Shaw Walters, founder of Eliza OS, said in a lengthy post that the ai16z token is finished, the foundation has been shut down, and there will be no buyback for holders.
Walters said holders can either sell or “find a group of people to pump it,” but should not expect him to do anything else for the token. He also described the crypto community as “a bunch of spoiled crybabies” and said lawyers had sued him on behalf of token holders. According to his account, the Eliza OS Foundation no longer had the money to fight those cases and its remaining funds had already been paid out.
He added that his wallet once held $25 million worth of ai16z tokens, none of which he sold, and that position eventually went to zero. Walters said he is now living off his savings, staying in a run-down small bedroom in San Francisco, and writing code every day.
The post has pushed the market back toward the crypto AI agent frenzy that swept through the sector in late 2024. ai16z was once one of the names most closely tied to that trade. Its collapse, now acknowledged by its founder in blunt terms, has reopened a broader question about how crypto and AI drifted apart even as AI products became more real.
How ai16z became an early symbol of the AI agent token boom
In October 2024, ai16z launched a crowdfunding campaign on DAOs.fun with a target of 420.69 SOL, roughly $75,000, to build an investment fund managed autonomously by AI.
The number 420.69 carried obvious crypto-culture symbolism from the outset. Even so, the market took the idea and ran with it.
In less than three months, ai16z reached a $2.6 billion market capitalization. It also helped drive an entire category higher. GOAT came earlier, VIRTUAL followed, and a growing list of agent-themed tokens entered the market, taking the broader AI agent segment from zero to nearly $10 billion.
That move looked especially aggressive in the context of the time. At the end of 2024, ChatGPT had only just turned two years old and was still frequently criticized for hallucinations. Claude did not yet have tools that could directly operate a computer. For most people, the idea of an AI agent was still largely conceptual. Crypto markets had already assigned it a price.
Questions over the “agent” appeared almost immediately
One detail, the article noted, drew far less attention at the time than it does now.
Less than a week after ai16z went live, a crypto media report questioned whether the project’s core AI agent, the so-called “Marc AIndreessen” that was presented as making autonomous investment decisions, was actually being operated by a human rather than thinking independently as an agent.
A similar line of criticism later appeared around aiIxbt, a well-known market-analysis agent in crypto.
Back then, though, those doubts had little impact. The market cap kept rising, the community kept cheering, and more AI agent tokens kept launching. Looking back, the article argues that this may have been the clearest snapshot of the crypto AI narrative: whether the product was fully real mattered less than whether the story was early enough to attract liquidity and speculation.
That was the trade in its pure form. The token came before the product, and price came before the technology.
By 2026, AI agents arrived — but not through token projects
The article then shifts to 2026 and says AI agents had by that point actually arrived. In Western markets, it pointed to CodeX and Claude. In China, it mentioned products such as Workbuddy that fit local usage patterns more closely. Many of the features once left inside crypto AI white papers — automated market analysis, workflow handling, sentiment monitoring and similar functions — had by then become real.
But the teams delivering those products were not the token issuers from the previous cycle.
Money moved instead to Anthropic, OpenAI, and coding teams in San Francisco and Hangzhou. Revenue flowed to aggregation platforms such as OpenRouter, to cloud providers’ GPU bills, and to real API calls being made at scale.
Walters wrote: “AI doesn’t have these problems. The people there are optimistic and strong, building the future instead of tearing each other down. Those are my people.”
That line became one of the sharpest parts of the piece. A founder once associated with the crypto AI agent narrative ended up saying, in effect, that he saw himself on the AI side rather than in crypto.
AI companies without tokens built workable businesses first
The comparison drawn in the article is direct. The successful AI applications and model companies of today do not appear to need tokens. Anthropic charges for subscriptions and API usage. OpenRouter runs a routing and aggregation business and, according to the article, generates annual revenue above $100 million. These companies did not issue tokens and did not build DAOs, yet their business models worked and developers kept using them.
For many crypto AI projects, the situation looked far harsher.
When token prices were high, everything seemed to be in place at once: treasury money, an active community, and a narrative the market wanted to buy into. Once prices broke down, all of that started disappearing at the same time. Developers left because salaries paid in tokens had lost most of their purchasing power. Communities turned into rights-protection groups. Teams that still wanted to build sometimes found they could not even afford to hire one engineer.
Walters also said the Eliza framework itself was not dead. The code is still being updated, and companies are still discussing partnerships around it. As an open-source AI framework, it can survive without a token, whether through services, acquisition, or another route entirely.
The article added another detail: among the active contributors to Eliza over the past week was Claude. In its reading, a developer or AI project without a token attachment may actually be worth more in the 2026 market than one tied to a token structure.
The link between crypto and AI is changing, not disappearing
The piece does not end in outright pessimism. Its argument is that what broke was the path of issuing tokens around AI concepts, not every possible connection between crypto and AI.
Some of the projects still standing are doing something very different from ai16z. It cited Bittensor, which is building a decentralized AI compute market, and Render, which operates a distributed GPU rendering network. In both cases, the token is used to coordinate a real computational resource. The article says there is still a market for that, even if the premium and the narrative force are weaker than they were during the bull run.
It also pointed to practical use cases inside crypto itself, including on-chain data analysis, vulnerability scanning, whale-wallet tracking, sentiment monitoring, and smart contract strategies on DEXs. AI, in its view, is gradually becoming useful across those areas. The piece gave one example: by 2026, a professional crypto trader opening Hyperliquid will likely have a Claude- or GPT-supported strategy window open nearby.
That leads to its final turn. The relationship between crypto and AI has not ended, but the direction needs to be reversed. Rather than turning AI into another token narrative, crypto may need to accept that AI is becoming a layer across almost every sector, including crypto itself.
As for how large the crypto market can still become, the article leaves that question tied to when BTC improves. On the individual level, its conclusion is more practical: developers who drop the token burden may find more value in today’s AI world than tokens can offer, while crypto participants who genuinely adapt to AI may gain more than the old model built on rough speculation and narrative momentum.