fund operator to sign off.
② The VASP amendment confirms fund tokens are not a "virtual asset issuance", so they sit under fund law rather than crypto law.
"Your March amendments require operator approval on every transfer of a tokenised interest. That assumes a named transferee the operator can actually approve. What happens when the holder is an autonomous agent transacting at machine speed? Either approval becomes the bottleneck that defeats tokenisation, or it becomes a formality nobody really performs. Is that gap something the framework anticipated, or is it just early?"
Why this lands. It is not a pitch, it is a stress test of her own framework, asked by someone who read it. And it is the same entity question you ran past Bacina last night, now aimed at the jurisdiction that has to answer it in statute rather than opinion.
Sequencing: take the front of the breakfast, slip out at 9:15 for Saliba, come back for the 9:45–10:00 tail — which is the networking half anyway.
Machine-to-machine micro-transaction ledger — the closest public analog to the UCW value-capture layer. Do not pitch your thesis as new. He launched Mosaic on Jul 16 for cryptographic identity; frame metering as the monetary complement to his identity end, not a competitor.
"HyperCycle's per-inference settlement is the rail I've been thinking about for cognitive rent. Where does node economics break at scale?"
Two targets, one panel. Khamitova runs startups & scaleups at Vector — Canada's flagship AI institute, and a home-turf credibility door. Ask her: "METAVENTIONS is a Toronto sovereign-AI company — how does Vector work with startups on the applied-research side?"
⚠ Check the badge before you open. She is still COO of SingularityNET, but the conference lists her as Co-CEO and Founder of Zarqa. Both are true — she co-founded Zarqa, Cogito, Rejuve and Jam Galaxy out of that ecosystem. What matters is that she chose the Zarqa billing for this room, so leading with "your ASI Alliance work" opens on the thing she is not here to talk about.
Zarqa is a neural-symbolic LLM — deep networks plus a symbolic reasoning core, aimed at the reasoning and context limits of ordinary LLMs. The part that matters to you is how they describe the economics: smart ownership and crowdsourced optimization.
"Zarqa's model gets better through crowdsourced optimization, which means many people improve one model. That is the attribution problem before agents even enter it: whose contribution moved the model, and what is that worth? I build the measurement layer for exactly that, one layer out — across agents rather than across contributors. How are you handling the split inside Zarqa?"
That opener says you read her current company, not her Wikipedia entry. If it lands, the DEEP Funding grants she sources for are the concrete follow-on — a real ask, not a coffee.
Two targets, one panel — but do not spend McMullen here. You get him again at 12:25 in Purple Room, so this is the warm hello that makes the second one land. Say your name, say what you build in one sentence, tell him you will be at the Agentic Economy panel. That is all.
Save the real question for 12:25 — it is below.
⚠ The board had Trevor Koverko here. The app says Ali Malik. Sapien is sending someone else, so do not sit through this expecting Koverko and do not walk up to a stranger with the Polymath opener. If Koverko is at the conference at all he is working the floor, not this stage.
Lands between Adams and Stornetta, so you are already in the seat — zero cost to stay. The Sapien thesis is unchanged, so the data-layer question still works on whoever presents: "Sapien's data layer plus a sovereign inference layer is a natural stack. How are you thinking about downstream value-share to labelers?"
Koverko is still on the official speaker list ("Co-Founder, Sapien"), so he is in the building — just not on this slot. Ali Malik does not appear on that list at all, which reads like a late substitution. Keep the Polymath/Polymesh material in your pocket for Koverko on the floor or in VIP Backstage. It is on /battle-station.
Free opener: Polymath ran the "$16 Trillion Migration" panel on Day 1 Main Stage — Dolan, Hirsch, plus two Halford keynotes. "I sat through the Polymath tokenisation panel yesterday…" costs nothing and lands instantly.
The real question. Polymesh requires every on-chain actor to be a verified identity — they already answered "who is the legal person" for humans and institutions:
"So what happens when the actor is an autonomous agent? Does it inherit the deployer's identity, or need its own? Inheriting breaks the moment six agents from four firms co-produce one settlement."
That is your entity question aimed at the one team who may actually have an answer.
The one that does not come around again. Haber & Stornetta, 1991 — "How to Time-Stamp a Digital Document" — is cited in the Bitcoin whitepaper and is the direct ancestor of every chain in this building.
His current SafeBets mandate is cryptographically verifiable prediction records: who forecast what, and when. That is the attribution problem with a different noun in front of it.
"Your timestamping problem and my attribution problem are the same problem. You proved you could establish when something happened without a trusted party. I'm trying to establish whose cognition produced an outcome — same absence of a trusted party, one dimension over."
The warm route: Gina Antoniello and SafeBets were on Day 1's 3:30 block. If you touched that, open with it — a sponsor-side introduction beats a cold stage-exit approach with someone this senior.
Purple Room stacks McMullen twice: Trust Layer 11:50–12:25 and The Agentic Economy 12:25–12:55. The first collides with Stornetta. The second does not.
Take Stornetta, skip Trust Layer, then walk to Purple Room for 12:25. "The Agentic Economy, Verifiable Infrastructure" is your thesis printed on a schedule — and you will already have met McMullen at 10:50, so you arrive as a second touch rather than a stranger.
Know the product before you speak. Billions Network verifies humans and AI agents — proof of uniqueness, KYC/AML, credentials, and machine reputation, using DIDs and verifiable credentials, non-biometric. $30M raised, official identity and reputation partner to Sentient, and used by Deutsche Bank, TikTok and HSBC. Their own framing for it is "Know Your Agent".
