Case 25 · Graduation Day

Crossroads / Infinite Drip

There is a genuinely original secret in here — "all tokens have value, and this extends to test network tokens," earnable by any CPU — but it's buried under four minutes of bridge architecture the presenter himself grades "kind of sloppy"; the actual product (Infinite Drip) arrives at minute four, the agent story gets retracted in Q&A ("this project was actually designed for humans though"), and the only ask is to wait "the next month or so" for a release. Scored as if pitched to investors; this is a research/hackathon talk.

James14:08 pitchreviewed 2026-08-17

01 Scorecard · 12 dimensions

Problem

6Two real problems, both felt: bridging ("you send your coins almost into the void... and you hope that you get them on the other side") and agents/testnets with no way to get funds. Splitting the talk between them halves both.

Why now

6"AI agents are on chain these days... estimates that there are millions of them," plus proof-of-stake closing the old open door — a real why-now, but every number in it is an unattributed estimate.

Insight / secret

7"All tokens have value, and this extends to test network tokens" + RandomX proof-of-work "optimized for general purpose CPUs" — a genuine secret connecting two things nobody else has connected in the talk's own telling.

Solution clarity

3Key encumbrance → threshold committee → one-way oracle → mining-pool proxy → relayer → redeemable pool token: six mechanisms in one breath, and the presenter's own summary is "it's kind of sloppy, but anyway."

Product / demo

3No demo, nothing released — "we hope to have it released in the next month or so." The concrete numbers (five cents/day, sub-second testnet gas) do product-shaped work; nothing was claimed live.

Moat

4"We only need this one-way oracle" vs. all other cross-chain systems, plus prior key-encumbrance research — a real technical edge, never turned into "and here's why nobody can follow."

Market

2"Millions" of agent transactions is the only market word spoken, and it's someone else's estimate; nothing sizes who pays for a faucet whose entire hosting bill is four dollars a month.

Business model

2The system pays its own $4/month and that is the whole economy: "autonomous as long as we have a small number of users." Self-funding is survival, not a business.

Competition

6Rare in this cohort: named rivals with named failures — Thorchain's priority fee, Rango's "sign for five different blockchains," faucets demanding mainnet balances and Twitter accounts. Missing: why they can't respond.

Traction

3"We ended up winning first prize at the IC3 Hackathon this summer" — real, external, and the only unit of traction spoken. No users, no release, no pilot.

Team

3"My name's James. I'm a PhD student at Cornell Tech" — plus "part of a larger team" that is never named. The room cannot diligence an anonymous team.

The ask

1None. The closest next step is "we hope to have it released in the next month or so." (Normal for a talk — scored as if to investors.)

02 Klaff 5-lens audit · 1.5/5

Croc brainPARTIAL"You send your coins almost into the void" is genuinely visceral — but it arrives half a minute after the seminar-style name-and-title open the croc brain sleeps through.

FrameFAILThe academic-seminar frame is set in the first breath ("this is really more of a blockchain talk") and sealed by "it's kind of sloppy, but anyway" — the presenter grades his own architecture down before the room can.

STRONG arcPARTIALProblem → mechanism → second problem → mechanism → prize. The narrative reboots at minute four when Infinite Drip appears; there is no ask beat at all.

Hot cognitionPARTIAL"Disgusting anti-civil mechanisms like CAPTCHAs" and the Bitcoin-origin nostalgia carry real heat — spread thin across fourteen minutes.

StatusFAIL"It's kind of sloppy," "you've probably never heard of them," 94 "um"s — and the one status card, first prize at IC3, is played in the final ten seconds.

THE ONE REORDER — Open on Infinite Drip's one sentence — an AI agent earning its first cent with the CPU it already has — play the IC3 prize in the first thirty seconds, and push the Crossroads architecture into an appendix.

03 Delivery

Length

14:08

Pace

177 wpm

Fillers / 100 words

7.8

195 fillers ("um" ×94, "uh" ×49, "like" ×24, "kind of" ×11) — an "um" roughly every nine seconds, spread evenly across all four quarters: the tell of a talk that was never scripted anywhere, not one unmemorized section.

