Eighteen Months to Midnight: A Telecom CEO's AGI Prophecy and the Authority Gap

Eighteen Months to Midnight: A Telecom CEO's AGI Prophecy and the Authority Gap

When the chief executive of a telecommunications company announces that artificial general intelligence will arrive within eighteen months, the correct reflex is not to update your calendar. It is to update your model of who is permitted to narrate the future — and then to ask why a crypto news outlet, of all the places on the internet, decided this particular prophecy deserved a headline.

The claim belongs to Dan Schulman, the former PayPal chief who now runs Verizon. The venue was some public appearance, the kind that produces a quotable fragment and then evaporates into the archive. The conduit was Crypto Briefing, a blockchain news platform whose readers, by and large, hold positions in tokens and protocols. And the substance, once you strip away the rhetoric, is nearly nothing: a time horizon, two derivative judgments about industry disruption and workforce retraining, and a word — "AGI" — that nobody in the sentence bothered to define.

I have spent the better part of two decades watching technically hollow words do enormous work in markets. I audited the early oracle mechanisms in Augur and Gnosis back in 2017 and found three logic flaws the token sales never mentioned. I sat inside Curve governance during DeFi Summer and watched a geometric invariant formula harden into a cultural identity. And I have learned, repeatedly and at cost, that the most dangerous sentence in this industry is never a lie. It is a true sentence with the wrong noun in it.

So let us take the noun seriously. "AGI" is doing something here. And "eighteen months" is doing something else entirely.

The signal chain

To understand what this headline is, you have to understand what it is not. It is not a research result. It is not a model card, a benchmark, a paper, a compute curve, or a funding round. It is a single human being, standing in front of a microphone, offering an intuition about the future of his own industry's supply chain. Everything downstream of that microphone is amplification.

The chain runs like this. A business executive speaks. A generalist reporter captures the most quotable fragment. A crypto publication re-packages that fragment for an audience that lives on narrative. The fragment then becomes a data point in a thousand threads about "AI x Crypto." At every hop, context dies. By the time it reaches you, you cannot know whether Schulman was answering a question about Verizon's internal tooling, whether he was being deliberately provocative, or whether he was simply speaking loosely in a fireside chat where the entire point is to be quotable. The original speech act is gone. What remains is the vibe.

This matters because the authority of a prediction is not a property of its confidence. It is a property of the predictor's track record in the relevant domain. A quantitative researcher who has tracked the scaling of training compute for a decade earns a specific evidentiary weight when she talks about capability timelines. A telecom operator who has spent his career in payments and connectivity earns a different weight — not zero, but different. When a news platform promotes the second voice using the rhetorical register of the first, that is what I call authority misattribution. The confidence travels. The competence does not.

I first learned to distrust the traveling confidence in 2017, when I was still a mathematician pretending to be a developer. I was auditing the oracle layers of Augur and Gnosis, the two prediction-market protocols that were supposed to make betting on reality trustless. The marketing was immaculate. The whitepapers had diagrams. The token sales cleared. And underneath, three logic flaws sat quietly in the resolution mechanisms, waiting for the precise market conditions that would expose them. None of the people amplifying those projects had read the resolution code. They had read the narrative. That is the same mechanism at work here, only the noun has changed from "prediction market" to "AGI."

We didn't ask whether the prophecy was true. We asked who benefits from it being believed.

The definitional war

The word itself has been quietly hollowed out, and the hollowing is the story. In corporate settings, "AGI" almost never means the academic definition — cross-domain generalization, autonomous goal-setting, continual learning. It means something more mundane and more useful to a boardroom: systems that can absorb most white-collar workflows. That is a legitimate thing to predict. But it is not AGI, and conflating the two is exactly how a modest operational forecast becomes an epochal headline.

I have watched this same sleight of hand in my own field. The word "decentralization" did the same work in 2017 that "AGI" does today. It was elastic enough to mean whatever a given deck needed it to mean, and precise enough to sound like a guarantee. A project could be decentralized in its governance, its token distribution, its validator set, or its rhetoric, and the word never told you which. The same is true of AGI now. It can mean a benchmark, a business capability, a philosophical horizon, or a marketing category. The ambiguity is not a bug. It is the product. A word that means everything cannot be falsified, and a claim that cannot be falsified cannot be wrong — which is precisely why it makes such an effective headline.

