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. ChatGPT had come out a few weeks earlier, and within a couple of years AI was the dominant topic in software.
The old article is more useful now than when we wrote it. A few years on, it is easier to see which hyped trends failed, which just took longer, and which we misjudged. The same lessons apply to the trends startups are being pushed toward today, AI agents included.
Below we go back over the biggest hype cycles of the early 2020s and turn them into practical rules for founders and product owners.
Mass crypto payments
The promise
Around 2021 many commentators expected people to start paying for everyday goods in cryptocurrency, 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. In most markets everyday crypto payments never went mainstream. Price swings, transaction costs and regulatory uncertainty worked against them, and paying in crypto was simply clumsier than paying by card or instant bank transfer.
What survived
The underlying technology is still here. Financial institutions and fintech companies keep doing serious work on stablecoins, tokenisation of financial assets and blockchain-based settlement, and regulation in several regions is clearer than it was. What failed was the consumer payments story, which was sold years ahead of the product. 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 pitched 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. Most buyers ended up holding speculative assets with little practical use. Many brand projects were wound down.
What survived
Verifiable digital ownership still has niches, for example in ticketing, loyalty schemes and some games. In each of them the value comes from what the token lets you do.
The consumer metaverse
The promise
When Facebook renamed itself Meta in 2021, it was betting 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 never came. For most people headsets stayed a niche product, virtual worlds struggled to bring users back, and many corporate metaverse projects were scaled down or closed. Meta has kept investing 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 in real use for training, design review, industrial maintenance and some healthcare work, and lightweight augmented reality on phones and smart glasses keeps developing. Again, the narrow practical applications survived and the all-encompassing vision did not.
The trend we got wrong: AI
Our 2022 article called AI one of that year's failures. We pointed to disappointing project results, a shortage of skilled specialists and teams focusing on the wrong tools, and all of that was true at the time. Within months, generative AI had changed the picture completely.
We think that is a lesson in its own right. The AI hype had been building for years with modest results, and scepticism was reasonable. Meanwhile the research kept improving, and once the technology crossed a usability threshold, adoption came very fast. By 2026 AI coding agents, AI features in mainstream products and AI and machine learning development are part of everyday business.
The 2022 diagnosis still fits many AI projects today. Unclear goals, poor data and the wrong tools still produce disappointing results, even though the technology itself now works.
Why hyped trends fail: common patterns
The same patterns show up across all of these cases:
- The story arrives before the product. Investors and media describe the end state years before the technology or the user experience can deliver it.
- Speculation hides real demand. When prices or valuations climb fast, you cannot 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, and 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 run into regulators sooner or later, and the rules can change the business model.
- Funding moves in herds. When investors move on to the next trend, projects that lived on hype instead of revenue run out of runway.
Lessons for startups
1. Start from a customer problem, not a technology
The startups that came through these hype cycles were usually solving a specific problem for a specific customer, with the technology as a means to an end. If you cannot describe your product without naming the trend, treat that as 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 all it has going for it is novelty, adoption will be slow.
3. Run small, cheap experiments
Testing an idea does not require a full metaverse team or a token launch. A prototype, a pilot with a few customers or one feature inside an existing product will tell you most of what you need to know. For most trend bets, a focused MVP built for learning is the right size.
4. Separate the durable core from the hype
Each failed trend left something useful behind: stablecoins and tokenisation from crypto, practical AR from the metaverse. Work out which part of a trend solves a real problem and put your money there instead of into the headline.
5. Do not dismiss a trend just because the hype failed once
Our own call on AI shows why. A trend can disappoint for years and then take off once 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, put it behind clear interfaces so you can switch providers or remove it without rewriting the product. Today that applies to AI model providers as much as it applied to blockchains a few years ago.
Applying this to today's trends
Startups now hear about AI agents that run whole workflows, apps generated almost entirely by AI builders, and new categories of AI-powered devices. The same questions apply to all of them.
Some of these trends are clearly paying off. Teams that use AI coding agents well ship faster, and our agentic AI engineers work this way every day. The familiar warning signs are there too: 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. Everything in the lessons above still holds.
Conclusion
Hype cycles are a normal part of technology. Most big trends contain something real, but the timing, the scale and the applications that win usually differ from what was promised. Startups that do well treat a trend as an option to test instead of a strategy to adopt wholesale, and stay focused on customer problems they can solve today.
If you are deciding whether a new technology belongs in your product, talk to Mefasoft. We have delivered more than 400 projects since 2019 and watched several hype cycles come and go, and we can give you a practical view on what is worth building now.