Tracing the ghost in the machine
Last week, a company that trains humans to wield AI tools quietly locked up $570 million in fresh capital, pushing its valuation to $2.1 billion. The market cheered. Founders posted victory laps. But as I read the sparse press release—no customer names, no unit economics, no forward guidance—I felt the familiar chill of a story being sold more than a story being told.
I’ve been here before. In 2017, at 32, I spent 60 hours auditing the Solidity of a hyped ICO called Ethos. I found three critical re-entrancy bugs. My report got me called a “shill” by the project’s Telegram army. Six months later, the project imploded. The ghost in that machine was code masquerading as trust. Today, the ghost is different: it’s centralized training infrastructure masquerading as the solution to AI’s talent gap.
Multiverse’s raise is not just a funding round. It’s a flag planted on the terrain where AI and human capital intersect—a terrain that crypto protocols have been eyeing for years. But the narrative being sold—that we need centralized tutors to scale AI literacy—is a comfortable lie. The truth, as always, lives on-chain.
Context: The Rise of the AI Apprenticeship
Multiverse, founded in 2016 by Euan Blair (son of former UK PM Tony Blair), is a vocational training platform that places apprentices into paid jobs at companies like Google, Morgan Stanley, and Visa. The curriculum spans software engineering, data analytics, and now, increasingly, AI skills. The model is classic B2B2C: enterprises pay for a cohort of apprentices, the government subsidises part of the cost, and graduates get a credential (often a level 4-6 UK diploma) and a job.
It’s a proven model. The problem is that it’s scaling at a moment when the very nature of “skill” is being redefined by generative AI. In 2023, Multiverse reported revenue of roughly $120 million. At a $2.1 billion valuation, that implies a price-to-sales multiple of 17-18x—rich for edtech, but justified by 50%+ growth. The $570 million war chest (likely from General Catalyst, Index, or a sovereign fund) is earmarked for U.S. expansion and AI curriculum development.
But here’s where the cryptographer in me raises an eyebrow. Multiverse’s core value proposition—trusted credentialing—is exactly what blockchains were invented to solve. Every time a graduate shows a Multiverse badge to a recruiter, they are relying on a centralized database to confirm the truth. The same database can be frozen, hacked, or politically sanitized. Code is law, but trust is fragile.
Core: The Narrative Mechanism and the Sentiment Trap
To understand why $570 million flowed into a centralized training platform, you have to map the narrative resonance of the moment.
The macro story is undeniable: AI job postings have surged 130% year-over-year. Companies are desperate for people who can fine-tune models, build RAG pipelines, and deploy agents. The supply of such talent is minuscule. Any solution that promises to mint more of it at scale is a magnet for capital.
But beneath that surface, a specific narrative mechanism is at work: the hero’s journey of the late adopter. Corporate L&D (Learning & Development) teams, burned by years of failed MOOC experiments, now see AI as the ultimate justification for outsourcing training. They don’t want to build internal academies—they want a vendor with a trusted brand to do it for them. Multiverse is that vendor.
The sentiment data backs this up. In the last six months, mentions of “AI apprenticeships” on LinkedIn grew 340%. The Crypto Briefing article that broke the news (yes, a crypto outlet covering edtech—a sign of narrative bleed) framed the raise as a “paradigm shift.” But what if it’s actually a reversion to a medieval paradigm—the guild master controlling the apprentice’s credentials?
Let me draw on a personal experience. During the 2020 DeFi summer, I worked with three independent researchers to analyze Compound’s governance. We found that a single admin key could change the COMP distribution logic. We published “The Illusion of Decentralization.” The protocol survived—but our caution saved us from over-leveraging. Today, Multiverse’s centralized database of credentials feels like that admin key. It works—until it doesn’t.
The real insight, however, is that Multiverse’s raise exposes a liquidity fragmentation analogous to what we see in Layer 2 scaling. Just as dozens of L2s split the same small user base, dozens of AI training platforms (Coursera, Udacity, Springboard, General Assembly) are fighting over the same pool of aspirants. Multiverse’s edge is its apprenticeship model—real jobs, real projects. But that model is capital-intensive and local. It doesn’t scale to the millions of learners in Lagos, Bangalore, or São Paulo.
Authenticity is the only scarce resource. And authenticity in credentialing cannot be centrally manufactured. It must be consensus-driven.
Contrarian: The Decentralized Credentialing Blind Spot
The contrarian angle is this: Multiverse’s massive raise will actually accelerate the adoption of on-chain skills verification, not slow it down. Here’s why.
First, as more people graduate from centralized programs, the demand for portable, censorship-resistant credentials will skyrocket. A Multiverse badge is worthless if the company goes bankrupt or changes its curriculum under political pressure. An on-chain soulbound token (SBT) issued by a DAO of peers—verified through zero-knowledge proofs of project contributions—cannot be revoked by a single board.
Second, the emerging Learn-to-Earn and Proof-of-Learning protocols (RabbitHole, LearnWeb3, Decentralized Academy) are already testing this model. They issue tokens for completing on-chain tasks, creating an immutable record of skills. The UX is terrible today—wallet creation, gas fees, fragmented interfaces—but the trends in account abstraction (ERC-4337) and Layer 2 scaling will solve that within 12 months.
Third, AI itself will commoditize the training curriculum. As foundation models like GPT-5 and Claude 4 become capable of generating personalized, adaptive tutorials, the value of a fixed syllabus plummets. What remains valuable is the verification that a human actually applied that knowledge to solve a real problem. And that verification is best performed by a network of peers, not a central authority.
Let me offer a hypothetical. Imagine a future where a developer in Nairobi wants to prove she can build a DeFi dapp. She doesn’t need a Multiverse apprenticeship. She deploys a smart contract on a testnet, stakes a small bond, and asks a DAO of experienced builders to review it. The reviewers earn reputation tokens for honest assessments. The developer gets a non-transferable SBT that records her achievement. Recruiters query the on-chain record via a ZK oracle. No intermediaries. No frozen databases.
This isn’t science fiction. It’s being built right now by projects like Questbook (audited by the team I respect), DeveloperDAO, and the OpenGuild ecosystem. The ghost in the machine is not a centralized server—it’s the consensus of the collective.
My contrarian take, therefore, is that Multiverse’s $570 million is a peak-Centralized-Training signal. It will inspire a wave of copycats, but their very success will highlight the fragility of their model. The smart money, after this raise, will start looking at decentralized credentialing infrastructure—just as, after the ICO boom of 2017, smart money turned to DEXs and L2s.
Takeaway: Listening to the Silence Between the Blocks
The market’s silence about Multiverse’s risks is deafening. No one is asking: What happens when a government demands to see the list of AI-trained graduates? What happens when an employer challenges the validity of a credential? What happens when the company’s incentives diverge from the learner’s (e.g., needing to fill quotas for lower-skilled roles)?
These questions don’t have answers in a centralized database. They have answers in a transparent, verifiable, immutable ledger.
So here’s my forward-looking judgment: The next narrative in AI + crypto will be the Tokenized Human Capital Stack. Not just compute tokens (Render, Akash) or model tokens (Bittensor), but learning tokens—tokens that represent a human’s authenticated competence. The funding that flowed to Multiverse today will flow to these protocols tomorrow. And when it does, the ghost will finally be visible: not a corporation, but a community.
Until then, I’ll keep listening to the silence between the blocks. Because that’s where the truth lives.