So when I read that a telecom CEO has predicted AGI in eighteen months, my first question is not "will it happen." My first question is "what did he mean by the word, and did the reporter ask." The answer, almost always, is that nobody defined it, and nobody asked, because the ambiguity is what made the quote travel.

The four gaps nobody put in the headline

Here is where my old life as a cryptographer reasserts itself. When I audited those early prediction-market oracles, the exercise was never "does the code run." The code ran. The question was whether the system's guarantees held at the edges — under adversarial input, under thin liquidity, at the precise moment when the assumption everyone relied on quietly stopped being true. Prediction markets taught me to hunt for the failure mode that the demo never shows.

AGI claims deserve identical treatment. So let me name the four gaps any honest capability timeline has to cross, and then ask whether eighteen months is even the right unit of measurement.

The first gap is continual learning. Current frontier systems are, in a precise sense, frozen after training. They can retrieve and reason within a context window, but they do not durably update their weights from experience. Ask a model to learn a new skill today and retain it tomorrow without a retraining run, and you are asking for something the architecture was not built to give. Continual learning without catastrophic forgetting is an open research problem, not an engineering backlog item. It has been open for decades, and no announcement at a telecom keynote closes it. I have watched my own field make the same mistake: promising that a protocol upgrade would deliver a property — finality, scalability, privacy — that the underlying cryptography could not yet provide. The upgrade shipped. The property did not. The gap between the two was filled with narrative, and the narrative held for about one market cycle.

The second gap is causal reasoning. Systems that predict the next token are extraordinarily good at correlation and extraordinarily fragile at intervention. They know that roosters crow before dawn. They are far less reliable at the question of what happens if you remove the rooster. For an agent expected to act in the world — to run a network, to manage a supply chain, to renegotiate a contract — the ability to reason about interventions rather than merely observe correlations is not a nice-to-have. It is the line between a tool and a liability. In DeFi I saw this distinction cost real money. A model that knows the historical correlation between a liquidity pool's depth and its slippage is not the same as a model that knows what happens to slippage when a whale withdraws during a cascade. Correlation describes the past. Causation survives the future.

The third gap is a stable world model. Genuine general intelligence requires an internal representation of reality that supports counterfactual simulation: if I do this, then that. Current systems approximate this with enormous statistical pattern-matching, and the approximation is remarkable, but it is not the same object. A model that can describe a physical process is not a model that can predict the process under conditions it has never seen. And the world, especially the adversarial world of markets and networks, specializes in conditions never seen. This is the same lesson that every risk model teaches, and the same one that Three Arrows Capital ignored in 2022. Their models were calibrated on a world where correlations held. The world stopped cooperating. The models did not update. The fund did not survive.

The fourth gap is sample efficiency. A human child learns the concept of "dog" from a handful of examples. Frontier models require orders of magnitude more data to reach comparable reliability, and they require it again for every new domain unless they can transfer — which loops back to the first gap. This is not a footnote to the timeline. It is the timeline. If learning remains data-hungry and non-transferable, then "general" intelligence is a word we are borrowing from a future we have not engineered.

None of these four gaps appeared in the headline. None of them will appear in the thread that quotes the headline. And yet they are the entire substance of the question. The prophecy gave us a date. The engineering gives us a distance. The distance is not eighteen months. It is however long it takes to close four research problems that nobody has closed in the entire history of the field.

The graveyard of timelines

There is a reason I treat aggressive AGI timelines with such suspicion, and it is not cynicism. It is the historical record. The graveyard of confident predictions is vast and well-marked, and I have walked through it more than once.

Researchers have been forecasting human-level machine intelligence on short horizons for as long as the field has existed. Each decade produced its own eighteen months, dressed in the vocabulary of its era. In the 1960s it was ten years. In the 1980s it was the expert systems that would run the corporation. In the 2010s it was the deep-learning wave that would generalize any day now. The dates changed. The confidence did not. What changed, occasionally, was that the goalposts moved — the definition of "human-level" quietly receded as the target approached, because the target kept turning out to be harder than the vocabulary suggested.

I do not cite this graveyard to say that AGI is impossible. I cite it to say that confident short-horizon predictions from people outside the research frontier have an atrocious base rate, and that the correct prior is skepticism, not excitement. When a lab scientist with a decade of compute-tracking says eighteen months, I listen carefully and update slightly. When an operator of pipes and subscriptions says eighteen months, I reach for the base rate, and the base rate says: wait for the budget.

