At the end of 2022 we published a short list of IT trends that had not lived up to expectations that year: mass crypto adoption, the metaverse and, somewhat boldly, artificial intelligence. A few weeks earlier ChatGPT had been released, and within a couple of years AI had become the dominant topic in software.
That makes the old article more useful now than when it was written. With a few years of distance, it is easier to see which hyped trends failed, which ones simply took longer, and which ones we misjudged. The lessons apply directly to the trends startups are being pushed toward today, including AI agents.
This article looks back at the biggest hype cycles of the early 2020s and turns them into practical rules for founders and product owners.
Mass crypto payments
The promise
Around 2021 many commentators expected cryptocurrencies to become a normal way to pay for everyday goods, with merchants and payment providers adding crypto checkout everywhere. A wave of startups launched exchanges, wallets and payment gateways to serve that demand.
What happened
The 2022 market downturn, often called the crypto winter, and the collapse of several large crypto companies, including the FTX exchange in November 2022, damaged trust and dried up funding. Everyday crypto payments did not become mainstream in most markets. Price volatility, transaction costs, regulatory uncertainty and a worse user experience than cards or instant bank transfers all worked against it.
What survived
The underlying technology did not disappear. Stablecoins, tokenisation of financial assets and blockchain-based settlement continue to attract serious work from financial institutions and fintech companies, and regulation in several regions has become clearer. The lesson is not that blockchain failed. It is that the consumer payments story was sold years ahead of the product, and that the useful applications turned out to be narrower and more regulated than the hype suggested.
NFTs as a mass market
The promise
Non-fungible tokens were presented as the future of digital ownership: art, collectibles, game items, event tickets and memberships, all traded on open marketplaces. Brands rushed to launch collections.
What happened
Trading volumes and prices fell sharply after the 2021 to 2022 peak, and many collections lost most of their value. For most buyers, NFTs turned out to be speculative assets rather than useful products. Many brand projects quietly closed.
What survived
The idea of verifiable digital ownership still has niches, for example in ticketing, loyalty schemes and some games. But in each case the value comes from a real use for the token, not from the token itself.
The consumer metaverse
The promise
When Facebook renamed itself Meta in 2021, it signalled a bet that people would increasingly work, shop and socialise in shared virtual worlds. Many companies opened virtual stores and offices and hired metaverse teams.
What happened
Mass adoption did not arrive. Headsets remained a niche product for most people, virtual worlds struggled to keep users coming back, and many corporate metaverse projects were scaled down or closed. Meta itself has continued to invest in virtual and augmented reality, but public attention, and much of the industry's budget, moved to AI.
What survived
Virtual and augmented reality are doing real work in training, design review, industrial maintenance and some healthcare uses. Lightweight augmented reality on phones and smart glasses continues to develop. As in the other cases, the narrow, practical applications survived and the all-encompassing vision did not.
The trend we got wrong: AI
Our 2022 article called AI a failure of that year, pointing to underwhelming project results, a shortage of skilled specialists and teams focusing on the wrong tools. All of that was true at the time. Within months, generative AI changed the picture completely.
This is an important lesson in its own right. The hype around AI had been building for years with modest results, and it was reasonable to be sceptical. But the underlying research kept improving, and when the technology crossed a usability threshold, adoption came very quickly. By 2026 AI coding agents, AI features in mainstream products and AI and machine learning development are everyday business, not experiments.
Even so, the earlier diagnosis still applies to many AI projects today: unclear goals, poor data and the wrong tools still produce underwhelming results. The technology works. Many projects built on it still do not.
Why hyped trends fail: common patterns
Looking across these cases, the same patterns show up again and again:
- The story arrives before the product. Investors and media describe the end state years before the technology or user experience can deliver it.
- Speculation hides real demand. When prices or valuations rise fast, it becomes hard to tell whether people want the product or just want to profit from the trend.
- The user experience is worse than the alternative. Crypto checkout was harder than paying by card. A headset meeting was harder than a video call. People rarely switch to something less convenient.
- Regulation is unclear or arrives late. Financial and data-heavy technologies eventually meet regulators, and the rules can change the business model.
- Funding moves in herds. When the money moves to the next trend, projects that depended on hype rather than revenue run out of runway.
Lessons for startups
1. Start from a customer problem, not a technology
The startups that survived these hype cycles were usually solving a specific problem for a specific customer, and the technology was a means to an end. If you cannot describe your product without the name of the trend, that is a warning sign.
2. Compare against the boring alternative
Ask what users do today without your product and whether your version is clearly better, cheaper or faster. If it is only newer, adoption will be slow.
3. Run small, cheap experiments
You do not need a full metaverse team or a token launch to test an idea. A prototype, a pilot with a few customers or a single feature inside an existing product will tell you most of what you need to know. A focused MVP built for learning is the right size for most trend bets.
4. Separate the durable core from the hype
In each failed trend, something useful survived: stablecoins and tokenisation from crypto, practical AR from the metaverse. Work out which part of a trend solves a real problem and invest there, not in the headline.
5. Do not dismiss a trend just because the hype failed once
Our own AI call is the reminder. A trend can disappoint for years and then take off when the technology matures. Keep watching, keep a small experiment running, and be ready to move when the evidence changes.
6. Keep your architecture flexible
If you build on a fast-moving technology, keep it behind clear interfaces so you can switch providers or remove it without rewriting the product. This applies today to AI model providers as much as it did to blockchains a few years ago.
Applying this to today's trends
The same questions are worth asking about the trends startups are hearing about now: AI agents that run whole workflows, apps generated almost entirely by AI builders, and new categories of AI-powered devices.
Some of these are clearly delivering. Teams using AI coding agents well are shipping faster, and our agentic AI engineers work this way every day. But the familiar warning signs are also there: claims that run ahead of what the tools reliably do, products where the AI feature is the whole pitch, and prototypes that look finished but have no tests or security behind them. The lessons above apply unchanged.
Conclusion
Hype cycles are a normal part of technology. Most big trends contain something real, but the timing, the scale and the winning applications are usually different from what was promised. Startups that do well treat trends as options to test, not strategies to adopt wholesale, and keep their focus on customer problems they can solve today.
If you are weighing whether a new technology belongs in your product, talk to EmphaSoft. With more than 400 projects delivered since 2010, we have seen several hype cycles come and go, and we are happy to give you a practical view on what is worth building now.