His stat is a gift — cite it back to him. Billions' own number is that non-human actors are already more than 51% of online and on-chain interactions. You do not need to argue the premise. He has already published it.
The distinction that makes you useful rather than competitive: reputation is a score about past behaviour. Attribution is a claim about a specific outcome. Billions answers who acted and can they be trusted. You answer what share of this result was theirs. Those compose; they do not overlap.
"You've built Know Your Agent — identity plus reputation, and your own number says non-humans are already past half of all interactions. Here's where I get stuck one layer up. Reputation tells me which agent to trust. It doesn't tell me what an agent contributed to a specific outcome. When six agents from four firms co-produce one settlement, I can verify all six and still have no basis for the split. Does Billions carry contribution, or only standing?"
If he says standing only, that is the opening: you are the layer above his, and he has the distribution. If he says they are building toward contribution, you have found either a partner or the strongest competitor in the space — and you would rather know today.
Backup, if the room is too big for a hallway conversation, ask from the floor: "Everyone here has solved whether value moves. Who is solving whose cognition caused it? AWS's own x402 docs say agents can't split fees dynamically — they hardcode partner splits. What replaces that guess?" Then /attribution-gap is what you send afterwards.
Not on the original board. Straight down your thesis: this is the room arguing about which rail wins, which is the argument that assumes rails are the constraint. Your whole position is that they are not — execution is solved, attribution is not. Good room to ask the question from the floor if the Purple Room ran cold.
The single most on-thesis session on Day 2, and it was not on the board. You have asked the entity question three ways in 24 hours — Bacina legally, Leith jurisdictionally, Polymesh at the protocol layer. This is the same question answered by people shipping code.
What Autheo actually is. They call themselves the Layer-0 OS for the agentic web, and announced their "Internet Operating System" on 30 June — three weeks ago. Four pieces:
① TheoID — W3C-compliant DIDs for users, services and AI agents
② PQCNet — post-quantum crypto on the NIST standards (ML-KEM, ML-DSA, SLH-DSA)
③ Sovereign Cosmos SDK Layer 0 with native IBC
④ EVM-compatible L1 on CometBFT
Testnet drew 350,000 wallets and 60,000 contracts in its first year. Founded 2021. This is not a deck.
They already answered Koverko's question. Polymesh made you ask whether an agent inherits its deployer's identity or gets its own. TheoID gives it its own. That is the cleanest answer anyone at this conference has to the entity problem — so do not re-ask it. Start from it.
"TheoID gives an agent its own DID instead of inheriting the deployer's, which is the cleanest answer to the entity question I've heard all week. But Autheo is a coordination layer, and coordination is exactly where my problem starts. When three agents with three TheoIDs coordinate to produce one outcome, across two chains over IBC, does the coordination layer record what each one contributed — or only that they participated? Identity answers who. I need what share, and I genuinely don't know whether that belongs inside the identity layer or above it."
Why this is the best conversation available today. Cross-chain multi-agent settlement is the hardest possible case for attribution, and they have built the substrate it would run on. Edward Johnson is their Chief Product Officer, not a sent presenter — and a workshop means a small room where a real question is welcome rather than an intrusion.
⚠ Now a three-way collision. Autheo runs 2:35–3:05 upstairs, Agnic 3:00–3:30 in Purple, Nadiminti (AWS) 3:05–3:25 on Upper Stage.
Revised call: Autheo first, then Nadiminti at the Upper Stage exit — both are on Level 2, so you never leave the floor. Autheo ends 3:05, Nadiminti ends 3:25, and you are already standing there. That costs you Agnic, which is the right trade: Moghadam is a workshop you can catch cold, Autheo is a thesis-level conversation you cannot.
You have been citing AWS's limitation all week. This is AWS. Do not soften it and do not turn it into a gotcha — you are handing him a problem his own product roadmap will hit, from someone who has thought about it longer than the room has.
Verified this morning, from AWS's own technical deep dive on AgentCore Payments. The service does exactly what it says: individual agents pay for services via stablecoin microtransactions, with per-session budget guardrails, orchestrating x402 v1 and v2, Stripe and Coinbase. "AgentCore payments orchestrates payment processing… applies the payment-limits guardrails, and signs the transaction to generate a payment proof."
What the document never mentions once: multi-party settlement, fee splits, revenue sharing, or attribution of contribution. Not deferred, not roadmapped — absent. It is a one-payer, one-payee model with excellent guardrails around it.
It also concedes the cross-provider problem rather than solving it: accessing paid services still requires separate billing accounts with each provider. AgentCore abstracts that away for one agent; it does not settle it between many.
"AgentCore Payments handles one agent paying one seller really well — guardrails, payment proof, x402 v1 and v2. What I can't find anywhere in the docs is the many-to-many case. When four agents from three vendors co-produce one purchase, AgentCore proves the payment happened but nothing records what each contributed, so the split has to be agreed in advance by a human. Is multi-party settlement out of scope on purpose, or is it just not this release?"
Either answer is valuable. "On purpose" means AWS has decided the attribution layer is somebody else's product — which is the clearest market signal you will get this week. "Not this release" tells you your runway.
Narrative amplification — the media venture door, not the protocol door.
"I have a film/media venture plus a sovereign-AI story that needs the right narrative push — what's Luna's sweet spot?"
Log today's three. Haymond, Sinclair, Chen are all still uncaptured on /contacts. The one-tap chips are pre-written with what actually happened — you only add what they said. Twenty seconds each, and it decides whether tomorrow's follow-ups land specific or generic.
Charge the phone. Nine board targets tomorrow against today's six.