The arithmetic of eighteen months

Let us do the arithmetic that the coverage skipped. If the remark was made in late 2024 or early 2025, then eighteen months points to mid-2026. That date is not random. It lands squarely inside the window that several frontier labs have publicly sketched for "agentic AI" — systems that can execute multi-step tasks with limited supervision, browse, call tools, and chain actions together. Agentic AI is real and rapidly improving. It will change a great deal of knowledge work. It is also not AGI, and the gap between the two is precisely the four gaps above.

Here is the part that should interest anyone who has watched a hype cycle from the inside: eighteen months is a psychological number, not a physical one. It is short enough to manufacture urgency — to justify a budget line, a strategy pivot, a headline — and long enough that the speaker cannot be embarrassed within the news cycle. By the time the date arrives, the original quote will be a forgotten hyperlink. This is the classic temporal framing strategy of corporate discourse, and I have seen its mirror image in crypto countless times. "Mainnet in six months." "Full decentralization by Q3." "The merge is imminent." The date is never a commitment. It is a rhetorical device with a calendar attached.

I am not saying Schulman is dishonest. I am saying he is doing what executives do: converting an uncertain future into a memorable interval that mobilizes present action. The problem is not the intention. The problem is what the interval does to everyone downstream who lacks the tools to discount it — the retail investor who reads the headline as a signal, the employee who reads it as a threat, the analyst who reads it as a data point. An interval is a tool for the person who wields it and a trap for the person who receives it.

The authority gap

There is a second, quieter signal buried in this story, and it is the one I find most revealing. The platform that ran the AGI headline is a crypto platform. That is not an accident. It is a symptom of a convergence.

For most of the last decade, AI and crypto ran on parallel tracks. Both claimed to be building the infrastructure of a new order. Both depended on the emotional fuel of retail conviction. Both promised to disintermediate institutions they quietly envied. What is new, in 2024 and 2025, is that the two narratives have fused into a single story: technology will dissolve the old world, and you can hold a position in the dissolution.

This fusion explains a great deal. It explains why a telecom CEO's speculative remark found a home on a blockchain outlet. It explains why the AI narrative has become the crypto narrative's most reliable borrowing source — tokens for decentralized compute, tokens for data sovereignty, tokens for AI agents that supposedly transact on-chain. And it explains why the industry's hunger for AI legitimacy has become, in my view, its most underrated vulnerability.

I want to be precise here, because this is where I part ways with the cheerleaders. The crypto industry's recurring fantasy is that AI will need blockchain — that autonomous agents will require trustless rails, that data provenance will demand on-chain attestation, that compute will be tokenized and traded permissionlessly. Some of this will be true at the margins. Most of it will not. The uncomfortable fact, the one my consulting work keeps confirming, is that traditional institutions do not need your public chain to do AI. They have private compute, private data, private compliance, and private incentives. They will build on permissioned infrastructure and call it innovation, and they will keep your token out of it entirely.

This is the same lesson RWA has been teaching for three years while everyone pretended not to hear. Real-world asset tokenization has been a storytelling exercise dressed as a technical migration. The institutions that tokenize are the ones already inclined to; the ones that are not keep their assets exactly where they are, on ledgers they control. I have sat in rooms with the compliance officers of mid-sized firms, and their question is never "how do we get on-chain." Their question is "how do we get the efficiency without the exposure." The answer they find is almost always a private ledger with a public relations campaign. AI will follow the identical script. The convergence is real at the level of narrative and mostly hollow at the level of architecture.

And that is why the authority gap matters so much. When a telecom CEO says "AGI in eighteen months," and a crypto outlet amplifies it, what is actually being transmitted is not a technical forecast. It is a permission slip. The industry is being handed a reason to believe that its moment has finally arrived — that the AI wave will wash it into relevance. Open source isn't a business model. It's a philosophy of transparency, and transparency about why a story is being told is the first discipline any reader of this industry should learn.

The provenance problem

There is a deeper connection between AI and my own field that the convergence narrative never mentions, because it is inconvenient. It is the question of provenance — of where the data came from, and who owns the output.

I co-founded an educational platform for digital artists in 2021, and I mentored fifty emerging creators through the mechanics of minting, ownership, and smart-contract provenance. The central lesson I taught them was simple, and it is the same lesson that AI has now made unavoidable for the entire creative economy. Art isn't the file. It's who owns it. The value of a digital artwork was never in the pixels, which are infinitely copyable. It was in the traceable, verifiable chain of ownership that made one copy scarce and the others worthless. Provenance was the product.