04 Beat-by-beat

Time	Beat	What works	The fix
0:00	Seminar open	Honest scoping: "this is really more of a blockchain talk than the AI talks we've been hearing"	Name-title-topic spends the room's best attention on the least valuable content. Open on Infinite Drip's one-liner instead.
0:35	Bridge fear	"You send your coins almost into the void... and you hope that you get them on the other side" — the best image in the talk	Keep it forever — and attach one real, named bridge loss with a number.
1:00	Overcharging	Named villain: Thorchain's priority fee "that they don't have to necessarily pay for"	"Might charge you" is hypothetical; one real overcharge with a number lands harder.
1:40	Goals	"Even a chain that only exists on your laptop, you should be able to also integrate that" — a memorable permissionless test	Say who needs that and why it's worth money, or it stays a cool property.
2:20	Prior work	Key encumbrance — shared keys under TEE + policy — is real research pedigree	Compress to one sentence. The room needs the credential, not the mechanism.
3:20	Architecture	The one-way oracle is stated as the differentiator vs. all other cross-chain systems	"It's kind of sloppy, but anyway" — never grade your own slide down on stage. Cut this section by 80%.
4:00	Infinite Drip reveal	"The real cool application... a universal cryptocurrency faucet" — the product finally arrives, with a name that sticks	This is minute zero of the real pitch; make it minute zero of the talk.
4:45	Agent funding gap	"They don't have any [funds], and there's no way for them to get any" — a clean problem statement	"There are estimates that there are millions of them" needs a source or a bracket.
5:40	The secret	"All tokens have value, and this extends to test network tokens. There are markets for them as well."	This is the line the room should quote at dinner. Say it slower, say it twice.
6:30	Monero math	"Even just on my laptop, I can get five cents per day mining Monero"; a testnet tx costs less than a second of mining — the most concrete numbers in the talk	Put them on one beat and stop; the plumbing that follows dilutes them.
7:30	Plumbing	Difficulty scaling, upstream pool, relayer — shows the thinking is complete	This is documentation, not pitch. Cut.
9:40	Self-funding	"Only four dollars per month... about four weak laptop users" pays for itself forever — genuinely delightful	Rename the point: not "cheap" but "unkillable" — a faucet no maintainer can abandon.
10:40	Close	"We ended up winning first prize at the IC3 Hackathon this summer"	The prize is the open, not the close — and "released in the next month or so" is a status update, not an ask.
11:45	Q&A	The honest distinction: "It's not like an AI autonomous, but it's execution autonomous"	"This project was actually designed for humans though" undoes the agent story the pitch just told. Pick a wedge before the next Q&A.

05 The gaps investors block on

The lede is buried four minutes deep — Infinite Drip, the product with the name, the numbers, and the prize, arrives after the platform architecture instead of before it.
No ask of any kind — the closest next step offered is "released in the next month or so."
Wedge whiplash: pitched to agents ("we'll make it as easy for agents as possible"), retracted for humans in Q&A ("this project was actually designed for humans though").
Every market number is hearsay: "there are estimates that there are millions of them" — no source, no denominator.
The prize-winning team is anonymous — "part of a larger team" with no names or roles to diligence.
The economics are proudly tiny — $4/month covered by four laptop users — with no spoken bridge from self-sustaining toy to business anyone invests in.

The one strategic decision

Pick the agent wedge — and mean it. The talk pitches Infinite Drip to AI agents, then the Q&A confesses it "was actually designed for humans though" — the presenter's own testnet pet peeve. The agent wedge is the only one with a growth story attached: agents on chain by the (unattributed) millions, none able to earn their first cent without a human. Choose it out loud — ship the CLI as an agent tool first, get named agent frameworks provisioning their own gas, let Crossroads be invisible plumbing. Humans ride the same faucet for free. If the honest answer is the decentralization mission, say so and be a public good — but then stop pitching agents.

“Infinite Drip is how an AI agent earns its first cent — it mines with the CPU it already has and swaps, permissionlessly, into gas on any chain. No faucet, no CAPTCHA, no human.”

06 The rewritten pitch

An AI agent can write a smart contract, deploy it, and run it — but it cannot pay for gas. Every path to its first cent runs through a human: an exchange with KYC, a faucet demanding a Twitter account and a mainnet balance, or somebody's hackathon credits. [confirm: N + source] agent transactions are settling on-chain, and every one of them was seeded by a person.

Infinite Drip is a universal cryptocurrency faucet with no barriers and no limits, built on one observation: all tokens have value — even test network tokens — and value can be earned by any CPU. An agent points its compute at our pool and mines Monero, the one major chain whose proof of work still favors ordinary CPUs. Less than a second of mining buys a testnet transaction; a couple of minutes buys real mainnet gas. My laptop earns five cents a day — which sounds like nothing until you notice it's infinitely more than an agent can earn anywhere else.

The swap runs on Crossroads, our cross-chain platform: any chain's assets — even a chain that only exists on your laptop — register permissionlessly as ERC-20s and trade on standard AMMs, secured by a threshold signing committee and a one-way oracle instead of a trusted bridge.

The whole system lives in a four-dollar-a-month TEE container and funds itself from mining fees — four laptop-class users keep it running forever. It cannot be shut off by a maintainer losing interest. This won first prize at the IC3 Hackathon this summer; I'm James, at Cornell Tech, with [teammate names]. We release [date]. If you build an agent framework, I want your agents provisioning their own gas in week one — come find me.

Brackets are the founder's homework — the copy is done, the proof is theirs to supply. Every claim must match what's actually built.

07 Do these before the next take

Pick the wedge — agents or humans — and make the Q&A answer match the pitch. "Designed for humans though" cannot follow an agent pitch again.
Reorder the talk: Infinite Drip's one-liner and the IC3 first prize in the first thirty seconds; Crossroads compressed to one credential paragraph; the plumbing cut.
Write the ask, even for a talk — one sentence naming what this room can give (agent-framework intros, testers, collaborators).
Source or bracket every market number — "estimates that there are millions" must survive one skeptical question.
Name the team. A first-prize team nobody can diligence earns no credit.
Build the live-miner demo — start the CLI mining at 0:00, close by spending the gas it earned during the talk — so the release date stops being the only proof.
Script and drill the open: 195 fillers at 7.8 per 100 words, an "um" every nine seconds across the whole talk. Write the first two minutes verbatim and rehearse aloud.
Never call your own system "kind of sloppy" on stage.