What Actually Happened to Web3: A No-BS Retrospective

In 2022, we published a guide explaining Web3, calling it "a digital Robin Hood" that would take power from big tech and give it back to communities. We meant it at the time. A lot of smart people did.

Four years later, some of that optimism looks naive. Not all of it, but enough that it's worth an honest look back.

NFTs as mainstream assets. The floor fell out. Monthly NFT trading volume dropped from $2.8 billion in January 2022 to under $100 million by mid-2023 and never recovered. The Bored Apes, the CryptoPunks, the profile picture projects that were supposed to be "the future of digital ownership" are now mostly held by people who can't sell them without taking a 90%+ loss. The underlying technology (token standards, provenance tracking) still works. The market for overpriced JPEGs does not.

The metaverse. Meta burned over $50 billion on Reality Labs between 2020 and 2025. The result: Horizon Worlds never reached mass adoption, and the term "metaverse" largely disappeared from corporate earnings calls by 2024. Decentraland and The Sandbox saw daily active user counts in the low thousands, not the millions that were projected.

Web3 gaming. Play-to-earn collapsed when the token incentives dried up. Axie Infinity's economy imploded after the Ronin bridge hack ($625M stolen) and the realization that new player money was funding old player withdrawals. The model was, in hindsight, structurally a Ponzi.

The "decentralize everything" thesis. Decentralized social media (Lens, Farcaster) remained niche. Decentralized storage (Filecoin, Arweave) found some use cases but didn't replace AWS. The average user never cared about owning their data enough to accept worse UX.

DeFi, in a reduced form. Decentralized exchanges (Uniswap, Curve) and lending protocols (Aave, Compound) are still running. Total value locked in DeFi stabilized around $50-80 billion, far from the $180B peak but also far from zero. Stablecoins (USDT, USDC) became genuinely useful for cross-border payments, particularly in countries with unstable currencies. This is probably the most real, lasting use case to come out of the entire Web3 era.

Blockchain infrastructure. Layer 2 solutions (Arbitrum, Optimism, Base) solved a lot of Ethereum's scaling problems. Transaction costs dropped from $50+ during the 2021 mania to fractions of a cent. The technology got better, even as the hype disappeared.

Zero-knowledge proofs. The privacy and verification technology that emerged from blockchain R&D has applications well beyond crypto: identity verification, supply chain provenance, and secure computation. This may end up being the most valuable technical export of the Web3 era.

Tokenized real-world assets. Banks and financial institutions quietly started tokenizing bonds, treasury bills, and real estate on blockchain rails. BlackRock launched a tokenized money market fund. This isn't the "democratize finance" revolution Web3 promised, but it's real institutional adoption, just not in the way anyone expected.

We're watching the AI industry run a similar pattern. The parallels aren't subtle:

The promise of transformation was real, but the timeline was compressed by hype. Web3 advocates in 2022 talked about blockchain the way some AI advocates in 2025 talked about agents: as if the technology was 6 months from replacing entire industries. The technology was real. The timeline was fiction.

Capital flooded in before use cases were proven. In Web3, this meant thousands of projects with tokens but no users. In AI, this looks like companies deploying AI tools before defining what problem they're solving, a pattern we see regularly when assessing companies' AI readiness.

The "build it and they'll come" approach failed. Web3 projects built decentralized apps that nobody asked for. AI projects that skip the problem-definition step face the same outcome: technically interesting solutions with no business impact. That's why measuring ROI from day one matters so much.

Vendor claims outpaced reality. In Web3, it was "agent washing" equivalent: projects calling themselves "decentralized" while running on a single server. In AI, Gartner found that only 130 of thousands of vendors claiming "agentic AI" capabilities were legitimate. The rest is rebranding.

Here's where the parallel breaks down: AI is producing real, measurable business outcomes in a way Web3 never did at scale.

Nobody saved 40% on compliance review time using blockchain. Nobody cut customer support costs by 30% with an NFT. AI is doing both of those things, today, in production, with provable numbers.

The risk isn't that AI is a bubble like Web3. The risk is that companies repeat Web3's mistakes: investing without strategy, chasing hype instead of use cases, and measuring activity instead of outcomes. Romania's AI adoption gap is partly a reflection of this: companies waiting because they've been burned by hype cycles before, and companies rushing in without a plan because they're afraid of being left behind.

The lesson from Web3 is simple: the technology can be real and the market can still fail. Execution, not enthusiasm, determines whether an investment pays off.

We published that Web3 guide in good faith in 2022. We were wrong about some things and right about others. The zero-knowledge proof section aged well. The NFT section did not.

Publishing this retrospective is part of the same instinct: tell the truth, even when it means admitting the last take was too optimistic. That's what "No BS" means in practice.

If you're evaluating AI for your business and you've been through a hype cycle before, good. That skepticism is useful. Channel it into asking the right questions: what problem are we solving, how will we measure success, and should we build custom or buy off the shelf?

The companies that got Web3 right were the ones that ignored the noise and focused on the infrastructure. The companies that will get AI right are doing the same thing.