AI has now dissolved that product. Generative models are trained on the accumulated creative output of the internet, often without consent, often without compensation, and they produce derivative works at a marginal cost of zero. The provenance chain that gave digital art its scarcity is precisely what AI destroys. And here is the irony that the convergence cheerleaders miss: the technology that could, in principle, restore provenance — on-chain attestation of origin and ownership — is being applied to the wrong layer. The industry is busy minting tokens for AI agents while the actual provenance crisis, in training data and creative rights, unfolds in courts and licensing negotiations that have nothing to do with any blockchain.

I find this genuinely sad, and I find it instructive. The crypto industry had a real, non-speculative contribution to make to the AI era — a provenance layer for creative and training data. Instead, it chose the more tradeable narrative. That choice tells you more about the industry's priorities than any whitepaper.

The retraining euphemism

Now the part of Schulman's remark that deserves the closest reading, because it is the part with human consequences. He paired his AGI prediction with a call for "workforce retraining." On its surface, this is the language of corporate responsibility. Read it the way an auditor reads a footnote, and it says something else.

In large enterprises, "retraining" is very often the soft perimeter around "role restructuring." Verizon employs roughly a hundred thousand people, a substantial share of them in customer service and back-office functions — precisely the workflows that agentic systems are already beginning to absorb. When a CEO forecasts capability eighteen months out and simultaneously emphasizes retraining, he is signaling a transition that has probably already been drafted in a strategy document somewhere. The retraining is the public face of the restructuring. It is not a lie. It is a framing.

I think about a day in the life of a claims adjuster, a first-line support agent, a junior network operations technician — the people whose tasks decompose most cleanly into the pattern-matching strengths of current systems. For them, the eighteen-month number is not an abstraction. It is a countdown with a paycheck attached. And the cruelest detail is that the retraining programs that follow such announcements have a documented tendency to underdeliver: skill mismatches, credential inflation, timing lags, and a quiet expectation that displaced workers will absorb the cost of their own redundancy.

I am not a technophobe. I have spent my career teaching people to use systems they were told they could not understand — stablecoin invariants, oracle mechanics, custody models. But teaching and replacing are different verbs, and honest writing requires keeping them apart. The ethical question is not whether AI will change work. It will. The ethical question is who bears the transition cost, and whether the people who bear it get a voice in the plan. A prophecy is cheap. A transition plan is expensive. Guess which one makes the headline.

Edge inference and the telecom repricing

There is a steelman for Schulman's remark, and intellectual honesty requires me to make it. As a telecom operator, he sits at a genuinely strategic chokepoint. If AI inference migrates to the edge — into cars, factories, hospitals, and real-time applications — then low-latency networks and edge nodes become newly valuable. The carrier stops being a dumb pipe and becomes a distribution surface for computation. That is a real thesis, and it is probably part of why a Verizon CEO is thinking publicly about AGI at all.

But notice the asymmetry. In that thesis, Verizon is a beneficiary of AI, not an author of it. It does not own the models, the chips, or the talent density that produce frontier capability. It owns the roads. Roads matter enormously — until someone routes around them. And the entities capable of routing around them are the hyperscalers, who build their own data centers, their own interconnects, and increasingly their own edge presence. The telecom operator's AGI optimism is, in part, a bid for relevance in a value chain where it risks becoming the least defensible layer.

This is a pattern I have seen before, and it is why I treat infrastructure optimism with care. In crypto, the equivalent move was the operator who insisted that "blockchain will need our validators." Sometimes true. Rarely decisive. The value accrues to whoever controls the scarce resource, and the scarce resource in AI is not connectivity. It is compute, data, and — above all — the small number of people who can push the frontier forward. Verizon has none of those in decisive quantity. It has an audience, and for eighteen months, an audience is enough.

The market signal

Let me now do what I actually do for a living, and read this as a market signal rather than a technical one. This is the work that pays for my newsletter, and it is the work I do for the institutional readers who are tired of hype-driven crypto journalism.

The signal is not "AGI is coming." The signal is that the AGI narrative has crossed from the technology sector into the boardrooms of traditional industry. When a telecom CEO — not a lab scientist, not a researcher, but an operator of pipes and subscriptions — feels compelled to speak publicly in AGI terms, it tells you that the cognitive threshold has been breached. Traditional-industry leaders now believe, or believe they must appear to believe, that AI is an existential variable. That belief has consequences that arrive faster than any capability does.

