How do you identify emerging trends before they become mainstream?

Watch the four leading indicators that precede almost every startup wave: regulation changes, new platform capabilities, cost curves crossing thresholds, and demographic or behavior shifts. Each shows up in a specific primary source (the Federal Register, platform changelogs, arXiv and earnings calls, census and survey data) 12 to 36 months before the trend hits mainstream coverage.

Is it too late to start an AI startup?

Too late for the obvious layer, early for the second-order layer. Generic chatbots and thin API wrappers are crowded, but each new capability (agents that execute multi-step work, cheap voice, long context) resets the clock for specific verticals. The question is never the technology; it is whether your specific wedge still has an open window.

What is a why-now for a startup?

The specific, recent change that makes an idea newly possible or newly urgent: a law that took effect, an API that launched, a cost that crossed a threshold, a behavior that shifted. Ideas without a why-now were either always possible (so incumbents own them) or are still impossible. Investors ask for it because timing explains more startup outcomes than idea quality does.

How to Spot Startup Trends Before They're Obvious (and Everyone Piles In)

You spot trends early by watching the inputs instead of the outputs. Every startup wave is preceded by a small set of leading indicators: a regulation changes, a platform ships a new capability, a cost curve crosses a threshold, or a behavior quietly shifts. These show up in primary sources 12 to 36 months before the trend has a name, and by the time TechCrunch calls it a trend, the wave is mostly priced in.

The good news is the primary sources are public and mostly free. Here is where to look, what each signal looks like, and three historical timelines that show how long the early-mover window actually stays open.

What are the leading indicators of a startup wave?

Four of them, and each has a specific place it appears first:

Leading indicator	What it looks like	Where it shows up first	Typical lead time
Regulation changes	New compliance obligations, bans, subsidies, reporting requirements	The Federal Register, EU official journals, state legislature trackers	12 to 36 months (laws announce their own effective dates)
New platform capabilities	An API, SDK, app store, or hardware feature opens up	Platform changelogs, developer blogs, WWDC/IO keynotes, API pricing pages	6 to 24 months
Cost curves crossing thresholds	Something drops from expensive to trivial: compute, sequencing, batteries, launch mass	arXiv papers, earnings call transcripts, industry price indices	12 to 48 months
Demographic and behavior shifts	Who has money, time, or a new habit changes	Census releases, BLS data, annual surveys, sub-culture forums	24 to 60 months

Regulation is the most underrated of the four, because a law is a trend with a published schedule. When a compliance deadline is set for a specific date two years out, you know exactly when thousands of companies will suddenly need software, audits, and services, and you know it before they do. Founders who read the actual text of a new rule are usually competing with nobody.

Platform capabilities are the fastest-moving indicator. A changelog entry that says "API now supports X" is an invitation letter with a short expiry: the window between "capability exists" and "category is crowded" has been shrinking every cycle.

Cost curves are the deepest signal. Almost every "sudden" boom is a cost quietly crossing a threshold: when something drops below the price where a new use case pencils out, that use case goes from research project to product category within a couple of years. Earnings calls are shockingly candid here; executives announce their cost trajectories to analysts every quarter, in public.

What do early signals look like in hindsight?

Three timelines worth internalizing, because the pattern repeats.

GPT-3 API to the AI-app boom. OpenAI opened API access to GPT-3 in June 2020, and for about 18 months the capability sat in plain sight while most of tech treated it as a toy. Copy.ai launched in late 2020, Jasper in early 2021, and Jasper reached a reported $1.5 billion valuation by October 2022, before ChatGPT even existed. The founders who moved in 2020 and 2021 had a two-year head start on the 2023 crowd. The signal was a public API announcement plus a pricing page; anyone could have read it.

The Shopify app ecosystem. Shopify launched its App Store in 2009, when Shopify itself was a small player. Building tools for merchants on a then-modest platform looked like a niche bet for years. The developers who committed early ended up owning categories as Shopify grew into millions of merchants, and companies that leaned into that ecosystem early, like Klaviyo in email, rode it all the way to an IPO in 2023. The signal was a platform opening a developer surface; the window stayed open for years because the platform's own growth kept expanding it.

Remote-work tooling. The behavior shift toward distributed work was visible in surveys and job postings through the 2010s, and the toolmakers who built before the forcing function (Zoom founded 2011, Loom 2015, Notion's rise through 2018 and 2019) were positioned when 2020 compressed a decade of adoption into a quarter. The founders who started remote-work tools in April 2020 were not early. They were responding to CNN.

Different indicators, same shape: public signal, quiet period where the opportunity looks weird, then a stampede.

Why does an idea that feels safe mean you're late?

Because "feels safe" is just the sensation of consensus, and consensus is what a crowded market feels like from the inside. By the time an idea feels safe, the window is already crowded. The discomfort you feel looking at a weird, early idea is the price of the open window; the comfort you feel looking at an obvious one is other founders, already in the room.

This flips how you should evaluate your own reactions. "Everyone will want this" is a warning label. "This seems too early, too narrow, or slightly embarrassing" is at minimum worth a weekend of validation. Timing explains more outcomes than execution polish does, which is why the strongest ideas can usually complete the sentence "this was impossible or pointless until [date], because [change]." If an idea cannot complete that sentence, an incumbent has probably had a decade to do it already.

How do you build a trend-watching habit that fits in 30 minutes a week?

A sustainable routine beats a heroic one. This takes 30 minutes weekly:

Pick two indicators to own. Choose the two that match your skills: developers should watch platform changelogs and arXiv, domain experts should watch their industry's regulation and earnings calls. Ignore the other two; depth beats coverage.
Subscribe at the source. Federal Register topic alerts, the changelogs of 3 platforms you build on, arXiv category feeds, earnings call transcripts for the 5 public companies nearest your niche. Primary sources only; newsletters that summarize newsletters are the lagging indicator.
Keep a why-now log. One line per signal: what changed, when it takes effect, who suddenly needs something because of it. Review monthly. Most entries die; the ones that keep accumulating supporting signals are your shortlist.
Let a researched feed cover the ground you can't. Two indicators is what one busy person can genuinely monitor, and waves come from all four. This is the gap Ignition was built to fill: every idea it delivers each morning includes an explicit why-now field stating what just changed, precisely because timing is the whole game. The why-now field exists so that no idea in the feed is of the "always possible, never urgent" variety, and the accompanying competition and market data tells you how crowded the window already is.

Act on a 2-of-4 rule. When one opportunity shows up under two different indicators (say, a new regulation and a cost curve both pointing at the same buyer), stop watching and start validating.

Your starter kit for this week

Subscribe to Federal Register alerts for one industry you know.
Bookmark the changelogs of the three platforms you would build on.
Add one arXiv category feed and the earnings call schedule for two companies near your niche.
Create the why-now log with three entries from this week's reading.
Calendar 30 minutes, same time weekly, to update it.

Trends are not predicted by geniuses. They are noticed by people who read the boring documents 18 months before everyone else reads the coverage of them. If you want to see what fully-formed, timed ideas look like right now, the 2026 ideas list is built entirely out of why-nows from the last 18 months.