Ivylane has been online since 1995 — before Google, before broadband, before the word "website" needed no explanation. That's not a nostalgia credential. Its only value is sample size: we've now watched roughly eight waves of technology arrive, peak, get oversold, crash in reputation, and then quietly become infrastructure. The pattern knowledge is the asset.
Every cycle, from inside
Each era had the same three acts. First, a genuine capability appears — online payments, search, the phone in every pocket, rented servers, machine learning, generative models. Second, a gold rush of claims forms around it, in which the capability is prescribed for every problem, including the ones it's bad at. Third, disillusionment prunes the nonsense — and what remains becomes boring, load-bearing, and permanent. Ecommerce survived the dot-com crash. Mobile survived the app-for-everything years. The cloud survived "cloud-washing." The technology almost always survives; the claims die.
The practical skill is telling, in act two, which parts are act-three material. A few tells have held up remarkably well:
- Does it collapse a real cost? Ecommerce collapsed distribution cost. Cloud collapsed server cost. AI-assisted development collapses production cost. Capabilities that collapse costs endure; capabilities in search of a problem don't.
- Does it survive the demo? Every era's vaporware shared a trait: incredible on stage, fragile against real data, real users, real edge cases. If nobody can show it running unattended for a competitor's use case, it's still theater.
- Who bears the failure? Durable technologies fail visibly and accountably. The ones that quietly transfer risk to the buyer — your data, your lock-in, your liability — end up in act-three's discard pile.
What survived all eight cycles
More interesting than what changed is the short list that never did. Every era rewarded the same fundamentals, usually while loudly claiming they were obsolete:
- Speed. Fast pages won on dial-up in 1996, won on mobile in 2010, and win with AI crawlers in 2026. The impatient reader changes species; the principle doesn't.
- Structure. Clean documents beat clever tricks in early search, and beat them again with language models. Machines have always preferred legible.
- Ownership. Every era mints a new landlord — platforms, app stores, page builders — and every era's casualties are the businesses that didn't own their domain, data, or code. Renting convenience is fine; renting your existence isn't.
- Small, finished things. The graveyard is full of platforms that were going to do everything. The survivors are mostly tools that did one thing completely. We scope accordingly.
- Accountable humans. Through every abstraction — agencies, offshoring, no-code, now AI — the projects that worked had someone specific who understood the whole thing and answered for it.
Where that leaves the current cycle
Generative AI is a genuine act-one capability — we use it daily and it has re-priced our whole category — currently wrapped in a perfectly ordinary act two. The claims will get pruned; the capability will become plumbing. So the advice from eight cycles is unglamorous: build the fast, structured, owned, small, accountable version of whatever you build, and adopt the new capability precisely where it collapses a cost you actually pay — not everywhere it's being prescribed.
Technologies change. Good judgment about them compounds. That compounding is the only thing "since 1995" is meant to claim.