The first consequence is budget. Belief precedes capital, and capital follows belief with a lag of six to eighteen months. Expect traditional-industry AI budgets to accelerate, not because the technology suddenly improved, but because the perception of competitive necessity hardened. The beneficiaries are enterprise AI platforms, systems integrators, and the training-and-transformation firms that sell certainty to anxious boards. This is where I would look for durable revenue, not in the speculative tokens that borrow the AGI word for a whitepaper.

The second consequence is the opposite of a signal: it is noise. A prediction from a non-technical authority has no bearing on the fundamentals of any AI company. It is pure sentiment, and sentiment-driven rallies in "AI + telecom" or "AI + edge" concepts are precisely the kind of theme trade that punishes late entrants. I have a rule born from the Terra and Three Arrows post-mortems I wrote in 2022, under the banner of the hubris of leverage. When the least-informed voices become the most enthusiastic, you are closer to a local top than a local bottom. Extreme optimism from non-specialists has historically clustered near cyclical highs. I do not know if that holds this time. I do know that I have never regretted treating it as a warning rather than an invitation.

The third consequence is the one I care about most, and it is the meta-signal. The fact that this story was surfaced by a crypto outlet, framed for a crypto audience, tells me that the industry's narrative engine is now running on AI fuel. That is useful to understand precisely because it is so easy to mistake for fundamental progress. Narrative is not nothing. Narrative moves capital, and capital builds things. But narrative is also the thing that gets repriced when it meets a date it cannot keep.

I have watched this movie before, and I know how it ends. In 2021, the narrative was NFTs and the metaverse, and every traditional brand was minting a collection to prove it understood the future. By 2022, most of those collections were worthless and most of those brands had quietly deleted their announcements. The technology did not fail. The narrative outran the technology and then paid the difference. The AGI narrative is now running the same race, and the same wall is waiting at the same distance.

The contrarian read: the timeline is the wrong question

Everyone is arguing about whether eighteen months is too aggressive. I think that is the wrong argument entirely, and the obsession with it is itself the tell.

Here is the blind spot. The AGI timeline debate assumes that the important variable is when general intelligence arrives. But the systems that will actually reorganize the economy over the next several years are not general at all. They are narrow, agentic, and relentlessly deployed — automating specific workflows with brutal efficiency. You do not need AGI to eliminate a customer-service tier, to compress a compliance review from days to minutes, to replace a first draft with a prompt. The disruption is already here, and it is arriving on a schedule that has nothing to do with the AGI prophecy. The eighteen-month number is a distraction that lets everyone argue about an unfalsifiable future while the falsifiable present quietly reconfigures their jobs.

There is a second blind spot, and it is specific to my field. As AI agents begin to act on-chain — executing trades, holding assets, signing transactions, participating in governance — the crypto industry is about to collide with a problem it has spent a decade avoiding: legal personhood and liability. Most DAOs already operate with the legal status of a rumor. When the actors inside them are software, and the software makes a decision that harms a counterparty, the question of who is liable becomes not a philosophical puzzle but a lawsuit. I have watched this exact pattern before: a governance structure that looks decentralized in a forum post and looks like unlimited personal liability in a courtroom. Add autonomous agents to that structure and you have not solved the coordination problem. You have automated the exposure.

Decentralization is not a tech stack; it is a set of promises about who bears the risk. And right now, nobody is volunteering. The AGI conversation is about to hand this industry its most consequential liability question yet, and the industry is still debating whether the date is eighteen months or eighteen years. The date is not the question. The question is who goes to court when the agent misbehaves.

The takeaway

So here is my forward-looking judgment, offered the way I offer it to the institutional readers who pay for my newsletter. Treat the eighteen-month prophecy as a thermometer, not a clock. It does not measure the temperature of AI capability. It measures the temperature of institutional anxiety — and that anxiety is running hot, which is itself the most tradeable fact in the story. Watch what Verizon actually does next: the capital expenditure, the vendor partnerships, the hiring and attrition data. The prophecy is free. The budget is the evidence. When the budget appears, the prophecy was real. When it does not, we will have watched, once again, a date do the work that a strategy could not.

We didn't need AGI to get here. We needed a headline, an audience, and eighteen months of not checking. The next time a CEO hands you a date, ask for the ledger. That is where the truth keeps its books.